* Add ctx-per-slot argument for unifid KV cache
* Swap out ctx fractions for ctx pool slots
* Formatting cleanup
* Remove ctx-pool-slots, make ctx-per-slot an int
* refactor it
---------
Co-authored-by: Xuan Son Nguyen <son@huggingface.co>
* gguf: add qwen4exp (Qwen3.8-Flash-Next) arch and converter
Adds the GGUF-side plumbing for HF model_type qwen4_exp:
- MODEL_ARCH.QWEN4EXP plus tensors for the low-rank hyper-connection
variant (hc_*_norm/down/up/inject) and the PLE n-gram hash embeddings.
The DeepSeek-V4 hc_*_fn/base/scale tensors are a different
parameterisation, so these are separate entries rather than reuse.
- Reuses the existing indexer, per_layer_token_embd, SSM and
compress_ratios keys unchanged.
- conversion/qwen4exp.py inherits the Qwen3.5 linear-attention V-head
reorder and interleaved mrope, concatenates the 128 PLE embedding
shards, and splits index_qk_proj into separate indexer q/k tensors.
The PLE hash multipliers reach ~2.4e13. prepare_tensors() casts every
non-float dtype to float32 before modify_tensors() runs, and GGUF array
writes infer INT32 from Python ints, so both paths are bypassed: the
constants are read from the pre-cast lazy tensors and written as
explicit UINT64 arrays.
Additive only; no existing arch changes behaviour.
* llama: load qwen4exp (Qwen3.8-Flash-Next) hparams and tensors
Adds LLM_ARCH_QWEN4EXP with its hparams and tensor loading. The graph
comes in the next commit; this makes the model load and report correct
metadata.
- hyper-connections set n_embd_out_impl = hc_count * n_embd, so the
residual stream is 4x wide and there is no output_norm: the final
mixer's hc_norm is the last norm in the model.
- registered as hybrid and given the same recurrent/attention memory
filters as Qwen3-Next and Qwen3.5.
- reuses the existing indexer, per_layer_token_embd, SSM and
compress_ratios keys as-is.
- the PLE table row count is read back from the file rather than
recomputing the vocab padding rule.
llama-model-loader gains UINT64 array support. That branch previously
threw, so no existing caller changes behaviour; it is needed because the
PLE hash multipliers do not fit in int32.
* qwen4exp: shorten comments
* llama: qwen4exp text graph with hyper-connections, GDN and MoE
Implements the decode graph for Qwen3.8-Flash-Next: the hyper-connection
residual stream, gated delta net layers, the MoE block with its gated shared
expert, and dense full attention. The QSA indexer and the PLE n-gram embedding
are not wired up yet and land in later commits.
Hyper-connections are implemented here rather than shared with deepseek4.cpp.
The two formulations agree on the [n_embd, hc, n_tokens] layout and little
else: DeepSeek-V4 mixes with a full-rank projection and Sinkhorn-normalises
it, whereas this model uses a low-rank down/silu/up sigmoid gate and collapses
by a plain mean. Only the ~10 line stream mean is genuinely common, so sharing
would mean touching DSV4's hot path and its three fused CUDA ops to reuse very
little. What is reused is the substantive part: the LLM_KV_HYPER_CONNECTION_*
keys, the n_embd_out_impl wide-residual support already in the loader, and the
layout convention.
Also allows a checkpoint to carry no PLE layers at all, which makes it
possible to bring the graph up and validate it in stages.
Validated against vLLM, the only working reference implementation. On a
scaled-down model with an init scale large enough to give non-uniform logits,
agreement with vLLM sits at the numerical noise floor: llama.cpp f32 against
its own bf16 gives 84.3% top-1 agreement over 255 positions, and this graph
against vLLM gives 85.1%. The comparison was calibrated by seeding three
deliberate bugs (silu instead of sigmoid on the delta net gate, dropping the
1/hc scale in the mix, dropping the 2x in the combine); each drops top-1 to
between 0% and 11%, an order of magnitude below the floor.
* llama: qwen4exp PLE n-gram hash embedding
Adds the per-layer embedding: a custom I32 graph input hashes each token with
its ngram_size-1 predecessors host-side and the result is a plain row gather
over the shared table, the same shape gemma3n's per-layer embedding uses. The
hash has to run on the host because the splitmix64-derived multipliers reach
2^45, so the products need 64-bit integers and an xor, neither of which ggml
has.
Predecessors that fall outside the ubatch come from a small per-sequence
history on the model, mirroring the per-request ngram_context the reference
carries. It is only trusted when contiguous with the incoming position, so a
fresh prompt or a rewound cache falls back to EOS padding rather than hashing
against stale tokens.
The depthwise conv is written out as a sum of shifted, per-channel-scaled
copies rather than through ggml_conv_1d_dw, which carries a correctness
warning upstream.
Verified two ways. The row indices match a transcription of the reference's
tensor formulation exactly, 1024 of 1024 rows, including sequences with EOS
tokens sprinkled through them to exercise the segment reset. Separately, with
PLE placed on layer 0 so its input is just the token embedding, ple_embd and
ple_gated_value match a PyTorch computation from the same checkpoint to every
printed digit.
End to end over 1023 scored positions the port sits the same distance from
vLLM with PLE as without it, 6.3 points of top-1 against 6.0, so PLE costs no
accuracy relative to the rest of the model. That common offset is vLLM's bf16
activations, which cannot be removed: its QSA kernel refuses float32.
Two bugs found along the way, both caught by the row-index check. The history
was read and updated in the same pass, so a token early in a ubatch could pick
up an earlier token of that same ubatch as prior context; it is now snapshotted
first. And an EOS token was cutting its own context, where the reference takes
the last EOS strictly before the position, so a boundary only hides tokens from
the positions after it.
Known gap: the conv carries no state across ubatches, so it is exact only for a
prefill that starts at position 0. Chunked prefill and decode need the conv
state wired into the recurrent memory, and the conv branch itself is still
numerically unverified because the fixture zeroes its weights.
* llama: carry the qwen4exp PLE conv state across ubatches
The PLE depthwise conv was zero-padding on the left, which is only right for a
prefill that starts at position 0. Decode and chunked prefill saw a truncated
history for the first (kernel-1)*ngram_size positions of every ubatch.
The PLE module sits on a layer that is also a delta-net layer, so both need a
conv history in the same recurrent row. Rather than plumb a per-layer state
size through build_rs and build_conv_state, the row is widened once and each
convolution addresses its own slice through a local helper. n_embd_r() gains
the extra span, which is zero for every other architecture because it is
derived from ple_n_heads.
Verified by feeding the same 1024 token sequence in chunks instead of one
shot: at 64 tokens per decode the logits are bit-identical to the single-shot
run, 1023 of 1023 top-1 and a maximum logprob deviation of exactly zero. At
one token per decode they differ slightly, but the no-PLE model differs more
under the same test (94.6% against 97.1%), so that is the usual gemv-versus-
gemm accumulation difference and not the state.
The conv branch is also no longer unverified. With non-zero conv weights the
port sits 6.3 points of top-1 below the numerical floor, the same distance as
with the weights zeroed and as the model with no PLE at all, so the branch
adds no error of its own.
test-llama-archs passes every existing architecture at 0.00e+00, including the
delta-net models that share this code path.
* llama: fix the qwen4exp PLE conv state and unblock test-llama-archs
build_rs writes into the state tensor in place, zeroing one row and copying the
carried-over states, so calling it twice for the same layer let the second call
clobber the first write-back. The PLE layer is also a delta-net layer, so that
is exactly what happened: both convolutions gathered the same row. They now
share a single gather per layer.
The earlier claim that the conv state was carried correctly was tested on a
fixture whose conv weights are zero, where the branch contributes nothing and
chunking matches trivially. Re-running with non-zero conv weights showed the
divergence, growing with the number of ubatch boundaries: 97.1% top-1 at one
boundary down to 90.2% at seven. With the shared gather it is bit-identical to
the single-shot run at every chunk size tried, 512, 128 and 64, with a maximum
logprob deviation of exactly zero over 1023 positions. The delta-net-only model
stays bit-identical too, so nothing regressed there.
Also derive the delta-net conv channel count the way load_arch_tensors sizes
wqkv instead of from ssm_d_inner. The two agree for this model, but n_embd_r()
only bounds the row and the convolution has to match the tensor feeding it.
test-llama-archs previously aborted on this architecture and took every later
architecture with it. qwen4exp is marked MoE-only, given the hyper-connection
keys and an ssm_d_inner consistent with its tensor derivation, and skipped for
now: the hyper-connection keys written by get_gguf_ctx are not reaching the
synthesised file, which needs a separate look. The suite completes again, 124
architectures at 0.00e+00.
* llama: optional indexer key cache in llama_memory_hybrid
Groundwork for qwen4exp's QSA sparse attention. Its indexer needs a per-token
key history for the full-attention layers, but a hybrid model cannot use
llama_kv_cache_dsa: that class derives from llama_memory_i rather than
llama_kv_cache, and llama_memory_hybrid constructs its attention cache
directly. No existing architecture pairs recurrent state with a sparse
indexer, so there was nothing to reuse wholesale.
llama_memory_hybrid therefore gains a third, optional cache, shaped the same
way llama_kv_cache_dsa shapes its lightning-indexer cache: a copy of hparams
with n_head_kv forced to 1 and n_embd_head_k_full set to indexer_head_size.
It is built only when a filter_idx callback is passed, which defaults to
nullptr, so every existing architecture gets exactly what it got before. The
per-sequence operations and the batch preparation forward to it under a null
check, matching how the DSA cache prepares its two caches over the same
ubatches.
test-llama-archs passes all 124 architectures at 0.00e+00, including the 12 in
the hybrid family that share this code. The qwen4exp fixtures are unchanged:
same logits against vLLM, and chunked evaluation still bit-identical to
single-shot.
* llama: QSA sparse attention for qwen4exp
The full-attention layers of this model do not attend to everything. An
indexer scores one mean-pooled key per block of compress_ratio tokens and
keeps a budget of the best blocks, plus the tail of tokens that do not yet
form a complete block. Below indexer_top_k + compress_ratio - 1 cached
tokens every block fits in the budget, so the result is exactly dense.
What is reused rather than rebuilt:
- the mask machinery. build_attn's DSA overload already turns a list of
token indices into a KQ mask via ggml_set_rows, so that block is lifted
out verbatim into build_attn_mask_top_k and shared with a new overload
on llm_graph_input_attn_kv. DSA's node sequence is unchanged; the new
overload exists because llama_kv_cache_dsa assumes MLA and cannot be
dropped into a hybrid model.
- the indexer key cache, which is the optional third cache added to
llama_memory_hybrid in the previous commit. It holds raw keys, because
pooling happens before the norm and the rotation.
The graph expands block scores rather than block indices: giving every
token of a block its block's score needs only a gather, where expanding
indices would need an integer multiply-add that ggml has no op for. Since
the budget is a whole number of blocks and a block's members tie exactly,
the cut still lands on a block boundary.
Everything that depends on cache layout is computed host-side in
set_input_qsa. Blocks are cuts of the position line rather than of the cell
array, so nothing assumes the cache is contiguous.
Measured on the tiny fixture against vLLM, comparing the selected token
indices directly rather than the logits:
below the budget selection identical, and 1024-token logits are
bit-identical to the pre-QSA dense path
above the budget mean jaccard 0.975
The direct index comparison is what made this correct. The reference
rectifies each head's dot product before summing over heads, which an
earlier reading of it had missed; on logits alone the resulting port looked
fine, because on a randomly initialised fixture the known-correct dense
path already disagrees with vLLM by more than the bug did. Comparing the
indices showed 0.794, and fixing the ReLU moved it to 0.975.
* llama: give the qwen4exp indexer cache the attention cache's slots
The indexer cache found its own slots, independently of the attention
cache. Both are the same size and see the same ubatches, so in a
straight-through prefill they agree, which is why every fixture and every
single-shot parity run passed. They drift once the context is being
rewritten between turns, and then the QSA top-k indices, which are applied
against the attention mask, point at the wrong cells.
The seven-turn chat test caught it on the third turn: llama-server aborted
on the assertion that the two caches report the same n_kv.
The cache is a side buffer addressed by the attention cache's cells, so it
now takes that cache's slot layout instead of computing one. Applying that
layout also marks its cells identically, so the two agree cell for cell by
construction rather than by coincidence, and the assertion can no longer
fire.
Inert where the caches already agreed: test-llama-archs green at 126 archs
and 0.00e+00, and the 4096-token tiny fixture is unchanged at max logit
delta 0.0.
* tests: record what the qwen4exp arch-test skip actually observes
The old note guessed that the hyper-connection keys never reach the file.
They do: dumping the gguf_context handed to llama_model_init_from_user
shows both among its 67 KVs, and the loader still reports one missing.
* tests: cover qwen4exp in test-llama-archs
The arch was skipped with a note guessing that the hyper-connection keys
never reached the synthesised file. They did. The suite builds a model, then
saves and reloads it, and llama_model_saver did not re-emit those keys, so
the failure was in the roundtrip leg rather than the first load. Three gaps,
all in shared code and all additive:
- add_kv_from_model wrote no hyper-connection, compress-ratio or PLE keys.
The PLE group only means anything whole, so it is written or omitted
together; the rest follow the file's existing style of writing every key
unconditionally, since an architecture that does not read one is
unaffected by a zero.
- the saver had no uint64 path at all, which the PLE hash constants need.
- add_tensors_from_model enumerates model-level tensors by hand and was
missing per_layer_tok_embd and the three final-mixer tensors.
Two smaller fixes on the qwen4exp side, both found by running the test:
- build_qsa_top_k divided by the compression ratio before asserting it was
non-zero, so a file without the key crashed instead of reporting.
- a layer with no compression ratio now falls back to dense attention,
which is what the model computes below the budget anyway. The test then
has to write a ratio to reach QSA at all, and an indexer key length no
narrower than n_rot, since the indexer ropes with the main attention's
rotary width.
Full suite: 126 archs, qwen4exp at 0.00e+00 with roundtrip OK. The tiny
fixture is unchanged, max logit delta 0.0 against the pre-QSA dense run.
* convert: stream the qwen4exp PLE table instead of concatenating it
The n-gram table arrives as 128 shards that were held in a dict and then
torch.cat-ed, so the peak was the shards plus the concatenation: around
300 GB of RSS on the real checkpoint, which rules out machines that could
otherwise convert this model.
Each shard is now written straight into a memory-mapped file at its final
row offset and dropped, so the resident set is one shard and the rest is
the page cache's problem. The temporary file sits beside the output and is
removed once the write finishes, including on failure.
Shards other than the last must be uniform for direct placement, which is
asserted rather than assumed, and a shard arriving before the stride is
known is held instead of misplaced.
Verified on the tiny fixture: the resulting GGUF is byte-identical to the
one the concatenating path produced (md5 2d274efac91ad1e9a6007efb0687e597).
* quantize: fall back to F16 for 32-block types with an odd ncols
tensor_type_fallback demotes a tensor whose ncols is not a multiple of the
target's block size, but its switch only enumerates the 256-block types. A
target that is already a 32-block type (iq4_nl, q4_0, q5_0, q8_0, ...) falls
into default: and throws, even though the function already knows how to answer
that case: the ncols check right below the switch resolves an unrepresentable
shape to F16.
Route those types into that check instead of throwing. Only paths that abort
today change, so no quantization that currently succeeds is affected.
Found on a 4-wide depthwise conv kernel. llama-quantize reported nothing but
"failed to quantize model from ...", with no tensor name and no exception text,
which made a quant recipe that had simply not pinned the tensor look like a
corrupt model. It now names the tensor and continues.
* quantize: let --tensor-type name per_layer_token_embd
per_layer_token_embd shares the TOKEN_EMBD category with token_embd.weight, so
--token-embedding-type is returned for it before any --tensor-type pattern is
consulted, and there is no way to give it a tier of its own.
That grouping is fine as a default and stays the default. It is a poor fit for
the size, though: on qwen4exp the table is 97.7 GiB of a 337.6 GiB BF16 file and
about 46% of a 4-bit one, roughly eighty times token_embd.weight, and it is
read by ggml_get_rows rather than a matmul so no imatrix ever covers it.
Allow an explicit --tensor-type pattern to name it, and only it. Nothing
changes unless such a pattern is passed, and token_embd.weight keeps the old
precedence in either case.
Measured on Qwen3.8-Flash-Next, Q4_K_M with an imatrix: the table lands at q8_0
(51.9 GiB, 113.5 GiB total) by following --token-embedding-type, and pinning it
q4_1 gives 30.5 GiB for 92.1 GiB total, 19% off the file.
* quantize: size the output buffer exactly instead of nelements * 4
The per-tensor output buffer was sized `nelements * 4`, described as an upper
bound. It is a very loose one: the output is at most 2 bytes per element
(f16/bf16) and usually well under 1.1 (q8_0 and below), so between 2x and 4x of
it is never touched. The exact size is already known here, since it is what the
quantization loop writes, what new_size sums to, and what the GGUF metadata is
asserted against a few lines later.
On a model whose largest tensor is a few GB none of this matters. On
Qwen3.8-Flash-Next it does: per_layer_token_embd is 51.2 G elements, so the
buffer was 205 GB where 54 GB is needed at q8_0 and 32 GB at q4_1.
Measured on that model, VmHWM of a live llama-quantize was 485 GB per process.
Three of them fit in 2 TB and five did not, which is what an OOM-killed quant
ladder looks like. This removes about 150 GB of that.
Byte-identical output, verified against the same binary built at the parent
commit: q4_K, q8_0, q5_K, q6_K and IQ4_XS, over BF16 and F32 sources, with and
without a PLE table present. Six cases, six matching md5s.
* qwen4exp: hash the image placeholder for multimodal batches
The PLE row indices are computed host-side from ubatch->token, and set_input
returned early when that was null. A multimodal ubatch is exactly that case:
the mtmd layer consumes the image placeholder ids and hands llama_decode
embeddings instead. The early return left the I32 index tensor uninitialised,
so ggml_get_rows indexed a 320 M row table with whatever the buffer happened to
contain, and aborted:
GGML_ASSERT(i01 >= 0 && i01 < ne01) failed
ggml_compute_forward_get_rows
mtmd_helper_decode_image_chunk -> llama_decode
Every image request crashed. Nothing caught it because the vision work had only
ever been verified by converting an mmproj, never by running one.
The reference computes the hash over input_ids, where those positions still
hold the image placeholder, so carry that id through as qwen4exp.ple.image_token_id
and hash it. The key is optional: a file converted before it existed falls back
to the PLE EOS token, which is defined and treats the image as a segment
boundary rather than crashing.
Verified end to end with llama-mtmd-cli, a Q4_K_M base and the F16 mmproj, on a
generated image with known content. The model names the red circle, the blue
square, the inverted green triangle and reads "UNSLOTH 42", each with the right
position.
* qwen4exp: support a non-unified KV cache in QSA
set_input_qsa asserted n_stream == 1, so llama-server could not serve this
model with more than one slot unless -kvu was passed. With a non-unified
cache each sequence owns its own cells, and a cell index means a different
token in each stream, so a single shared mapping is wrong.
- cell_blk, blk_cells and bias gain a stream dimension. At n_stream == 1
these collapse to the shapes they had, so the unified path is unchanged.
- Scoring is now batched over streams. ggml_mul_mat matches ne[2] on both
operands, so stream s's queries only ever meet stream s's blocks; without
this sequences would score against each other's context.
- set_input_qsa loops per stream and resolves cells through
v_cells[seq_to_stream[seq_id]], following set_input_kq_mask_impl, instead
of hardcoding v_cells[0].
- llama_kv_cache_context::get_n_stream() is added, mirroring the ns that
get_k and get_v already derive from the slot info.
build_attn_mask_top_k needed no change: it already expects
[n_top_k, n_batch, 1, n_stream], so the top-k result is reshaped to meet it.
set_input_qsa has exactly one caller, so the blast radius is qwen4exp only.
Validation, UD-Q4_K_XL on one B200:
- unified cache unchanged within noise: 1802.9/68.85 -> 1807.2/69.11 t/s at
batch 1, 2262.5/192.43 -> 2270.1/193.75 at batch 4.
- non-unified now runs at npl 1, 4, 16 where it previously aborted, and is
22% faster than the -kvu workaround at batch 16 (1205 vs 984 t/s total),
since per-stream cells avoid the cross-sequence masking a unified cache
pays for.
- no cross-stream contamination: four concurrent sequences each carrying a
distinct secret all recall their own and no other, on both cache modes.
- test-llama-archs green on qwen4exp, deepseek2, gemma3n, qwen3next, llama.
Note on testing: comparing concurrent output against solo output exactly is
not a valid check. It failed 0/4 with no bug present, and the unified-cache
control failed the same way, because batch composition changes the
floating-point reduction order and near-tied tokens flip. The contamination
test above is what the exit code gates on.
* llama: keep the qwen4exp top-k attention mask arch-local
The QSA graph needed a build_attn that attends only to the cells named by a
top_k tensor, and the first version got it by adding a llm_graph_input_attn_kv
overload to llm_graph_context and factoring the mask construction out of the
existing MLA sparse path into a shared build_attn_mask_top_k.
That put a new arch on the shared attention path and made the deepseek32 and
glm-dsa attention build depend on a helper introduced for qwen4exp. Build the
mask in src/models/qwen4exp.cpp instead and leave llama-graph.{h,cpp} exactly as
they were: the MLA path keeps its own copy of the same node sequence.
The nodes emitted are unchanged, so this is bit-identical.
* llama: hold the qwen4exp indexer cache in a new llama_memory_hybrid_idx
The indexer key cache was added by extending llama_memory_hybrid with an
optional third cache, and the host-side cell/block mapping that drives QSA was
added as set_input_qsa on llama_kv_cache. Both are shared classes that every
hybrid and every attention model goes through.
Move both into a new memory type, llama_memory_hybrid_idx, following
llama_kv_cache_msa: the indexer cache and the pos<->cell translation live with
the sparse-attention memory rather than in the classes that serve every other
architecture. llama-kv-cache.{h,cpp} and llama-memory-hybrid.{h,cpp} are
restored to their unmodified state.
init_batch is repeated from llama_memory_hybrid because the indexer cache has to
be handed the attention cache's slot infos, and those are not reachable through
the context the base returns. Allocating them separately lets the two caches
drift, which is what pointed QSA's top-k at the wrong cells before.
The context derives from llama_memory_hybrid_context so build_inp_mem_hybrid
keeps working unchanged, and get_n_stream is computed from the slot infos
exactly as llama_kv_cache_context did.
Behaviour is unchanged: logits over an 8192-token sequence are bit-identical to
the previous implementation, sparse and dense alike.
* llama: save and restore the qwen4exp indexer KV cache
llama_memory_hybrid_idx forwarded clear, seq_rm, seq_cp, seq_keep, seq_add and
seq_div to the indexer cache but not state_write / state_read, so a saved
session dropped the indexer keys and a restored one selected QSA top-k against
an empty cache. The effect is invisible until the context passes
indexer_top_k + compress_ratio - 1 cells, because QSA is exactly dense below
that and the indexer contents cannot change the result.
The indexer section is written last rather than next to the attention cache it
mirrors. As a suffix, a reader that does not expect it stops early and the
trailing bytes are caught by the size check in state_load_file; placed between
the attention and recurrent sections it would instead be parsed as recurrent
state, which can succeed and restore silent garbage. It follows the same
LLAMA_STATE_SEQ_FLAGS_PARTIAL_ONLY gate as the attention cache, since a partial
checkpoint deliberately skips the token-level attention caches.
The indexer restores its own cells instead of taking the attention cache's
restored slots. The two caches share size, padding and every sequence
operation, and init_batch hands the indexer the attention cache's slot infos,
so both state_read_meta calls run find_slot over identical occupancy and land
on identical cells.
The overrides live on llama_memory_hybrid_idx, the only memory type that owns
an indexer cache, so llama_memory_hybrid and every architecture that uses it
write and read exactly the bytes they did before.
The session and sequence state versions are bumped because the qwen4exp state
layout changed. The session path already rejects a short read via its size
check, but llama_state_seq_load_file accepts one silently, so only the version
check stops a pre-fix blob from being half-restored by a fixed build.
(cherry picked from commit 2721542354f8e158c3217625f4e2e7b83e51e3fe)
* llama: make the qwen4exp PLE n-gram history per context and serialise it
The PLE hash of a token mixes in the ple_ngram_size - 1 tokens before it, which
a decode ubatch does not carry, so they were remembered in a map on
llama_model_qwen4exp. That is the wrong owner twice over.
A llama_model is shared by every context that loads it, and the map was keyed
only by llama_seq_id, so two contexts running the same sequence id - two server
instances on one model, or a draft/target pair - overwrote each other's window.
The next_pos guard turned that into EOS padding instead of a crash, so it
degraded quality silently.
The map was also in no state blob: grep found ple_hist in neither
llama-kv-cache.cpp nor llama-memory-*.cpp nor llama-context.cpp. A restored
context therefore failed the next_pos check on its first ubatch and hashed the
first tokens after the restore against EOS padding. This is why a session blob
round-tripped byte for byte while the restored context computed different
logits: the state was never in the bytes.
It moves to llama_memory_hybrid_idx, which is per context, is the memory type
qwen4exp always builds, and already does the per-sequence bookkeeping this
needs. Every sequence operation now carries the window with it:
seq_rm a rewind (p1 < 0) truncates the window to the surviving prefix and
moves next_pos to p0, so a rollback keeps exact context; a hole
punched in the middle leaves the window non-contiguous, so it is
dropped
seq_cp the destination inherits the source's window, truncated to the
copied position range - a copied sequence continues with the same
n-grams the source would have used
seq_keep every other sequence's window is dropped, like its cells
seq_add a shift that moves the whole window keeps it and moves next_pos with
it, which is the context-shift case; one that cuts through it drops
it
seq_div positions stop being consecutive, so an overlapping window is
dropped
clear everything is dropped
Dropping means next_pos = -1, which set_input turns into full EOS padding: the
same thing a fresh sequence gets, and the same thing this code did before it
followed the sequence operations at all, so no case is worse than before.
The state payload is a self-delimiting list, u32 count then per entry
{ i32 seq_id, i32 next_pos, u32 n_toks, i32 toks[n_toks] }, so a whole-context
save and a single-sequence save share one format and a single-sequence restore
can retarget the window at its destination seq_id. It is written after the
indexer section, last, for the same reason that one is: as a pure suffix an
older reader stops early instead of parsing these bytes as something else.
Unlike the indexer section it is not under LLAMA_STATE_SEQ_FLAGS_PARTIAL_ONLY.
The window is recurrent state - it is the input the PLE convolution's own
recurrent state is derived from - and the recurrent cache beside it is written
for partial checkpoints too. Gating it would leave the server's speculative
decoding checkpoints restoring the conv state without the window that produced
it.
No further version bump: LLAMA_SESSION_VERSION 10 and LLAMA_STATE_SEQ_VERSION 3
were introduced for the indexer section in the same unreleased series, and both
changes are qwen4exp-only additions to the same blob layout.
Also fixes the padding of a short window. set_input pads a window shorter than
ngram_size - 1 up to that length, but prev() indexes the snapshot with the most
recent token last, and resize() pads at the back, so the filler EOS landed where
the immediately preceding token belongs. It now pads at the front. A window is
short at a sequence start after a one-token prefill, and after a seq_rm rewind,
which the new bookkeeping makes common.
Every architecture other than qwen4exp builds llama_memory_hybrid rather than
llama_memory_hybrid_idx, has no PLE table and never asks for a history, so
nothing about its graph, its sequence operations or its state bytes changes.
(cherry picked from commit de170364c052c68fcf63285cc0028095edb9f23c)
* qwen4exp: tidy comments and simplify image token read
Rewrite the comments this series adds to the AGENTS.md rules: one or two lines,
no prose hard-wrapped mid-sentence, no narrative or history, and no comment that
only restates the code. Net 146 fewer comment lines, no code change.
Correct the PLE image comment: mtmd does not consume the placeholder ids. An
image is decoded as an embeddings-only batch, so ubatch->token is null and the
per-position ids never exist here. gemma3n and gemma4 hit the same case and
stand in row 0 of per_layer_token_embd; qwen4exp stands in the configured image
token id instead.
Read image_token_id straight from self.hparams in the converter. base.py merges
text_config into the root of hparams, and the key sits at the root of
config.json, so the config.json re-read was redundant.
(cherry picked from commit 205840c12169057da3e8d2f65ec4ceec3e18b980)
* qwen4exp: support a quantized KV cache in the QSA attention path
(cherry picked from commit 4c30574f81dc1115d08078c47b6cf8c789c0a842)
* llama: give qwen4exp a large-graph node budget
(cherry picked from commit 37c8c194e6a30e4c46ac29bee3fb264f091596ef)
* qwen4exp: drop an unused variable that breaks -Werror builds
(cherry picked from commit 528d032b51fa3cf935ed3ef6e0fb1c7401df53b5)
* quantize: dequantize and quantize large tensors in row bands
f32_conv_buf held the whole dequantized tensor, which is 204.8 GB for
per_layer_token_embd alone and dies with std::bad_alloc long before the
work buffer is reached. Dequantize and quantize in bands of whole rows
instead, capping the f32 staging at 1 GiB per band.
Rows are independent and the imatrix is indexed by column, so band
boundaries cannot change any output byte. Bands nest inside the existing
per-expert loop so each expert slice keeps its own imatrix, and a band is
kept to at least one quantization chunk per worker thread so the existing
multithreading still has work. F32 sources still stage nothing and are
banded by pointer arithmetic into the tensor.
llama_tensor_dequantize_impl now takes a first element offset; the single
caller is updated.
(cherry picked from commit 658c22549613555dbce57a772be4de8509eba3ee)
* llama: segment the qwen4exp fused QKV for tensor split
qwen4exp was missing from the gated delta net branch of get_split_segments,
so its attn_qkv.weight, shaped {n_embd, 2*key_dim + value_dim}, fell through
to the generic fused QKV rule and tripped
GGML_ASSERT(tensor->ne[axis] == n_embd + 2*n_embd_gqa) while loading with
--split-mode tensor. --split-mode layer was unaffected.
qwen4exp broadcasts K to the V heads by tiling, k_conv is grown with a plain
ggml_repeat_4d over the head axis so that v head j pairs with k head
j % n_k_heads. That is the Qwen 3.5 pattern, not the repeat interleave that
Qwen 3 Next builds explicitly, so qwen4exp takes the else branch and its V is
segmented on the scale of K.
Reported by benklop.
(cherry picked from commit 353d753f595dc81634ae6130188b31f06018f5ae)
* llama: fix the qwen4exp PLE history seq_rm(-1) iterator invalidation and the fatal-warning build
ple_hist_rm recursed over ple_hist with a range-based for and the recursive call
erases the entry it is iterating when the whole sequence is removed (p0 <= 0,
p1 < 0), so the loop then increments an invalidated iterator. It is unreachable
today only because llama_memory_recurrent::seq_rm rejects seq_id < 0 before
llama_memory_hybrid_idx::seq_rm reaches the history, which is a guard in another
class. Advance past the entry before recursing.
Two smaller things in the same area:
- the n_toks sanity bound in ple_hist_state_read was the literal 64, which is
the value of LLAMA_MAX_PLE_HEADS, not of the quantity being checked. The
window is at most ple_ngram_size - 1 tokens, so the bound is
LLAMA_MAX_PLE_NGRAM - 1, eight times tighter.
- build_conv_state_at left mem_size unused, so -DLLAMA_FATAL_WARNINGS=ON does
not compile. Predates this series; drop the line.
(cherry picked from commit 6eba44a89d5f328eb4859b844e1d28fb564cbe3e)
* qwen4exp: include llama-impl.h explicitly for llama_mul_mat_hadamard
(cherry picked from commit b634fd4d250d181ef82bf78bd00c1ae3b96a7af6)
* convert: fix the qwen4exp lint and type-check failures
flake8 flagged an unused MmprojModel import, and ty flagged seven errors in
the PLE streaming path: eos_token_id can be absent, and _ple_map, _ple_path,
_ple_row_dim and _ple_rows_per_shard are all Optional at the declaration but
were dereferenced without narrowing.
The map is opened and the stride fixed before the first shard is written, and
_finish_ple_table only runs once every shard has landed, so the invariants
hold. Assert them so the checker can see it. A missing eos_token_id now raises
with the reason instead of a TypeError from int(None).
* llama: give the qwen4exp indexer cache its own tensor names
The indexer KV cache and the attention KV cache both named their tensors
cache_k_l%d, so the Meta backend matched the indexer cache against the
attention split pattern and aborted in handle_set_rows. Tag the names
instead, and mirror the indexer cache: it has one key head and its
projections are mirrored.
(cherry picked from commit a1cdc8181134659766763a17762545a1f0e5db7b)
* qwen4exp: double the Q split granularity for tensor parallelism
qwen4exp fuses the attention gate into attn_q.weight the same way qwen3next
and qwen 3.5 do, so a device boundary must fall on a whole q+gate pair or the
Q heads stop lining up with the K/V heads and attn_output rows.
(cherry picked from commit 6c9a592f0a425a459ab6efae3b897cf68460e244)
* qwen4exp: keep the indexer cache in step across server slots
The QSA indexer keeps a side cache addressed by the cells of the attention
cache, so cell j has to hold the same token in both: the top-k indices it
produces are applied to the attention KQ mask. init_batch already hands the
indexer the attention cache's slot layout rather than letting it look for its
own, but the restore path did not. state_read called llama_kv_cache::state_read
on the two caches in turn and each ran its own find_slot over its own occupancy.
That agrees only for as long as nothing has already pushed the two caches apart,
which is the property a restore is supposed to re-establish rather than one it
can lean on.
The failure path was the worse half, and it is reachable from the public API
with nothing more than a short buffer. Truncating a good blob at 35 offsets and
feeding it to llama_state_seq_set_data left the two caches disagreeing at 5 of
them, and every one of 23 truncations of a whole-context blob did. Four of those
five land inside the attention section, so the attention cache drops the
sequence and the indexer keeps it; only the cut that lands in the indexer
section gives the opposite direction. llama_kv_cache::state_read cleans up its
own cache and rethrows, so whichever way it falls, nothing is left to bring the
two back together. The server papers over this by clearing the slot when a
prompt cache load fails; a caller of llama_state_seq_set_data that does not is
left with an indexer addressing cells that no longer mean what it thinks.
llama_kv_cache::state_read_sinfo reports the cells a restore landed in, or takes
a copy of them, and state_read_meta uses a supplied layout in place of find_slot
once it has checked that those cells are free here too. The indexer now adopts
the attention cache's restored layout by construction instead of reproducing it
by coincidence, and a layout that does not fit fails the read rather than being
applied over cells that already drifted. The hybrid restore is wrapped so that
any failure drops the sequence, or for a whole-context restore the context, from
all three caches at once, which is a state they do agree on.
* kv-cache: clear the cache once when restoring a whole context
state_read walks the streams of the cache in turn, and for a whole-context restore
each stream went through state_read_meta, which starts by calling clear(). clear()
resets every stream at once, so each stream after the first threw away the streams
already restored, and the K/V buffers with them. A non-unified cache holds one
stream per sequence, so a context saved with N sequences in it came back with only
the sequence in the last stream that carried any cells - the highest sequence id.
A unified cache has one stream and never showed it.
The cache is now emptied once, before the loop, which is what a whole-context
restore means. A blob whose streams are all empty now empties the cache as well,
where before it left the old contents in place.
* kv-cache: check the mirrored slot layout on a whole-context restore too
state_read_meta only looked at the layout it was given on the single-sequence path.
A whole-context restore lays the cells out from 0 in both caches, so they agree as
long as they restore the same number of cells, but nothing checked that they did: an
indexer section belonging to some other context was read over cells the attention
cache had filled from a different one, which is the state the indexer must never be
left in.
* qwen4exp: give the PLE conv history its own mirrored recurrent row
n_embd_r() reserved n_conv + ple_conv_state() so that one cache_r_l row could
carry both the delta-net conv state and the PLE dilated conv history, but the
QWEN4EXP arm of get_split_segments only described n_conv. Under -sm tensor the
segment sum came up short by ple_conv_state() and llama_memory_recurrent
construction aborted in ggml_backend_meta_alloc_ctx_tensors_from_buft.
Widening the segment list is not the fix. The Meta backend propagates a view's
split descriptor from its parent unchanged, so a view of one sub-range of a
split axis is sized as the whole row on every device; declaring the PLE tail as
a second segment merely moves the abort to "shape mismatch for VIEW" at graph
allocation. The two histories also want opposite policies: the delta-net state
is split by head to match wqkv and ssm_conv1d, while per_layer_tok_embd,
ple_conv1d and ple_norm_conv are all mirrored, so every device computes the
whole dilated conv and needs the whole history. One tensor cannot be both, and
the split state has no per-segment mirroring.
Move the PLE history into its own cache_ple_r_l%d row, mark it MIRRORED, and
return n_embd_r() to n_conv. The row is allocated only on layers where is_ple
holds, so mirroring one 92160-element row per device replaces a 92160-element
tail on all 36 recurrent rows: the recurrent R footprint drops rather than
grows. build_conv_state_at now takes its width from the tensor it was handed
and keys its gather on that tensor, which also drops a cont of a strided view.
* no more ple_hist (use master version)
* llama: give the qwen4exp full memory context its indexer cache
graph_reserve() walks a full memory context, and qwen4exp builds its
sparse attention only when the context exposes an indexer cache. the
full-context constructor left ctx_idx null, so the reserved worst case
was the dense fallback: a smaller graph than the one decode executes.
ggml-alloc then had to grow the compute buffer on the first decode,
past the size reported at load.
with -np 4 -c 32768 -fa on -ctk q8_0 -ctv q8_0 on an IQ1_S qwen4exp,
the reserved CUDA0 buffer was 217.00 MiB against 275.71 MiB actually
used, and CUDA_Host 42.31 MiB against 191.14 MiB. reserving the sparse
graph makes both match exactly, in unified and non-unified cache mode.
Co-authored-by: Pascal <admin@serveurperso.com>
Assisted-by: Claude
* qwen4exp: shrink the PLE hparams storage
llama_hparams is held by value inside llm_graph_params and every llm_graph_input_*,
and llm_graph_params is a stack local in graph_reserve and process_ubatch, so its
width is paid on every worker thread stack.
is_ple_impl spent 2048 bytes carrying 512 bits. It is the one per-layer flag that is
not moved through the loader's uint32 array templates, so a bitset costs nothing in
call sites and also removes the uninitialized read that non-qwen4exp archs had, since
nothing filled the array for them.
The PLE head offsets and vocab sizes are token-space indices; the gather that consumes
them already truncates to int32, so 64-bit storage was never reachable. The gguf arrays
stay uint64 for file compatibility and are narrowed on load.
sizeof(llama_hparams) 34440 -> 31944, sizeof(llm_graph_params) 34872 -> 32376.
* llama: opt-in random-access mmap advice for host-resident gather tables
qwen4exp keeps per_layer_token_embd on the host: 26.8 GiB at IQ4_NL, read
by ggml_get_rows as 16 gathers of ~90-170 bytes per token, spread across
16 head regions ~20M rows apart. Measured over 4.75M gathers, no two
consecutive gathers land on the same 4 KiB page, so the readahead the
loader asks for buys nothing here and the whole table ends up cached to
serve about 4% of itself.
llama_mmap applies POSIX_FADV_SEQUENTIAL, MAP_POPULATE and a whole-file
POSIX_MADV_WILLNEED unconditionally. Those are right for streaming the
file once into buffers and wrong for whatever stays mapped afterwards.
Under LLAMA_MMAP_RANDOM the eager pull-in is skipped and the mapping is
advised random once every tensor has been read, so the load itself keeps
its sequential readahead. That alone drops the table to 4.4% resident but
serializes one NVMe latency per gather.
The second half is what pays for it: the PLE input already computes every
row index for the ubatch before the graph runs, so the pages those rows
fall on are handed to the kernel in one batch and the reads overlap.
POSIX_MADV_WILLNEED on POSIX, PrefetchVirtualMemory on Windows, which
takes the discontiguous ranges in a single call.
Off by default and off for every other model: the batched prefetch keys
off "this mapping was advised random", which nothing sets unless the user
opts in.
-c 512 --chunks 60, cold, IQ1_S, mean of 3:
default 35.3 s 26.82 GiB resident (100%)
advice only 104.5 s 1.19 GiB resident (4.4%)
advice + prefetch 34.2 s 1.19 GiB resident (4.4%)
PPL 4.2346 +/- 0.07862 in all three. IQ1_S KLD is unchanged in every
field, including Mean KLD 0.396070 +/- 0.001931 and Same top p 77.325%.
* llama: narrow the random-access mmap advice to the gather table
The advice was applied per mapping: every mapping the model kept got
POSIX_MADV_RANDOM plus a whole-file POSIX_FADV_RANDOM, and the eager
pull-in was skipped for every file. On qwen4exp that also hit
token_embd.weight, which sits 0.33 GiB past the PLE table in the same
shard and is read densely, not by sparse gathers. Measured over
-c 512 --chunks 60 on IQ1_S it fell to 8.45% resident, against 100% with
the feature off.
A model now nominates its gather tables (qwen4exp: per_layer_tok_embd)
and only those byte ranges are advised. The range is rounded out to
whole pages, which on this model takes in 832 bytes before and 192
after. token_embd goes back to 86.55% resident and the PLE table still
drops to 4.44%; smaps shows one VM_RAND_READ VMA of exactly the table
instead of one over all 27.16 GiB that stays mapped.
posix_fadvise is dropped from the narrowed path. POSIX_FADV_RANDOM
ignores its offset and length and marks the whole open file, and the
FMODE_RANDOM it sets is only read by page_cache_sync_ra() on the read()
path, which a fault on a MADV_RANDOM vma never reaches. POSIX_FADV_
DONTNEED does take a range, so the drop mode keeps it.
The eager pull-in is now skipped only for the files holding a nominated
table, and re-issued as WILLNEED over the rest of such a file, so other
shards load exactly as before.
prefetch_rows() keys off the tensor being nominated rather than off a
mapping-level flag, so the batched readahead lands only where the advice
did.
-c 512 --chunks 60, cold, IQ1_S, mean of 3, total wall:
default 32.50 s
whole mapping 30.05 s
narrowed 30.35 s
PPL 4.2061 in all three. IQ1_S KLD is bit-identical with the feature on
and off, including Mean KLD 0.396070 +/- 0.001931 and Same top p
77.325%. tg128 73.65 +/- 0.33 narrowed against 73.49 +/- 0.34 whole.
Assisted-by: Claude
* llama: fold the random-access prefetch into its own feature flag
LLAMA_MMAP_RANDOM_PREFETCH existed to measure the two halves of the feature
apart, and the measurement is done: on a cold cache over the same wikitext
run, MADV_RANDOM without the batched readahead takes 94.4 s against 36.7 s
for an untouched mapping, while the pair together take 34.1 s. Suppressing
the kernel's readahead only pays if we replace it, so the split let a user
select a 2.6x regression through a documented switch.
Keep the accessor, since the call site reads better than a mode comparison,
but derive it from the mode alone.
* FACP (Fewer Acronym Classes Please)
* qwen4exp: bias the QSA selection per block, not per cell
The QSA bias is a graph input, so it is pinned on the host and uploaded every
decode, and at -c 32768 -np 4 its twelve copies were 768 of the 815 MiB of
reserved host compute buffer.
Only one half of it needs a cell: whether the cell sits in the always-visible
tail, and whether its block was pooled. Both are properties of the block. The
other half - empty, other sequence, or in the future - is the plain visible/not
test the attention mask already carries over the same cells, so add that mask
instead of repeating it. The bias then holds one value per block.
A block sits wholly inside or wholly outside the tail because the tail starts on
a block boundary, so one value per block is exact. Cells no block covers keep
their -inf from the mask.
The mask is F16 and the bias F32, and a mixed ggml_add reinterprets the F16
buffer as float rather than converting it, so the cast is required.
reserved host compute buffer at -c 32768 -np 4:
--kv-unified 814.86 -> 238.86 MiB, CUDA0 721.07 -> 421.07 MiB
--no-kv-unified 214.86 -> 70.86 MiB, CUDA0 317.07 -> 265.07 MiB
Selection is unchanged: over 8192 tokens, four times the budget, every QSA
layer returns identical top-k indices and the logprobs are bitwise equal.
Two things a reviewer should know. A cell whose position divides past the last
block is guarded by an assert rather than handled, because no run reached it.
And the mask's same-position M-RoPE rule cannot fire for text and was never
exercised for images, so the 2D case is unverified.
* clean up code comments
* clean up new comments
* revert LLAMA_MMAP_RANDOM
* nits
* replace some changes with #27795
* improve the m-rope image for get_prev_tokens
* LazyChunkedTensor
* fix lint
* add some validations
* reduce input nodes
* trim output tokens
* nits
* some more sanity checks
* fix llm_graph_input_ple reuse
* exclude from webgpu test
---------
Co-authored-by: danielhanchen <danielhanchen@users.noreply.github.com>
Co-authored-by: danielhanchen <unslothshared@gmail.com>
Co-authored-by: Xuan Son Nguyen <son@huggingface.co>
Co-authored-by: Pascal <admin@serveurperso.com>
Co-authored-by: Sigbjørn Skjæret <sigbjorn.skjaeret@huggingface.co>
* spec : add DFlash2 support (local convolution + candidate selector) (#27342)
* support DFlash2
* Add p_min in DFlash2
Assisted-by: Claude Opus 5
* Revert unnecessary changes
Assisted-by: Claude Opus 5
* Revert draft sampling in rejection sampling
Assisted-by: Claude Opus 5
* Refactor code structure
Assisted-by: Claude Opus 5
* Delete embedding scaling
Assisted-by: Claude Opus 5
* Gate output transforms on DFlash2
Assisted-by: Claude Opus 5
* Optimize Dflash 2 cost
Assisted-by: Claude Opus 5
* Avoid using atoi
Assisted-by: Claude Opus 5
* Modify comments
Assisted-by: Claude Opus 5
* Move llama_model_dflash_selector_top_k to llama-ext.h
Assisted-by: Claude Opus 5
* Formatting
Assisted-by: Claude Opus 5
* Apply patch to fix the mrope bug
Assisted-by: Claude Opus 5
* fix ci
Assisted-by: Claude Opus 5
* Fix graph number calculation
Assisted-by: Claude Opus 5
* rename hid and unary
Assisted-by: Claude Opus 5
---------
Co-authored-by: Jian Chen <jianchen0311@gmail.com>
Co-authored-by: Xuan-Son Nguyen <son@huggingface.co>
* revert top-k.cu changes
---------
Co-authored-by: Zihan Zhang <tiancaizhangdaxian@sjtu.edu.cn>
Co-authored-by: Jian Chen <jianchen0311@gmail.com>
* ui : strip trailing container-format segments from parsed model names
* ui : show reasoning and modality icons on model options and search by modality
* ui : keep reasoning submenu visible regardless of model state
* ui : add show-org-name-in-trigger display setting
* ui : move model list into a submenu within the model selector
* ui : make model option hover and focus highlight override the active state
* ui : add raw model id tooltip to model selector options
* feat: Enable microphone input as default for audio models
* ui : fix eslint issues in chat form and model selector
* ui: show modality icons instead of file submenu in chat add menu
Assisted-by: pi
* chore: Format
* chore: Format
* ui: add ModelCapability enum and shared modality/capability icon constants
Assisted by: pi:GLM-5.3-Flash
* ui: derive modality badge icons and labels from shared constants
Assisted by: pi:GLM-5.3-Flash
* ui: split model option icons into capabilities and modalities
Replace the supportsThinking flag on ModelId with a capabilities object
keyed like ModelModalities, so future capabilities (tool calls, etc.)
slot in alongside reasoning. Icons and labels now come from the shared
CAPABILITY_ICONS/MODALITY_ICONS constants.
Assisted by: pi:GLM-5.3-Flash
* ui: replace per-conversation MCP overrides with per-conversation tool policy
MCP server enabled state is now global (server.enabled); per-conversation
control moves to disabled tool keys and categories seeded into each new
conversation. Aligns the add sheet with the dropdown options and flattens
MCP tool groups in the tools submenu.
Assisted-by: pi
* ui: keep tool policy migration running when defaults parse fails
A corrupt disabledToolKeys localStorage entry no longer aborts the
migration; it falls through with empty defaults so legacy MCP server
overrides still get converted.
Assisted-by: pi
* ui: fall back to global defaults when agentic flow has no tool policy
Passing empty disabled sets bypassed the global defaults and could
enable tools for callers that do not pass a policy yet.
Assisted-by: pi
* ui: align preferences section headers with their methods
The Reasoning Effort and Working Directory headers sat above tool
policy methods; move them above setCwd and setReasoningEffort. Also
clarify the disabled tools JSDoc: existing rows with an unset field
have an empty policy, defaults apply only when there is no active
conversation.
Assisted-by: pi
* ui: gate MCP server avatars on conversation tool policy
Servers whose tools are disabled for the current conversation (MCP
category or server-scoped key) no longer show as enabled for the chat.
Assisted-by: pi
* ui: drop unused MCP category toggle from tools panel hook
Per-conversation MCP control is server-granular; no component renders
a whole-category toggle, so remove the dead API.
Assisted-by: pi
* ui: skip MCP init when flow policy disables the MCP category
Resolve the effective tool policy before deciding whether to
initialize MCP so flows that will not send any MCP tools skip the
init work. Callers without a policy keep falling back to global
defaults.
Assisted-by: pi
* chore: format
* ui: restore reasoning section in mobile add sheet
The sheet rewrite dropped it; the desktop dropdown still has it.
MCP Prompts and Resources stay out of the sheet on purpose.
Assisted-by: pi
* ui: clear MCP server group key in enableAllToolsForServer
The group key disables every tool of the server regardless of
per-tool keys, so re-enabling a server from Settings did nothing
while it was set.
Assisted-by: pi
* ui: skip MCP init when no policy-enabled server remains
Extends the category-level check: the flow also skips MCP init when
every globally-enabled server has its server-scoped group key
disabled in the tool policy.
Assisted-by: pi
* ui: make Settings tools tab edit defaults with category toggles
Adds per-category checkboxes and a caption stating the tab applies
to new conversations; tool picks inside a chat only affect that
chat.
Assisted-by: pi
* ui: gate cwd picker and mention picker on effective tool policy
Both checked the global disabled set directly, so a conversation
that disabled file_search still showed search as available.
Assisted-by: pi
* ui: clean up tool key helpers and store docs
Documents getEnabledToolsForLLM properly, unstacks the JSDoc at
isEntryEnabled, makes setToolEnabled persist like setCategoryEnabled
(toggleTool now delegates to it), and routes the serverId-less MCP
branch of toolKey through getMcpServerToolsKey so both key formats
come from one place. Preferences banner comments become plain
comments so they no longer read as class member docs.
Assisted-by: pi
* ui: indeterminate group checkboxes and inert grayed rows
A category that is on with nothing enabled under it now shows the
mixed checkbox state instead of a checked box next to 0/N. Rows
grayed out by a disabled parent no longer stay clickable behind
opacity.
Assisted-by: pi
* ui: gate MCP prompt and resource capabilities on tool policy
hasPromptsCapability and hasResourcesCapability accept an optional
set of usable server ids; ChatFormActions resolves it from global
enablement minus the active conversation's policy. Restores the
per-chat gating the old mcpServerOverrides provided; callers without
arguments keep global behavior.
Assisted-by: pi
* ui: remove unmounted MCP submenu component
Never rendered anywhere; its entries are duplicates (prompts and
resources live in the attachment menu, servers in the add menu and
sheet) that would need capability wiring maintained for nothing.
Assisted-by: pi
* ui: fix model information dialog width on all screen sizes
The dialog sets container-type: inline-size, so auto width ignores
its contents and collapses to padding. Give it an explicit viewport
width on mobile and cap at 60rem on desktop.
Assisted-by: pi
* ui: scroll wide chat template in model information dialog
Long unbreakable Jinja tokens blew out the table and dialog width;
the block now scrolls horizontally instead of stretching.
Assisted-by: pi
* ui: use fixed table layout in model information dialog
Auto table layout sizes columns to content min-content, so the chat
template's long lines kept inflating the dialog despite the scroll
wrapper. Fixed layout pins the first column and gives the value
column a definite width the wrapper can scroll within. min-w-0 on
the grid item guards the same path on the grid side.
Assisted-by: pi
* ui: make model information dialog full-screen on mobile
Matches the settings dialog pattern: full viewport below md,
calc-sized and capped at 60rem on desktop.
Assisted-by: pi
* ui: stack chat template row in model information dialog
Label above the block in a single full-width cell, so the template
gets the whole table width and its horizontal scroll is usable on
narrow screens.
Assisted-by: pi
* ui: scroll model information header with the content
The base dialog header is sticky; this dialog overrides it to
relative so the title and description scroll away with the body.
relative keeps the header as the close button's containing block.
Assisted-by: pi
* ui: replace literal comment text in sheet group snippet
A // line inside the Svelte snippet rendered as visible text; use an
HTML comment.
Assisted-by: pi
* ui: let indeterminate state win over checked in group checkboxes
The checkbox indicator snippet renders the check icon whenever
checked, so the mixed state never showed. Pass the checked prop
as false while indeterminate.
Assisted-by: pi
* ui: initialize only policy-enabled MCP servers for a flow
ensureInitialized accepts an optional server id set; the agentic
flow passes the servers its tool policy leaves usable, so servers
disabled for the conversation no longer get connected. Callers
without arguments keep the global behavior.
Assisted-by: pi
* ui: derive group checkbox state in useToolsPanel
Moves the mixed-state derivation out of the submenu and sheet
snippets into one getGroupCheckState accessor; the snippets just
consume checked and indeterminate.
Assisted-by: pi
* ui: gate /prompt command on the conversation tool policy
The slash command's availability now follows the same rule as the
agentic flow instead of the global capability check, so it disables
itself when the conversation's policy leaves no usable MCP server.
Assisted-by: pi
* ui: remove dead MCP prompt menu trigger chain
The /prompt slash command is the surviving trigger; the menu-button
path (onMcpPromptClick, hasMcpPromptsSupport, showMcpPromptButton,
the MCP_PROMPT attachment item and its unrendered item arrays) has
no consumer left. Message display for inserted prompts is untouched.
Assisted-by: pi
* ui: render dash for mixed-state group checkboxes
The accessor refactor dropped the checked-and-not-indeterminate
guard, so the category-on flag won and the dash never showed. The
tooltip keeps using the raw parent flag since clicking a mixed
group still disables it.
Assisted-by: pi
* ui: fix group checkbox sticking checked after disable
Clicking a mixed-state group box let bits-ui optimistically flip
its internal checked flag; the derived checked prop did not change
across the transition (both mixed and off map to checked=false),
so Svelte never applied the settled value and the check icon stuck
while the count already read 0/7.
Pass the parent flag as checked and the mix as indeterminate, so
every group toggle changes checked; render the dash on top of a
checked box for the mixed state.
Assisted-by: pi
* fix: UI for Model Information dialog
* ui: keep MCP connections stable across policy switches
ensureInitialized folds the policy into its config signature, so
alternating two conversations with different policies tore down and
reconnected every server with health checks included. Tool collection
already filters by the flow policy, so initialize every
settings-enabled server instead and never pass a policy into the MCP
config. The duplicated policy-server check becomes one accessor on
ConversationPreferences.
Assisted-by: pi
* ui: remove dead MCP resources menu trigger chain
Same shape as the earlier prompt trigger cleanup: nothing renders the
MCP resources menu button, and the only live entry into resource
browsing is Settings > MCP Servers plus the attachment resource
picker. Drop onMcpResourcesClick, hasMcpResourcesSupport,
MCP_RESOURCES_CLICK, the AttachmentItemVisibleWhen enum and
hasResourcesCapability; the resources display, browser and picker
components are untouched.
Assisted-by: pi
* Add benchmark-only synthetic speculative acceptance to llama-server and llama-cli
* Address review comments
* Address review comments
* Add some comments in the code
* common : dedupe --n-cpu-moe / --spec-draft-n-cpu-moe override loops
* common : add --n-cpu-ffn to CPU-offload dense FFN weights of first N layers
* common : generalize llm_ffn_block_regex over the FFN regex, drop TODO
* hexagon: use non-host bufs by default and make the backend fully async
* hex-hb: remove optional hostbuf support and fix async copy
* hex-unary: relax supported unary check
* hex-bufs: use same get_alignment for host bufs
* snapdragon: bump android_platform to 34
* hex-rows: super hacky get/set rows for q8_0
* hex-get-rows: fix q8_0
* hex-get-rows: supprot for f16 and cleanup for q8_0
* hex-get-rows: generic macros and specialized thread funcs
* hex-get-rows: add DMA pipeline, vtcm_layout and kernel params
* hex-set-rows: fix q8_0 support, add dma and tracing
* hex-tests: override nmse threshold for HTP of Q8_0 quants
* hex-fa: add support for Q8_0 with inplace dequantizers
* hex-get-rows: simplify type dispatch
* hex-rows: simplify GET/SET_ROWS DMA pipeline
* hex-async: add events, set/get-tensor-async and rest of the async api support
* hex-repack: use slice instead of expert in repack functions
* hex-cpy: update event/async-cpy logging
* hex-set-rows: optimize smaller tensors
* hex-geglu: fix perf regression with larger tensors
* hex-get-rows: add missing header
* hex-set-rows: add missing header
* hex-bufs: ressurect GGML_HEXAGON_HOSTBUF but disable it by default
* hexagon: do not reject ops with non-heaxon buffers
* hex-get-rows: apply >=32 restriction only for q8_0
* hex-res: bump vtcm acquire timeout to 10 seconds
* hex-bufs: add support for cloning buffers between sessions to speed up tensor copies
* hex-async: rework event recording and batch flushing and integrate with meta backend
* hex-bufs: improved handling of repacked tensors
* hex-repack: handle get_tensor_2d offsets
* hex-dev: add support for devices with multiple NPUs
* hex-sync: add support for sync tokens to synchronize npu devices for async splits
* hex-mmap: cleanup mmap calls and add a retry for robustness
* hex-sync: add failsafe if sync wait gets stuck
* hex-sync: use sync_seq to check for completed events
* hex-sync: rotate tokens for extra robustness
* hex-devs: add supprot for legacy device names for now
* hex-bufs: add support for auto-cloning buffers from diff sessions
* hex-fusion: simplify and optimize htp-opnode fusion handling
* hex-sync: override opnode name so that it shows up in the profiles
* hex-trace: update scripts to handle multiple devices
* hex-sync: bump the size of the opbatch queue and number of sync tokens
* hex-cpy-sync: do not explicitly flush opbatches in cpy_tensor_async and add support for cpy-dma
* hex-sync: add graph-flush threshold to avoid single op batches
* hex-sync: add sync_peer so that we can flush peers we depend on during cross-device ops
* hex-bufs: introduce tensor->extra and shadow_bufs for repacking
* hex-l2: flush tiny tensors inline
* hex-sync: use explicit l2flush for sync tokens
* hex-extra: track weight flags via tensor extra
* hex-fence: rename sync to fence
* hex-repack: proper handling of set-tensor-2d in the shadow_buf
* hex-trace: remove obsolete opstage mask that we used for profiling
* hex-env: remove obsolete use_hmx variable
* hexagon: new unified run.py and build.py and updated docs
* snapdragon: update run script to auto-escapt test-backend-op -p argument
* hex-scripts: fix trailing spaces
* hex-scripts: fix flake8 warnings
* snapdragon: cleanup dst lib/bin dirs before copying new build
* hex-ops: add support for allreduce
* hex-ar: improved allreduce with dma pipeline
* hex-ar: align macros
* hex-ar: consistent use of fence_seq
* hex-ar: add AR_SELECT env var to select ALLREDUCE kernel or fallback
* hex-ar: add proper synchronize handling for ALLREDUCE
* hex-opbatch: looks like we now just rely on backend.synchronise to flush the batches, no need to flush them by threshold
* hex-ar: bump block size to improve dma efficiency
* hex-ar: fused ALLREDUCE+ADD
* hex-ar: cleaner fence buffer management
* hex-ar: futher allreduce tweaking to remove race conditions
* hex-ar: add simple solver and remove non-dma kernels
* hex-ar: add row-broadcast to fuse with bias ADD
* hex-fence: pass seq numbers via op_params
* hex-ar: allow for both entry/exit seq for completing entry wait
* hex-ar: align macros
* hex-ar: do not refetch broadcast row
* hex-fusion: move all fusion into opbatch::add_op for consistency with ALLREDUCE and things
* hex-fusion: fix incorrect MUL_MAT reordering
* hex-mm: make fused 2x and 3x matmuls more generic
* hex-fusion: move tensor fusion tagging to graph_compute
* hexagon: make sure to copy tensor->extra by value
* hex-get-rows: fix offset calc with row-chunking
* hex-repack: get_tensor_2d fixes for non-zero offsets
* snapdragon: make profile/trace scripts more robust and donot mix stdout/stderr by default
* hex-devices: use legacy device nameing by default to ease the transition
* hex-devices: hardcode CDSP domain IDs for current devices for now
* hex-optrace: improve multi-NPU timestamp alignment and overall handling of cycle values
* hex-optrace: more robust handling of the fence events
* ui : open MCP servers in a dialog from the chat form
Replace the MCP servers submenu with a single "MCP Servers" item that opens
a new DialogMcpServers dialog instead of navigating to the /mcp-servers route.
Assisted-by: pi
* ui : browse MCP resources from the server card
Make the Resources capability badge clickable so it opens the MCP resources
browser dialog, and drop the page-only chrome from SettingsMcpServers.
Assisted-by: pi
* ui : remove mcp-servers route and sidebar entry
MCP servers are now managed in a dialog, so drop the dedicated route and the
sidebar icon that navigated to it.
Assisted-by: pi
* ui : remove unused MCP servers submenu component
The submenu was replaced by the MCP servers dialog, so delete the component
and its export.
Assisted-by: pi
* feat(ui): add DialogSettingsChat dialog
* refactor(ui): switch SettingsChat to in-app section navigation
* feat(ui): open settings as dialog from sidebar
* refactor(ui): remove settings route and URL-based settings navigation
* fix(ui): adjust MCP dialogs for new base sizing
* chore: Formatting & linting
* feat(ui): make base dialog responsive and support sticky headers
* ui: move dialog close button to the sticky header
Assisted-by: pi
* chore: Formatting & linting
* ci : inline UI version resolution into ui-build.yml
* ci : build UI once and reuse the artifact in release jobs
Server jobs now extract the ui-build artifact into tools/ui/dist instead of npm-building the UI. Also removes the get-version job and the no-op -DHF_UI_VERSION flags.
Assisted-by: pi:Kimi-K3
* ui : disable the npm UI build by default (LLAMA_BUILD_UI=OFF)
The flag now only controls building the UI from source via npm. The UI
is still embedded by default from local tools/ui/dist or the prebuilt
download (LLAMA_USE_PREBUILT_UI=ON). CI jobs no longer npm-build the
UI; server-sanitize does not need node anymore.
Assisted-by: pi:Kimi-K3
* ci : rename the ui-build artifact to llama-ui.zip
Consistent with the other artifact names in the Actions summary.
Assisted-by: pi:Kimi-K3
* ci : clarify the windows artifact merge in release.yml
The windows-cuda/vulkan/sycl jobs build only the backend library;
llama-server (with the embedded UI) is injected into their zips from
the windows-cpu package during the release. State this in the job
comments and use accurate wording in the merge step.
Assisted-by: pi:Kimi-K3
* cuda: unblock mmq for MoE on sm_60
* cuda: duplicate mmq-config-pascal for dp4a and older
* cuda: reduce occupancy on non-dp4a pascal for Q2_K, Q4_K, Q5_K, Q6_K
This commit contains a fix for the conversion of NVIDIA Nemotron 3.5
Lightning which currently incorrectly converts when using a transformers
version later than 5.5.1.
When converting using [convert](https://github.com/ggml-org/convert) the
transformers version is 5.13.1 and this produces the following:
```console
WARNING:gguf.gguf_writer:Duplicated key name 'nemotron_h_moe.attention.head_count_kv', overwriting it with new value [0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0] of type ARRAY
```
This does not happen with transformers 5.5.1. The reason seems to be
that the configuration is different in later versions, for example when
using 5.13.1 the configuration block looks like this:
```console
transformers 5.13.1
raw has layers_block_type: True
autoconfig has layers_block_type: True
autoconfig layers_block_type: [
'linear_attention',
'moe',
'linear_attention',
'moe',
'linear_attention',
'full_attention',
'moe',
...
]
```
And with 5.5.1 we get:
```console
transformers 5.5.1
raw has layers_block_type: True
autoconfig has layers_block_type: True
autoconfig layers_block_type: [
'mamba',
'moe',
'mamba',
'moe',
'mamba',
'attention',
'moe'
...
]
```
In our conversion script we only match for attention, not full attention
which is causing this issue.
With the changes in this commit the output with transformers 5.13.1 will
be:
```console
(venv) $ gguf-dump models/NVIDIA-Nemotron-3.5-Lightning-30B-A3B-BF16.gguf | grep head_count_kv
INFO:gguf-dump:* Loading: models/NVIDIA-Nemotron-3.5-Lightning-30B-A3B-BF16.gguf
29: [INT32] | 52 | nemotron_h_moe.attention.head_count_kv = [0, 0, 0, 0, 0, 2, ...]
```
Resolves: https://github.com/ggml-org/llama.cpp/issues/27718
Refs: https://github.com/ggml-org/convert/actions/runs/32949047680/job/98116096069#step:5:2391
* metal: WIP chunked SSD SSM_SCAN kernels for multi-token prefill
* metal: drop scalar SSD path; MMA + sequential tail
* drop WIP ssm scan test noise
* remove state_from_dst and rename CS and NSG constants
* remove unrelated added whitespace padding
* added clarity to mma_tokens calculation
* added clarity to use_mma bool checks
* added comments to metal ssd op constants for clarity
* reserve K tokens for sequential kernel rollback snapshots
* reset concurrency between mma and seq tail
* remove print args no longer used
* fixed comment to no longer point to specific line
* add FC_SSM_SCAN so seq path skips token offlset unless it's mma tail
* added changes to new ssm.metal for rebase after ggml-metal.metal refactor
* specialize ssm_scan tail with a template instead of a function constant
---------
Co-authored-by: dpantaleoni <dominikpantaleoni@gmail.com>
Co-authored-by: forforever73 <690105611@qq.com>
* rpc: support apple RDMA as an RPC transport
* remove set_tensor micro optimization, rpc socket pinning per CR
* remove transparent reconnect
* trigger apple builds on RPC changes
---------
Co-authored-by: Ryan Churaman <rschu@meta.com>
* devops: use GGML_NATIVE=OFF for OpenVINO
Same as in other Dockerfiles.
Should fix#23100
* enable backend dl and cpu all variants
---------
Co-authored-by: Sigbjørn Skjæret <sigbjorn.skjaeret@huggingface.co>
* server: fix tool calls getting silently stripped with --prefill-assistant
Last assistant carries tool_calls + --prefill-assistant is on → request
flips into continuation mode, add_generation_prompt forced off, tail
rebuilt from reasoning_content + content only. Tool calls just vanish.
- Auto-continuation now skips trailing assistant msgs that have tool calls
- continue_final_message on those throws a clear error instead of
silently corrupting the prompt
- Regression tests included, red before / green after
Fixes#27588
Developed with AI assistance, disclosed per the contribution policy.
* server : address review: fail on prefill-assistant + trailing tool_calls
Move validation into oaicompat_chat_params_parse (next to the existing
two-or-more-assistant check) and remove it from common_chat_templates_apply,
which has no precedent for validation. Drop the regression tests.
Per review: --prefill-assistant with a trailing assistant message
containing tool calls is not supported and should fail loudly.
* add ccache-buckets action
* use ccache-buckets
* only save on master
* install python3-venv for hip
* add jq and python3 for cuda
* only delete caches older than 5 minutes
* metal : null-check ggml_metal_buffer_init result to avoid OOM crash
ggml_backend_metal_buffer_type_alloc_buffer used the result of
ggml_metal_buffer_init without checking for NULL. ggml_metal_buffer_init
returns NULL when the underlying Metal allocation fails (e.g. an
out-of-memory condition), and the following ggml_metal_buffer_is_shared(res)
call dereferences it, turning a recoverable allocation failure into a hard
crash (EXC_BAD_ACCESS). This is easy to hit on memory-constrained devices
such as iOS when a model/context exceeds the available Metal budget.
Log the failure using the existing GGML_LOG_ERROR convention and return
NULL so the allocator surfaces a diagnosable error up the stack instead of
crashing.
* cont : fix log
---------
Co-authored-by: Georgi Gerganov <ggerganov@gmail.com>
* grammar : accept "\-" escape in character classes
gbnf_escape_char_class() escapes '-' as "\-" but parse_char() rejected
that escape, so generated tool-call grammars failed to parse.
Assisted-by: Claude Code <claude@anthropic.com>
* tests : add parser test for "\-" in char classes
Assisted-by: Claude Code <claude@anthropic.com>
* tests : add integration test for "\-" in char classes
Assisted-by: Claude Code <claude@anthropic.com>
* tests : drop integration and parser tests
* metal : per-device tuned (Q, NE) for flash-attn vec (#25750)
* rebase Q-generic FA vec body from 01dc93607 (#23114)
* add 53 f16 (Q,NE) flash-attn vec instantiations (vec 80 -> 133)
* add FA vec (Q,NE) tuning table + dispatch wiring + SMEM cap fallback
* add FA vec (Q,NE) perf sweep
* fill tuning result
* fold family table into a per-family representative SKU
* refactor tuning result format
* extend FA vec tuning to quantized KV caches
* sync fa vec tuner bucketing with runtime, use pointwise tuning regret
* update tuned table
* format and cleanup
* prefix fa_vec tuning procs with ggml_backend_metal_tuning_, drop unused fa_vec_override_active
* add device id -> token lookup for the offline tuning tool
* add ggml-metal-tuning skeleton
* add op-agnostic perf cell + median timing for the tuner
* add FA-vec graph build + tensor init to the tuner
* tools : add FA-vec (Q,NE) sweep, compression and table emit
* cool down and re-measure the dirty window on thermal drift
* test-backend-ops : replace the FA vec tune mode with a bounded (Q,NE) slice
* tools : document the Metal tuner, point the table comment at it
* abort on unknown KV type, single-source fa_vec_legal_ne
* cleanup
* honor -o in the FA vec (Q,NE) slice
* retune FA-vec (Q, NE) under a pointwise no-harm gate
* cont : add fa-vec tunings for M1 Pro, M2 Ultra, M5 Max
---------
Co-authored-by: Georgi Gerganov <ggerganov@gmail.com>
* metal : per-op source split + parallel compile (#24021)
* preliminary extract common header
* op source split
* split metallib into 8 libs && load in parallel
* derive kernel->library routing from functionNames
* x-macro lib list + underscore filenames, dedup QK_NL, MRC fixes
* op source split 8 to 20
* improve robustness of source fallback
* clean up
* change bool -> atomic_bool
* only prepend headers that source actually includes
* no semaphore, use GCD global queue
* dedup library compile path, fix NSError lifetime, rename gla
* relocate upstream concat/rope_back/repeat kernel changes into split files
* move ggml-common.h from common.h into dequantize.h to shrink binary size
---------
Co-authored-by: lvyichen <lvyichen@stepfun.com>
* metal: add col2im_1d op (f32/f16/bf16) (#25176)
* metal : add set_rows with src0 f16 (#25434)
* metal : add CONV_2D_DW (depthwise convolution) support (#21565)
* metal : add Q2_0 support (#25419)
* metal: fuse snake activation (mul, sin, sqr, mul, add) (#25459)
* ggml-metal: FWHT kernel for metal backend (#25924)
* metal : port new kernels into the split sources
Move the kernels added on master after the split (lightning indexer,
DSv4 hyper-connections, silu_back, f16 bin ops, TQ2_0, the flash-attn KV
dequantization pass, rope offset/inplace, ssm_scan rollback, packed q8_0
dequantization and the tensor-API mat-mat K clamp) into the corresponding
kernels/*.metal sources. Copied verbatim, no functional change.
---------
Co-authored-by: lvyichen <lvyichen@stepfun.com>
Co-authored-by: Georgi Gerganov <ggerganov@gmail.com>
`repetition_penalty` is standard HF key for repetion penalty.
Currently, only `penalty_repeat` is mapped, read `repetition_penalty`
and map it to `metadata.sampling_penalty_repeat`.
* ci : apply ccache-clear with older/min/dry-run to all ccache jobs
Assisted-by: llama.cpp:DeepSeek-v4-Flash-0731
* ci : install gh in ccache-clear if missing (container jobs)
The ccache-clear action relies on the gh CLI, which is not present in
container-based jobs. Install it on demand so those jobs can clear caches.
Assisted-by: llama.cpp:DeepSeek-v4-Flash-0731
* ci : install gh via apt repo in ccache-clear
The install.sh script used previously is no longer served (404). Switch to
the official GitHub CLI apt repository, which is still available.
Assisted-by: llama.cpp:DeepSeek-v4-Flash-0731
* ci : pass --repo to gh cache commands in ccache-clear
In container jobs gh cannot auto-detect the repository from git, so
gh cache list/delete fail with 'failed to run git: not a git repository'.
Pass the repository explicitly via --repo using GITHUB_REPOSITORY.
Assisted-by: llama.cpp:DeepSeek-v4-Flash-0731
* ci : drop -new suffix from vulkan ccache key
The -new suffix was only needed to force a fresh cache. With
ccache-clear now evicting stale caches, the original key can be used
again. The old ccache-vulkan-ubuntu-24.04-arm-new entries still match
the ccache-clear key prefix and are cleaned up automatically.
Assisted-by: llama.cpp:DeepSeek-v4-Flash-0731
* ci : fix ccache-clear date parsing on macOS (BSD date)
macOS ships BSD date, which has no -d option. The older cutoff check
was silently disabled there: 'date: illegal option -- d' errors in the
log and the loop was only stopped by the min limit, risking deletion
of caches not older than the cutoff (e.g. saved by a concurrent job).
Parse the ISO-8601 timestamps with GNU date when available and fall
back to BSD date otherwise (TZ=UTC, fractional seconds dropped).
Assisted-by: llama.cpp:DeepSeek-v4-Flash-0731
* ci : extract ccache-clear logic into scripts/ccache-clear.sh
The composite action now consists of a dedicated step that installs the
GitHub CLI when missing (e.g. in container jobs) and a thin step that
calls the new script. The script follows the make-release-checks.sh
conventions (usage/env header, set -euo pipefail, CLI flags) and only
checks that gh is available. The action inputs are unchanged, so the
workflow steps are untouched.
Assisted-by: llama.cpp:DeepSeek-v4-Flash-0731
* ci : remove unused apple ccaches
* DSV4: sm tensor
* set coarser granularity for head splits
* fix dspark
* add model saving for dsv4 + allow dflash to return on specific device
* add comment about dsv4 seq_rm
* simplify
* add shared expert delayed allreduce
* remove special test for dsv4
* readme : update links
* readme : update maintainer PRs list
Add the new members of the `ggml-org` `maintainers` team to the
author filter of the maintainer PRs link (nikwen, marty1885,
Titaniumtown), keeping the canonical team ordering. The list now
matches the team exactly (35 members).
Assisted-by: pi:llama.cpp/Qwen3.8-27B
Run test-llama-archs with 1 to 4 GGML_METAL_DEVICES, mirroring the
existing CUDA runs, and dispatch the job unconditionally since the
per-backend guards now decide what to run.
Assisted-by: llama.cpp:DeepSeek-v4-Flash-0731
The device_info loop iterates over the discovered devices and gets
the available and total memory counts. With the CUDA backend (and
possibly others too) this requires creating a GPU context, which,
in case of CUDA, results in a 550 MB VRAM allocation.
For this information to be used in any way, the log verbosity must
be set to LOG_LEVEL_TRACE. If it's not, including in the default
configuration, the contexts get created, memory sizes get queried,
then the log function quietly discards the data.
In certain cases the user may not want to use any GPU resources.
The device_loop iteration is the only place touching the GPU that
cannot be skipped.
Fix by checking the verbosity level and skipping the loop if there
would be no output.
* DeepseekV4: fix rollback with multi-seq
* fix model loading
* make pending rollback single use
* only clear cache for seq_id for full load
* add assert for compress ratio
* make graph topology static
* pass true instead of flags in clear_compressed
* cont : clean-up + TODOs
---------
Co-authored-by: Georgi Gerganov <ggerganov@gmail.com>
* ui : add browser-style conversation tabs store
Track open conversation tabs in order, persisted to localStorage and
pruned against the loaded conversation list on init. The chat layout
syncs the route's tab on every navigation, so any way of reaching a
conversation opens a tab for it.
* ui : add temporary new-chat tabs
New-chat tabs are unsaved conversations carrying a temporary id used
directly as the route (#/chat/<id>). They live in memory and are only
persisted to the database - keeping the same id so the route and tab
stay stable - when the first message is sent. Deleting one drops it
without confirmation, and deleting conversations now closes their tabs.
* ui : render conversation tab bar in chat layout
Desktop-only tab bar above the chat screen, one tab per open
conversation or new-chat tab. The active tab follows the route id;
clicking navigates, middle-click or the close button closes (switching
to the left neighbor), and a trailing + starts a new chat. Tabs appear
only on chat-id routes; the bare #/ new-chat view has none. The bare
route stays put unless a prompt/model deep-link routes it to a new-chat
tab.
* ui : route new-chat entry points through tabs
The sidebar New chat item, Cmd+Shift+O, the search page and the
arrow-key fallback now open a new-chat tab instead of navigating to the
?new_chat URL, which is removed. New chat is no longer a special route
but a tab like any other conversation.
* ui : track sidebar expanded state in a shared ui store
Move the desktop sidebar expanded/collapsed state out of deviceStore into a
dedicated uiStore so the chat tab bar can react to it.
Assisted-by: pi
* chat : add opt-in conversation tabs setting
Add a Display setting that turns browser-style conversation tabs on or off,
enabled by default.
Assisted-by: pi
* chat : add browser-style conversation tabs with a new-chat screen
Track open conversations as tabs above the chat, one per open chat, plus a
single New chat tab for the bare `#/` route. New chat is just the `#/`
screen - no temporary conversations - and its tab is dropped when navigating
away. Sending the first message creates a real conversation and opens a tab
for it.
Assisted-by: pi
* chat : turn tab bar into a horizontally scrollable carousel
Make the tab bar a horizontally scrollable carousel with edge scroll buttons
and active-tab centering, and align its styling with the sidebar.
Assisted-by: pi
* chat : restyle the scroll-to-bottom button to match tab styling
Assisted-by: pi
* chat : add close-tab keyboard shortcut
Assisted-by: pi
* chat : soften tab bar fade and dim inactive tabs
Assisted-by: pi
* feat: Add stop button to tabs
* refactor: Componentize
* ui : fix carousel scrollability detection
Observe the content wrapper as well as the container, since adding overflowing items does not change the container's own box size. Also expose an onScrollableChange callback.
Assisted-by: pi
* ui : add unified ScrollCarousel component
Single carousel component with top/center variants, gap and scroll options, and hover-revealed chevrons. Rename the HorizontalScrollCarousel accessibility story accordingly.
Assisted-by: pi
* ui : migrate carousels to ScrollCarousel
Switch the settings mobile header, attachments list, thumbnail strip, and MCP resources to the unified component, and drop HorizontalScrollCarousel.
Assisted-by: pi
* ui : improve chat tabs carousel UX
Scroll newly added tabs into view, fade overflowing tabs at the edges, and hide the New chat button while a new-chat tab is open.
Assisted-by: pi
* refactor: Naming
* chat : add keyboard shortcut to jump between conversation tabs
Shift+Cmd/Ctrl+Left/Right cycles the open tabs, mirroring the existing
Shift+Cmd/Ctrl+Up/Down conversation navigation.
Assisted-by: pi
* chat : make the whole tab item act as a link
The full tab is now a link instead of only the inner label button, while
the stop and close buttons stay interactive by swallowing their clicks.
Assisted-by: pi
* chat : adjust tab bar width and use a shared offset variable
Widen the tab bar for the expanded sidebar and rename the tab bar height
variable to --chat-tabs-offset with a smaller value so the chat screen
min-height accounts for the overlay without overshooting.
Assisted-by: pi
* chat : account for the tab bar offset in the assistant min-height
Subtract the tab bar offset when it is shown so the last assistant message
does not overflow the available viewport space.
Assisted-by: pi
* refactor: Post-review fixes
* ui : restore deep links on the chat start page
- handle ?model selection, with ?load=true eager router loading
- ?q now creates a conversation, sends the prompt, and clears the params
- show the not-available-model dialog for unknown models
- never block mount on the conversation list
Assisted-by: pi
* ui : fix tab item link nesting and centralize tab constants
- the tab anchor covers the whole item while stop/close stay siblings,
so interactive elements are never nested inside the anchor
- cmd/ctrl/middle clicks are left to the browser (new window)
- extract the tab labels, the active-tab data attribute, and the
sidebar-offset max widths into constants
Assisted-by: pi
* ui : tidy scroll carousel hook and keep mobile header arrows on
- drop the dead scrollLeft/scrollRight helpers and the unused
onScrollableChange/scrollBy props
- init the carousel once instead of inside a derived
- restore items-start on the center variant
- always show the settings header arrows on touch
Assisted-by: pi
* ui : keep the new-chat tab across reloads and fall back on close
- the new-chat sentinel is no longer pruned on init, so reloading on
the bare new-chat route keeps the tab the user is on
- closing the active conversation falls back to the new-chat screen
when Conversation tabs are off
Assisted-by: pi
* ui : don't block startup on the conversation list
- prune persisted tabs after the list loads in the background instead
of awaiting it during init
- openNewChat now returns void; its return value was never read
Assisted-by: pi
* ui: fix routing nits
* chore: Update doc comments
* refactor: Mark fire-and-forget openNewChat calls as `void`
* chat: fix the deep-linked prompt, the tab width and the tab shortcuts
The chat start page creates the conversation and hands the prompt over
to the chat route, which still sees it in the query string. Sending it
on both sides queues the second copy as a pending message, which shows
up as a stray user bubble once the answer lands and vanishes on reload
since it never reaches the database.
The tab bar takes the max width of the collapsed sidebar while it is
expanded, and the other way round.
The tab list is pruned against a snapshot of the loaded conversations,
so a conversation created while that list is still loading loses its
tab even though the route just opened it. The active tab then falls out
of the list and the cycling shortcut jumps to an edge on every keypress
instead of moving one tab over. Tabs synced from the route are kept as
they are, only the persisted ones are pruned.
The rich chat input claims ctrl or alt with shift and an arrow for its
badge-aware word jump, which now belongs to the tab cycling shortcut.
Holding shift hands the key combination over, the plain word jump is
unchanged.
The close-tab shortcut consumes the event before checking whether the
setting is on, and the logo background loses its importance flag.
---------
Co-authored-by: Pascal <admin@serveurperso.com>
* fit: also take into account n_streams
* server: make the draft context follow the target context
With a non-unified KV cache the target context now holds n_ctx_train
tokens per sequence, while the draft context was still created with
n_ctx = 0 and fell back to n_ctx_train / n_streams per sequence. A slot
filled beyond that point makes the draft batch fail to decode, and the
server answers 500 on the request.
The draft context now takes its size from the target context, so both
hold the same number of tokens per sequence. Contexts that share their
cells with the target no longer need the kv_size override.
The memory reserved for the draft model before fitting is measured at
the largest context the target can take, since the draft context grows
with the target and a fixed byte margin cannot express that.
* fit: take an optional second model into account
Illustrates the alternative discussed on the draft context fix. The
memory of a draft or MTP context is currently handed to the fit as a
fixed byte margin, which cannot express a memory that grows with the
context the fit is still deciding on.
common_fit_params now takes an optional second model that shares the
devices of the main one. Its context follows the main context and its
memory is measured again whenever that context changes, so the reduce
path stays exact instead of conservative. A model that cannot be
measured on its own, such as a shared cell MTP context, is skipped with
a warning and the main model is fitted alone.
This drops the reservation block in the server, which no longer has to
probe the trained context size of the target to guess an upper bound.
---------
Co-authored-by: Pascal <admin@serveurperso.com>
* Revert "ci : disable ubuntu-rocm (#26969)"
This reverts commit 9558fa44c9.
* ci: set ccache compiler_check=content for ROCm build
The ROCm toolchain is pip-installed fresh on every run, so the clang binary's
mtime changes each time. With ccache's default compiler_check=mtime that
invalidates the whole cache and warm builds only reached ~70% hits. Hash the
compiler contents instead so the cache survives toolchain reinstalls.
* Update ccache size to 1GB
We're waivering with so many architectures built, we need a bigger
ccache limit.
* merge fix
---------
Co-authored-by: Jim Wu <ywu@xilinx.com>
Co-authored-by: Sigbjørn Skjæret <sigbjorn.skjaeret@huggingface.co>
* ci : add older, min and dry-run options to ccache-clear
Assisted-by: pi:llama.cpp/Qwen3.8-27B
* pi : add note about not wrapping lines in PR descriptions
[no ci]
Assisted-by: pi:llama.cpp/Qwen3.8-27B
* opencl: fold the gpt-oss MoE bias adds into swiglu_oai
Default on, opt out with GGML_OPENCL_FUSE_MOE_BIAS_GLU=0.
* opencl: fold the MoE down-projection bias into the combine
Default on, opt out with GGML_OPENCL_FUSE_MOE_BIAS_COMBINE=0.
The 'Create release' step had no id, so steps.create_release.outputs.id
resolved to an empty string in the 'Upload nightly-tag.txt' step. The
uploadReleaseAsset call then hit /releases//assets and failed with HTTP
404 (Unhandled error: HttpError), e.g. run 32513839499.
Add id: create_release to the step; the action already exposes the id
output.
Assisted-by: pi:llama.cpp/Qwen3.8-27B
* ci : run ccache-clear as the last step of release jobs
Assisted-by: pi:llama.cpp/Qwen3.8-27B
* update disabled job too to force rebase
---------
Co-authored-by: Sigbjørn Skjæret <sigbjorn.skjaeret@huggingface.co>
Similar to ggml/scripts/release.sh: validates repo state, creates a
release candidate branch (llama-rc-vX.Y.Z), bumps LLAMA_VERSION_* in
CMakeLists.txt and commits the version bump.
Usage: ./scripts/release.sh [major|minor|patch] [--dry-run]
Assisted-by: pi:llama.cpp/Qwen3.8-27B
* ui : rework the settings registry into ordered raw-data sections
SETTINGS_REGISTRY becomes an ordered SettingsSectionEntry[] array; the
array order is the sidebar display order. Section titles, color mode
options and title radio options are declared inline in their section or
entry. Entries gain showInUi; MCP servers, the system-message toggle and
the title LLM flag become hidden entries of their own section.
Derived values (config defaults, help info, chat sections, numeric field
lists, syncable parameters) are still derived here; they move to their
actual consumers in follow-up commits.
* ui : extract settings localStorage persistence into SettingsService
Stateless load/save of the settings config and user-override keys, plus the
legacy theme key migration. Business logic (default merging, mobile
sendOnEnter default, applying the migrated theme) stays in the store.
* ui : move the settings exit route into ROUTES
SETTINGS_FALLBACK_EXIT_ROUTE is just a route, so it lives with the other
routes as ROUTES.SETTINGS_EXIT.
* ui : derive the syncable parameter list in the parameter sync service
The syncable parameter mapping is only consumed by the sync service, so
derive it there from the registry instead of exporting it from the
constants file.
* ui : restore isPrivate for API key masking
* ui : clean up settings registry and router fetch guard
Drop the per-entry section field (duplicates the parent slug and is
never read) and guard the router model fetch on fields?.length so the
Tools/Import-Export pages with empty fields are excluded again.
Assisted-by: pi
* ui : merge sampling and penalties settings into one section
Assisted-by: pi
As agreed in ggml discussion #1579, the official semver releases now
include a nightly-tag.txt asset containing the tag of the corresponding
nightly release (e.g. b10485). The Web UI assets are published to the
HF bucket under the nightly tag, so this makes them discoverable for
each official release.
- make-release-desc.sh: expose the resolved nightly tag as a
nightly_tag output
- make-release.yml: create nightly-tag.txt from that tag, upload it
as a release asset (skipped on dry-run), mention it in the release
body and in the dry-run summary
Assisted-by: pi:llama.cpp/Qwen3.8-27B
* Add DMMV Q4_K and Q6_K ESIMD kernels
Configure cmake build with -DGGML_SYCL_ESIMD=ON to enable.
Signed-off-by: Todd Malsbary <todd.malsbary@intel.com>
* Refactor ESIMD kernels to share common code
Signed-off-by: Todd Malsbary <todd.malsbary@intel.com>
* Move control of ESIMD from compile to runtime
Signed-off-by: Todd Malsbary <todd.malsbary@intel.com>
* Use ESIMD by default when available
Signed-off-by: Todd Malsbary <todd.malsbary@intel.com>
* Fix possible error when using ESIMD by default
While not an issue in the current version, this will become an
issue when additional QK ESIMD kernels are added (such as Q2_K).
Signed-off-by: Todd Malsbary <todd.malsbary@intel.com>
* Add explicit unroll to ESIMD kernels
Signed-off-by: Todd Malsbary <todd.malsbary@intel.com>
* Tidy up ESIMD kernels a bit
Signed-off-by: Todd Malsbary <todd.malsbary@intel.com>
* Add a reordered Q2_K MMVQ kernel
Signed-off-by: Todd Malsbary <todd.malsbary@intel.com>
* Add DMMV Q2_K ESIMD kernel
Signed-off-by: Todd Malsbary <todd.malsbary@intel.com>
---------
Signed-off-by: Todd Malsbary <todd.malsbary@intel.com>
Resolve the TODO in test_flash_attn_ext: the branch that creates V as a
sub-view of K (MLA-based models) was hardcoded for the 576/512 head shapes.
Add a v_is_view_of_k test case parameter (default false) and select the
sub-view branch on it; the existing 576/512 (DeepSeek MLA) cases now pass it
explicitly, so the test coverage is unchanged.
Also add more V-is-sub-view-of-K cases: the 320/256 (Mistral4 MLA) and
192/128 head shapes, and full views with equal head sizes (128/128 F16,
64/64 q8_0).
Assisted-by: pi:llama.cpp/Qwen3.8-27B
* Add DMMV Q4_K and Q6_K ESIMD kernels
Configure cmake build with -DGGML_SYCL_ESIMD=ON to enable.
Signed-off-by: Todd Malsbary <todd.malsbary@intel.com>
* Refactor ESIMD kernels to share common code
Signed-off-by: Todd Malsbary <todd.malsbary@intel.com>
* Move control of ESIMD from compile to runtime
Signed-off-by: Todd Malsbary <todd.malsbary@intel.com>
* Use ESIMD by default when available
Signed-off-by: Todd Malsbary <todd.malsbary@intel.com>
* Fix possible error when using ESIMD by default
While not an issue in the current version, this will become an
issue when additional QK ESIMD kernels are added (such as Q2_K).
Signed-off-by: Todd Malsbary <todd.malsbary@intel.com>
* Add explicit unroll to ESIMD kernels
Signed-off-by: Todd Malsbary <todd.malsbary@intel.com>
* Tidy up ESIMD kernels a bit
Signed-off-by: Todd Malsbary <todd.malsbary@intel.com>
* Add DMMV Q5_K ESIMD kernel
Signed-off-by: Todd Malsbary <todd.malsbary@intel.com>
* Remove redundant copyright notice
Signed-off-by: Todd Malsbary <todd.malsbary@intel.com>
---------
Signed-off-by: Todd Malsbary <todd.malsbary@intel.com>
* opencl: keep the vocab-scale K-quant lm_head on the CPU on the Adreno A7X
* opencl: revise comments
---------
Co-authored-by: Li He <lih@qti.qualcomm.com>
The Tensor API mat-mat path of kernel_mul_mm (GGML_METAL_HAS_TENSOR) fed a
static K=32 tile to the matmul2d op on every iteration. On the last, partial
K tile (ne00 % 32 != 0) the src1 slice extends past the K extent of the
tensor, and the op reads those out-of-bounds elements (undefined behavior per
the MSL specification, section 2.22.2). Depending on stale memory contents,
this corrupted the result or produced NaN.
Make the matmul2d op use dynamic_extent for K, and clamp the K extent of both
operand tensor views to the remaining valid K range (min(32, K - loop_k)) per
iteration, so the op reads exactly the valid K range on every iteration
(mirroring the tail handling of the MPP matmul2d examples). On K-aligned
inputs the clamp degenerates to the full 32-wide tile: the only difference
from the static-K op is that the dynamic-K op derives K from the operand
extents and edge-checks the tile against the tensor extents (a handful of
integer ops per iteration).
Add test-backend-ops MUL_MAT cases with K not a multiple of 32 to exercise
the unaligned K path.
Assisted-by: pi:llama.cpp/Qwen3.8-27B
* opencl: decline KV-convert flash_attn variants on Adreno A7X (compiler SIGSEGV)
The Adreno 740 (A7X) compiler E031.41 crashes inside clBuildProgram when
building the flash_attn programs whose KV path is mixed-type or dequantized:
flash_attn_f32_f16, flash_attn_f32_q8_0, flash_attn_f32_q4_0. It is a driver
crash rather than a compile-error return, so build_program_from_source_ex()
cannot catch it. The uniform f32 and f16 programs build correctly.
Decline the three KV-convert variants on the A7X in supports_op so they never
lazy-compile; those attention layers run on the CPU backend instead. Same
idiom as the existing Intel DK=512 and X1E carve-outs.
test-backend-ops FLASH_ATTN_EXT on the 740: 226 OK / 0 FAIL, previously exit
139. Other parts are unaffected - the gate is dead code there.
* opencl: fix q6_K flat mul_mat on older Adreno E031 compilers, gated
kernel_mul_mv_q6_K_f32_flat produces ~10x-wrong output on the older Adreno
E031 compilers while q4_K and q5_K are correct. Four codegen defects, each
confirmed on-device against the CPU reference:
1. 64-bit ulong arithmetic is miscompiled, so every weight and scale read
hit the wrong address - the primary cause, and why q5_K (int offsets)
was unaffected. The block index is computed in int and widened only
inside the pointer expression.
2. The vectorized dequant (int4/float4 bit-ops, convert_*4, dot()) is
miscompiled; the 6-bit weights are reconstructed and the dot done
scalar.
3. vload4 of the f32 activations is miscompiled; replaced by a
scalar-indexed load.
4. The accumulation is miscompiled unless a side effect forces the partial
sums to materialize. A printf under a guard the compiler cannot prove
false acts as a zero-cost optimizer barrier; its placement is
load-bearing.
The defect tracks the compiler, not the GPU generation: it reproduces on
E031.38 (Adreno 642L) and E031.41 (Adreno 740) and is fixed by E031.45
(Adreno 619), so the workarounds are gated on the compiler version. Where
they are not needed they cost real throughput - 42.4 -> 35.1 GFLOPS on an
Adreno 840 q6_K GEMV. The explicit compiler-type check is required, not
redundant: newer_than_or_same() is false for every non-E031 compiler, so
negating it alone would enable the workarounds on E17 and DX.
test-backend-ops MUL_MAT is 919/919 on the Adreno 740, 642L, 619, 840 and
850; the 740 and 642L were 909/919 before. The 642L additionally needs the
A6X per-kernel-program support to reach these tests at all.
* ui: Extract server stream lifecycle from chatStore into ChatStreamManager
Discovery, attach/replay, resume retry and the remote-running snapshot
formed a cohesive cluster inside chatStore. It now lives in
chat-streams.svelte.ts as ChatStreamManager, owned by chatStore, which
keeps the public entry points as delegates so components are
unchanged. chatStore: 2877 -> 2418 lines.
* ui: Extract user interaction gates from agenticStore into AgenticGates
Tool permission requests, turn-limit continue prompts and queued
steering messages are the state the loop waits on between turns. They
had no coupling to session state, so they now live in
agentic-gates.svelte.ts; agenticStore keeps delegates so components
are unchanged. agenticStore: 1196 -> 1073 lines.
* ui: Compose MCP resources under mcpStore.resources
Resource state was a second import scope next to mcpStore. Consumers
now go through mcpStore.resources, so the MCP surface is one store;
mcp-resources.svelte.ts stays a separate file owned by mcpStore.
* ui: Reorganize stores into domain namespaces
* fix: Update stale doc comments
* ui: Consolidate conv running-state into a chat activity ledger
Running-state was split across chatStore.chatLoadingStates (local
pipes), ChatStreamManager.remoteRunningConvs (backend sessions) and
attachingConvs (attach lifecycle), unioned by hand in
getAllLoadingChats and cross-cleaned by setChatLoading calling
streams.clearRemoteRunning - the 'spinner ghosts until tab toggle'
workaround.
chatActivityStore now owns both sets with one transition per event:
markLocal / localEnded (local pipe end also drops the stale remote
hint, no cross-owner call) / applyRemoteSnapshot (diffed). The
sidebar reads chatStore.activity.loadingConvs through the unchanged
getAllLoadingChats entry point.
Consequences:
- isStreamingActive and its five manual writers are gone; isStreaming()
now reports whether the active conversation has a live streaming
pipe, which is what all four consumers (assistant row, stop action,
context gauge, chat screen) actually check
- isLoading/isReasoning become derived from the per-conv maps plus
the active conversation, dropping the manual resync in
syncLoadingStateForChat and clearUIState
- attachingConvs and the last-attach coordination disappear from
ChatStreamManager
- getAllStreamingChats (no consumers) is removed
* ui: Give store collaborators narrow host interfaces
Collaborators took 'host: typeof <store>', i.e. the store's entire
public surface, which is how chatStore's streamChatCompletion,
createAssistantMessage, getApiOptions and setStreamingActive got
widened to public. Replace with per-collaborator interfaces carrying
only the members each one drives:
- ChatStreamHost (chat/streams) - activity, processing, streaming
states, abort controller, loading/streaming setters
- ChatFlowsHost (chat/flows) - streaming core, message creation,
per-conv state setters
- McpHealthHost (mcp/health) - connection registry + reconnection
- ModelPropsHost / ModelStatusHost (models) - model rows, feed
updates; the managers write modalities/status back onto the host's
rows, so those members stay writable
- ConversationsPreferencesHost (conversations) - the active row and
the conversation list
The store classes now declare 'implements <Host>' so the contract is
visible at the class level, and the 'import type { <store> }' back
references in the collaborators disappear entirely - the host
contract is local to each collaborator file, and collaborators can
no longer reach around their slice. Members stay public (structural
typing), but the collaborator side is now compiler-enforced.
* test: Chat Activity store test
* refactor: Cleanup
* chore: Remove legacy architecture docs
* ui: Memoize findMessageIndex for the streaming hot path
Streaming looks up the same message index on every chunk, a linear
scan of activeMessages each time. Cache the last lookup and reuse it
after validating the id still sits at the same position (O(1)); any
structural change to the array fails validation and falls back to a
full scan.
* ui: Throttle per-chunk stream state writes to localStorage
saveStreamState ran JSON.stringify + a synchronous localStorage.setItem
on every decoded chunk of the stream. The read loop now goes through a
new saveStreamStateThrottled (one write per conversation per 500ms,
latest value held pending); the public saveStreamState keeps its
immediate-write contract for stream start and pre-fetch, and also
resets the throttle window.
A pending offset is force-flushed at resume boundaries (resumeStream
reads the offset back from localStorage), on visibilitychange->hidden
and on pagehide, so a reload always finds a usable offset. The resume
offset only needs to be roughly current since the server retransmits
from a line boundary and the client discards its partial line.
Adds unit tests for the throttled/flush/clear interplay.
* ui: Compute context gauge timing stats in one pass
currentRead/Fresh/Cache/Output were separate deriveds, each running a
full reverse scan of activeMessages for the last assistant timings,
and cumulative ran its own forward scan plus an agentic filter - 4-5
O(n) passes per chunk while streaming. Replace with a single
summarizeAssistantTimings() pass (last assistant timings, last
agentic llm totals and the cumulative sums) feeding a shared derived
snapshot. Semantics unchanged, including the live-stats overrides and
the agentic llm-totals branch.
* agentic : clear session state when a conversation is deleted
Every conversation that ran an agentic flow left an AgenticSession in the
store forever; clearSession was never called. conversationsStore now
notifies deletion listeners and agenticStore drops the matching sessions,
avoiding a circular import back into conversationsStore.
* chat : extract ChatService.normalizeMessagesForApi
The DB->API message normalization (convert + drop empty system messages)
was duplicated in sendMessage, preEncode and the agentic flow. Extract it
into one shared method and call it from all three.
* sse : share record splitting and data extraction
splitSseRecords and extractSseDataPayload centralize the record-boundary
splitting and data: line extraction used by parseSseJsonStream and the
models status feed. chat.service keeps its own line-based parser for
resume support.
* api : delegate apiFetchWithParams to apiFetch
apiFetchWithParams duplicated apiFetch's headers/fetch/error handling
body-for-body; it only differs in URL construction. Build the URL and
delegate.
* chat flows : dedupe title, timings and cleanup handling
- conversationsStore.applyTitleFromContent centralizes the title-from-first-
message logic duplicated in 5 places
- ChatProcessingStore.applyStreamTimings centralizes the onTimings handler
shared by the chat and continue flows
- host.cleanupStreaming centralizes the loading/streaming/processing reset
repeated across the continue flow's exit paths
* conversations : centralize conversation update mirroring
rename, pin, mcp override, reasoning effort and cwd all repeated the same
write-DB-then-mirror-into-list-and-active dance. A single
applyConversationUpdate(id, updates) on the host collapses all five and
removes the forgot-to-mirror-one-field bug class. Drops the redundant
array reassignment in setCwd (deep field assignment is reactive).
* mcp : dedupe tool execution, server parsing and tool indexing
- executeTool delegates to executeToolByName (only diff was argument parsing)
- drop the private #parseServerSettings copy; use parseMcpServerSettings
- cache getServers() keyed on the raw config value (hot path)
- indexServerTools() unifies the three identical toolsIndex rebuild loops
Assisted-by: Claude
* mcp : share cursor pagination and tool indexing
- MCPService.paginate() collapses the identical do-while loops in
listAllResources and listAllResourceTemplates
- promoteHealthCheckToConnection now uses indexServerTools like the other
connect paths
Assisted-by: Claude
* database : share message parent-child bookkeeping
- addChildToParent() dedups the append-to-children update in createMessageBranch
and createSystemMessage
- removeChildFromParent() dedups the remove-from-children cleanup in deleteMessage
and deleteMessageCascading
- bulkAdd the cloned messages when forking a conversation instead of one add
per message
Assisted-by: Claude
* chore: Lint/format
* fix: `pagehide` event from `window`
* refactor: Api Fetch util
* docs : rewrite architecture sections in README
Update the high-level diagram, routes, hooks, stores, services and data
flow tables to match the current UI structure (mcp/settings/search
routes, agentic/tools/mcp stores, MCPService/ToolsService/SandboxService,
/tools API). Fix stale architectural patterns for per-conversation state
and modality validation.
* chore : add ESLint rule for blank lines between accessors
Enforce a blank line between consecutive class accessors. The core
padding-line-between-statements rule does not cover class members, so a
local rule is needed.
* refactor : reorder store members and unify naming
Order store class members as public fields, private fields, constructor,
getters, public methods, then private methods. Normalize private naming
to the `private` keyword (drop `#` and the `_` prefix where there is no
matching public getter). Rename conversationsStore.init() to
initialize() to match the other stores.
* refactor : prefix lookup methods with get in agentic and chat stores
Unify bare-name lookup methods with the get* prefix used across the
other stores (mcp, models, tools, settings). Renames currentTurn,
totalToolCalls, lastError, streamingToolCall, executingToolCallId,
pendingPermissionRequest, pendingContinueRequest,
pendingSteeringMessageContent, pendingSteeringMessageExtras in the
agentic store and pendingMessageContent, pendingMessageExtras in the
chat store. Updates the two consuming components and a doc comment.
* refactor: Clean up comments in stores' and services' code
* chore : add ESLint rule for class member ordering
Enforce structural order (public fields -> private fields -> constructor ->
getters -> setters -> public methods -> private methods) with alphabetical
sorting within each group via perfectionist/sort-classes. Dependency
detection keeps Svelte $derived fields in a valid dependency order instead
of alphabetizing them, since Svelte rejects forward references.
Assisted-by: Claude
* refactor : reorder class members to match new ESLint rule
Apply the sort-classes rule across stores, services, hooks and utils.
Pure reordering - verified no logic changes by comparing sorted line
multisets before/after. All tests and svelte-check pass.
* feat: add --mmproj-device arg & backwards compatible MTMD_BACKEND_DEVICE env var
* feat: load mmproj device backend immediately, add -mmdev shortflag
* fix: its a pointer now get the name
* clean up
* gen docs
* nits
---------
Co-authored-by: Xuan Son Nguyen <son@huggingface.co>
* CI: Use LLVM's OpenMP over MSFT_DEBUG_non_redist on Windows
Currently, we ship the non-redist debug version of microsoft's libomp.
This PR changes this to official LLVM's release, also packaging
the license as needed.
* Remove LLVM SHA from job name to increase legibility
* Add temp validations to CI
* Revert "Add temp validations to CI"
This reverts commit eef97c88b5.
* Build OpenMP in CI
* Make OpenMP fetch self-contained in cmake and cache in CI
* Robustify Licens-packaging
1. Ship OpenMP license, not LLVM's.
2. Invalidate cache also on checksum of the license
* Remove stale reference in docs/build.md
* No longer package base license in release
This was scope-creep
* Add explanatory comment to OpenMP license
* Remove arm64 smoke
Forgot this during conflict resolution during rebase of
c54c0e9cf6
* Remove GGML_OPENMP_FETCH_CACHE_DIR as requested by @CISC
* whitespace changes
* CUDA: runtime GGML_CUDA_MMVQ_MAX to tune the mvq->MMQ decode crossover
Add a runtime override of the mul_mat_vec_q -> MMQ batch crossover
(default MMVQ_MAX_BATCH_SIZE). Lowering it routes batches above the
threshold from the CUDA-core vector kernel to the int8 MMQ tensor-core
path, which is faster once quantized decode becomes compute-bound at
B>1 (measured +23-41% at B=8 on RTX 5090 for Q4_K dense, no low-batch loss).
The value is parsed once and clamped to [1, MMVQ_MAX_BATCH_SIZE], since
mul_mat_vec_q asserts ncols_dst <= that; invalid input warns and falls
back to the default. The override is applied consistently in both the
mul_mat_vec_q and MUL_MAT_ID dispatch paths. Default behavior unchanged.
* Added Blackwell specific switch point, to reduce dependence on runtime env var.
* Add per-HW switch point values for DGX Spark and removing runtime env var
* Adding switch points for Ada, tested on RTX 4090
* Modifying DGX Spark numbers based on latest run and adding some comments and small functional changes relating to MoE
* Reverting an unnecessary conditional
* Update ggml/src/ggml-cuda/mmvq.cu
---------
Co-authored-by: praneshgo <227579474+praneshgo@users.noreply.github.com>
Co-authored-by: Oliver Simons <osimons@nvidia.com>
* metal: dequantize q8_0 KV to f16 before flash attention
Add a preprocessing pass for GGML_OP_FLASH_ATTN_EXT on the Metal backend:
when the KV cache is quantized (Q8_0 for now), dequantize K and V into a
contiguous F16 scratch buffer and run the existing F16 flash attention
kernels on it, instead of the in-kernel dequantization path.
- new kernel kernel_flash_attn_ext_dequant_to_f16<block_t, QK, deq_t4x4>:
one thread per quant block (K then V), stride-aware so permuted KV is
supported; instantiated for Q8_0 (extending to Q4_0/Q4_1/Q5_0/Q5_1 is
one instantiation + one gate case)
- the gate is type-only: dequantize whenever the KV is quantized,
regardless of head sizes, GQA ratio or n_kv; the attention kernels
themselves are untouched
- the F16 copies live in the op's own scratch allocation
(ggml_metal_op_flash_attn_ext_extra_dequant_f16); the KV pad kernel
reads the dequantized buffers when the path is active
- the FA pipeline getters gain a use_f16_kv flag selecting the existing
f16 kernels and contiguous strides
- ref: https://github.com/ggml-org/llama.cpp/pull/25556
Verification (M2 Ultra):
- test-backend-ops test -o FLASH_ATTN_EXT: 4798/4798 pass, including the
new q8_0 eval cases (decode/prompt, permuted, sinks+ALiBi+softcap,
kv=113 pad path, kv=16384)
- llama-perplexity on Qwen2.5-0.5B with -ctk q8_0 -ctv q8_0 matches the
f16 KV reference (PPL 1.0008 vs 1.0008)
Assisted-by: pi:llama.cpp/Qwen3.8-27B
* metal : launch the FA KV dequant kernel separately for K and V
Simplify kernel_flash_attn_ext_dequant_to_f16: it now dequantizes a single
tensor (its own ne/nb and dst) with no is_v branching, and the op dispatches
it twice with the same pipeline - once for K and once for V. The kargs
struct shrinks to a single ne/nb set plus nblocks.
Assisted-by: pi:llama.cpp/Qwen3.8-27B
* metal : dequantize q4_0, q4_1, q5_0 and q5_1 KV to f16 before flash attention
The dequant pass now covers all quantized KV types supported by the Metal
flash attention kernels. The dequant kernel, kargs, scratch allocation and
dispatch are type-generic, so each type is one kernel instantiation plus one
gate case.
Assisted-by: pi:llama.cpp/Qwen3.8-27B
* metal : skip the redundant V dequant when V is a view of K
In MLA-based models, the V of the FA op is a view of K (the first ne20
elements of each K row); the dequantized V is then a view of the dequantized
K, so skip the second dequant dispatch, do not reserve the V scratch region,
and let the pad and attention kernels read V from the K F16 buffer with K's
strides. The detection follows the CUDA backend:
V->view_src && (V->view_src == K || (V->view_src == K->view_src && V->view_offs == K->view_offs))
Also fix the FA pipeline getters: ns10/ns20 are function constants baked into
the kernels and must be the actual K/V row widths as seen by the kernel. The
dispatch now passes them explicitly (nb11_attn/nb10_attn, nb21_attn/nb20_attn)
instead of the getters assuming contiguous F16 KV (ns20 = dv), which was wrong
when V is read from K with K's row pitch (e.g. 576 vs 512).
New test cases: 576/512 q8_0 (MLA shape, V is a view of K) at kv=113 (KV pad),
nb=1 (vec) and nb=64 (non-vec).
Assisted-by: pi:llama.cpp/Qwen3.8-27B
* test : remove backend-specific wording from test-backend-ops comments
Assisted-by: pi:llama.cpp/Qwen3.8-27B
* pi : avoid backend mentions in test-backend-ops comments
Assisted-by: pi:llama.cpp/Qwen3.8-27B
* metal : rename the FA dequant_f16 identifiers to kv_f16
Assisted-by: pi:llama.cpp/Qwen3.8-27B
* cont : clean-up
* cont : remove TODO
* ggml: fix backend split scheduler race condition
splits without input were running concurrently with other splits, while potentially reusing memory the other split is accessing
* only sync when split has no inputs
* convert: fix get block count error for Nemotron
Signed-off-by: Rock Chen <rockchen.tw@gmail.com>
* fix this in NemotronHModel.__init__ instead.
This reverts commit ca689cbc87.
---------
Signed-off-by: Rock Chen <rockchen.tw@gmail.com>
* provide static workspace for cuBLAS handles
* account for concurrent streams when using GGML_CUDA_GRAPH_OPT
* drop cublas_handle overloads and remove direct cublasSetStream calls
* Update ggml/src/ggml-cuda/common.cuh
---------
Co-authored-by: Oliver Simons <osimons@nvidia.com>
build_attn with the llm_graph_input_attn_k_iswa input was using the cached K
tensor itself as V. Create V as a view of K (the first v_cur->ne[0] elements
of each row), like the other K-only build_attn overloads.
The deepseek4 MTP call site now passes the kv tensor as v_cur.
Assisted-by: pi:llama.cpp/Qwen3.8-27B
* backend: propagate buffer usage in meta backend
* ggml-meta: make sure to call init_tensor for all new tensors
* meta: remove explicit check for meta backend in ggml_backend_meta_get_split_state
I can't seem to reproduce the original failure in the latest code.
* hexagon: fix FA HMX queue ordering in the pipelined path
* hexagon: double buffer D matrix, store diagonal tile only
* format code
* align the indentation
* opencl: port fused ssm_scan kernel (Mamba-2, d_state in {128, 256})
Fold the fused per-token SSM_SCAN recurrent step from opencl/gdn-qwen36-35b
onto the unified base. Previously SSM_SCAN fell back to CPU here; now scalar-A
Mamba-2 with d_state in {128,256}, all-f32, runs on GPU. Other shapes (incl.
Mamba-1 element-wise A) still fall back. test-backend-ops -o SSM_SCAN passes on
Adreno X2-90. opt-out via GGML_OPENCL_DISABLE_SSM_SCAN=1.
* opencl: cleanup
* opencl: require K == 1
---------
Co-authored-by: Li He <lih@qti.qualcomm.com>
* ggml-cpu: gate __fp16 on __ARM_FP16_FORMAT_IEEE
__ARM_NEON only signals NEON availability. The __fp16 type also needs
the IEEE half format, implied on AArch64 but selected with
-mfp16-format=ieee on 32 bit Arm, where the compiler otherwise rejects
the type.
The guard keeps every toolchain that provides the type on the same code
and sends that one configuration to the generic lookup path.
* ggml-cpu: gate the NEON+FMA block on __ARM_FP16_FORMAT_IEEE
Both halves of the F16 section dereference __fp16, so armv7 with
neon-vfpv4 hits the same unknown type error. Without the IEEE
format the configuration now falls back to the scalar path.
Address review from @JonathanC-ARM
* vulkan : dequant q8_0 KV once in coopmat1
Assisted-by: Claude (Opus 4.8)
* vulkan : fall back instead of aborting when FA scratch exceeds maxStorageBufferRange
* vulkan : require KV-cache layout in FA dequant path
Assisted-by: Claude (Opus 4.8)
* vulkan : skip FA dequant path on coopmat2
Assisted-by: Claude (Opus 4.8)
* tests : add contiguously-allocated quant K/V FA tests
Assisted-by: Claude (Opus 4.8)
* vulkan : trim comments
* vulkan : tighten permutation checks for FA path
* vulkan : set prealloc_x_need_sync after the FA dispatch
* vulkan : exclude Intel Xe1 from FA dequant path
* feat(convert): Add conversion for GraniteSWAForCausalLM
Branch: GraniteSWAForCausalLM
AI-usage: full (Bob, OpenCode + Qwen3.6-35b)
Signed-off-by: Gabe Goodhart <ghart@us.ibm.com>
* feat(llama): Add granite_swa support
Branch: GraniteSWAForCausalLM
AI-usage: full (Bob, OpenCode + Qwen3.6-35b)
Signed-off-by: Gabe Goodhart <ghart@us.ibm.com>
* feat(conversion): Add conversion infra for rope_pattern array
NOTE: There is other work also targeting this, so this may be
removed depending on merge order.
Branch: GraniteSWAForCausalLM
AI-usage: full (Bob)
Signed-off-by: Gabe Goodhart <ghart@us.ibm.com>
* fix(conversion): Fix SWA pattern logic and support for non-rope layers
Branch: GraniteSWAForCausalLM
AI-usage: full (Bob)
Signed-off-by: Gabe Goodhart <ghart@us.ibm.com>
* feat(conversion): Add support for GraniteMoeSWA
Branch: GraniteSWAForCausalLM
AI-usage: full (Bob)
Signed-off-by: Gabe Goodhart <ghart@us.ibm.com>
* feat: Add llama_hparams::has_rope and arch constants
NOTE: This shadows the work done for Granite Speech
https://github.com/ggml-org/llama.cpp/pull/25107
Branch: GraniteSWAForCausalLM
AI-usage: full (Bob)
Signed-off-by: Gabe Goodhart <ghart@us.ibm.com>
* feat: Add support for per-layer rope determination
Branch: GraniteSWAForCausalLM
AI-usage: full (Bob)
Signed-off-by: Gabe Goodhart <ghart@us.ibm.com>
* style: Fix failing flake8 for extra newlines
Branch: GraniteSWAForCausalLM
AI-usage: none
Signed-off-by: Gabe Goodhart <ghart@us.ibm.com>
* test: Write out SLIDING_WINDOW_PATTERN in llama-model-saver
Branch: GraniteSWAForCausalLM
AI-usage: full (OpenCode + Qwen3.6-35b)
Signed-off-by: Gabe Goodhart <ghart@us.ibm.com>
* fix(convert): Fix missing registration for GraniteMoeSWAForCausalLM
Branch: GraniteSWAForCausalLM
AI-usage: none
Signed-off-by: Gabe Goodhart <ghart@us.ibm.com>
* fix: Load MoE params as optional
Branch: GraniteSWAForCausalLM
AI-usage: draft (OpenCode + Qwen3.6-35b)
Signed-off-by: Gabe Goodhart <ghart@us.ibm.com>
* feat: Handle MoE params in conversion
branch: GraniteSWAForCausalLM
AI-usage: full (OpenCode + Qwen3.6-35b)
Signed-off-by: Gabe Goodhart <ghart@us.ibm.com>
* style: Remove unnecessary newline
AI-usage: none
Signed-off-by: Gabe Goodhart <ghart@us.ibm.com>
* fix: Remove unnecessary tensor additions to GRANITE architecture
Branch: GraniteSWAForCausalLM
AI-usage: none
Signed-off-by: Gabe Goodhart <ghart@us.ibm.com>
* fix: Correctly handle naming for ffn gate inp
Branch: GraniteSWAForCausalLM
AI-usage: none
Signed-off-by: Gabe Goodhart <ghart@us.ibm.com>
* fix: Always default hparams.rope_pattern to 1s
This isn't strictly necessary, but it will allow other models to rely on
hparams.has_rope(il) without needting to prepopulate.
Branch: GraniteSWAForCausalLM
AI-usage: none
Signed-off-by: Gabe Goodhart <ghart@us.ibm.com>
* feat: Move to has_rope for all granite model architectures
Now that we have a proper hparam for this, it's better to use it and not
require a hacky fallback in the hparam method itself.
Branch: GraniteSWAForCausalLM
AI-usage: none
Signed-off-by: Gabe Goodhart <ghart@us.ibm.com>
* feat: No hacky rope_finetuned fallback in has_rope
Branch: GraniteSWAForCausalLM
AI-usage: none
Signed-off-by: Gabe Goodhart <ghart@us.ibm.com>
* fix: Fully remove rope hparam filling in granitemoe
There are no granitemoe models that use NoPE (it's not actually used in the
layer building below), so this was just dead code.
Branch: GraniteSWAForCausalLM
AI-usage: none
Signed-off-by: Gabe Goodhart <ghart@us.ibm.com>
* fix: Save out rope_pattern in model-saver
Branch: GraniteSWAForCausalLM
AI-usage: none
Signed-off-by: Gabe Goodhart <ghart@us.ibm.com>
* fix: Set hparams.rope_finetuned for round trip
Since the value is _read_ from rope_finetuned, we need to persist it when
the model is saved with the saver.
Branch: GraniteSWAForCausalLM
AI-usage: none
Signed-off-by: Gabe Goodhart <ghart@us.ibm.com>
* fix: Code review cleanup
Signed-off-by: Gabe Goodhart <ghart@us.ibm.com>
Co-authored-by: Sigbjørn Skjæret <sigbjorn.skjaeret@huggingface.co>
Co-authored-by: Sigbjørn Skjæret <sigbjorn.skjaeret@huggingface.co>
* refactor: Keep gate/up fused for MoE path
Branch: GraniteSWAForCausalLM
AI-usage: full (Claude + Sonnet 5)
Signed-off-by: Gabe Goodhart <ghart@us.ibm.com>
* fix: Skip GRANITE_SWA in model saver
https://github.com/ggml-org/llama.cpp/pull/25505#discussion_r3773175651
Keeping is_swa_impl in the saver can break other models.
Branch: GraniteSWAForCausalLM
AI-usage: none
Signed-off-by: Gabe Goodhart <ghart@us.ibm.com>
* add sliding window pattern for model in test
* style: Fix indentation
Branch: GraniteSWAForCausalLM
AI-usage: none
Signed-off-by: Gabe Goodhart <ghart@us.ibm.com>
* fix: Fix \r\n
Thanks Claude!
Branch: GraniteSWAForCausalLM
AI-usage: none
Signed-off-by: Gabe Goodhart <ghart@us.ibm.com>
* feat: Keep shared expert fused
Branch: GraniteSWAForCausalLM
AI-usage: full (Claude + Sonnet 5)
Signed-off-by: Gabe Goodhart <ghart@us.ibm.com>
* style: More indentation fixes
Signed-off-by: Gabe Goodhart <ghart@us.ibm.com>
Co-authored-by: Sigbjørn Skjæret <sigbjorn.skjaeret@huggingface.co>
---------
Signed-off-by: Gabe Goodhart <ghart@us.ibm.com>
Co-authored-by: Sigbjørn Skjæret <sigbjorn.skjaeret@huggingface.co>
* add params
* cpu kernel
* metal kernel
* add test backend ops
* gate other backends
* ggml: (cuda) support ggml_rope_set_offset (#27121)
* rm cuda supports_op guard, fix webgpu clang-format
* ggml: support ggml_rope_set_offset on vulkan (#27344)
* ggml: support ggml_rope_set_offset on vulkan
* remove inplace optimization
The route loads run ahead of the root layout script, so validateApiKey
read the settings store while it still held factory defaults and probed
/props without the stored key. initStores() now hands the same startup
promise to every caller and the chat loads await it before probing.
The one-time admin baseline no longer overwrites a key the user has
already set: on a first visit the config carries factory values only, so
a diverging key comes from the user and wins.
* vulkan: tiled transpose for 0<->2 permuted CONT
-ggml_vk_get_cpy_pipeline only routed to the tiled shared-memory transpose
shader when dim1 was the innermost dimension, i.e. ggml_transpose (a 0<->1
swap). A 0<->2 swap -- ggml_cont(ggml_permute(x, 2, 1, 0, 3)) -- fell back to
the generic per-element strided copy, whose source reads stride by ne0*ne1
elements: one cache line per lane.
-DeepSeek-V4's lightning indexer performs exactly that permute on a
[n_kv, n_tokens, n_head] tensor. On Vulkan/RADV gfx1151 it ran at ~1-9 GB/s of
a ~200 GB/s part and accounted for 43% of total prefill time.
-Add copy_transpose_02.comp, mirroring copy_transpose.comp but tiling over dst
dims (0, 2) with dims 1 and 3 as the batch, so reads walk src dim2 and writes
walk dst dim0 -- both contiguous. The selection condition additionally requires
a non-contiguous source and a contiguous destination so it cannot take cases
the contiguous-copy shader already handles.
-test-backend-ops only exercised ggml_transpose for CONT, so the strided path
was untested. Add test_cont_permute covering (2,1,0,3), (1,2,0,3) and (0,2,1,3)
over f32/f16 at tile-aligned, tile-unaligned and large shapes. The large shapes
are in the eval set rather than only in perf because perf mode does not verify
results.
-Measured on gfx1151, ne=[n_kv,64,64,1], perm=(2,1,0,3), f32:
n_kv=1024: 9.08 -> 579.85 GB/s
n_kv=1280: 20.03 -> 153.71 GB/s
n_kv=2048: 7.11 -> 91.68 GB/s
n_kv=2304: 16.24 -> 86.49 GB/s
-The ~2.2x penalty previously seen at power-of-two n_kv (destination-stride
aliasing) is gone. End to end, DeepSeek-V4-Flash IQ3_XXS prefill on a 9k-token
prompt goes from 56.33 t/s to 103.74 t/s (+84%).
-Note: at n_tokens=512 a single slow-path dispatch takes ~273 ms and looping it
in perf mode can trip the GPU watchdog, so the perf cases use n_tokens=64.
* tests: fold test_cont_permute into test_cont, add L2-exceeding perf shapes
Review feedback: test_cont gains a permute parameter ({0,0,0,0} = none),
matching test_mul_mat's pattern, and the separate struct is gone. Perf
adds [n_kv, 512, 64, 1] variants (~0.5 GB per run) that exceed GPU L2,
since the 64-token shapes fit in cache on large parts and read above
memory bandwidth.
* tests: trim perf-case comment to the two-line summary
* vulkan: trim comments on the 0<->2 transpose path
Drop the shader file header, the read/write block comments and the
rationale prose in the CONT test cases. Keep the tile-shape and
bank-conflict notes and the permute parameter documentation.
---------
Co-authored-by: Kevin Hopper <no-reply@maestro.press>
* gguf-py : add size guards to GGUFReader
Guard kv_count, tensor_count, string length, and array length
against crafted values that cause unbounded allocation or hangs.
Assisted-by: opencode
* gguf : validate tensor data section fits within file
When no_alloc=true, gguf_init_from_reader accepted files where the
tensor data section (computed from header claims) exceeded the remaining
file size. This allowed crafted GGUF files to pass validation while
having insufficient data, leading to OOB reads when the loader later
mapped tensor data from the file.
Assisted-by: opencode
* gguf-py : move size limits into gguf_reader.py
Per review feedback, the limits are not part of gguf.h but are
arbitrary limits defined in gguf.cpp, so define them locally in
the reader instead of exporting them from constants.
Assisted-by: opencode
* remove gguf.ccp changes
---------
Co-authored-by: Sigbjørn Skjæret <sigbjorn.skjaeret@huggingface.co>
* ui: Move stream lookup and replay fetches into ChatService
chatStore called fetch() directly for /v1/streams/lookup and the
/v1/stream replay. These now live next to the other stream-session
methods in ChatService, so services stay the only API I/O layer.
* ui: Move /models/sse feed reader into ModelsService
ModelsService.watchModelEvents owns the byte stream, reconnect loop
and SSE record parsing; modelsStore keeps only event routing and
state.
* ui: Extract conversation import/export into ConversationTransferService
The JSONL session format, ZIP archiving and browser downloads are
pure I/O with no store state, so they move out of
conversationsStore. The store keeps the DB orchestration
(bulkExportConversations, downloadConversation,
importConversationsData) and delegates the format work.
* ui: Consolidate active model resolution into modelsStore.activeModelId
The same resolution chain was duplicated in useChatScreenActiveModel,
ChatForm, ChatFormActionModels and contextStatsStore, with slight
drift in the single-model fallback. The canonical getter now lives in
modelsStore, and the shared last-assistant-model lookup moved to
utils as getConversationModel.
* ui: Initialize stores explicitly via initStores()
Store constructors and module-level side effects ran migrations and
localStorage reads in import order. Migrations rename and rewrite
localStorage keys, so a settings load racing ahead of them could
clobber migrated values. initStores() is called once from the root
layout and runs migrations first, then the stores that read
localStorage, then the conversations DB load.
* refactor: Constants for stream query params
* ui: Remove dead code from stores
- persisted() helper was exported but never used
- messageUpdateCallback / registerMessageUpdateCallback were never wired up
- conversationsStore.initialize() alias, single caller moved to init()
* ui: Merge device, theme and viewport into a single deviceStore
All three are reactive browser-environment signals, now exposed as one
class store: deviceStore.isMobile, deviceStore.isIOSDevice / isIOSSafari
/ isWKWebView / isStandalone and deviceStore.systemTheme.isDark. The
systemTheme name disambiguates the OS preference from the user theme
preference in settingsStore. Drops the unused viewport export (only
isMobile was consumed).
* ui: Merge build info into version store
One VersionStore class with build (llama.cpp build number from
build.json) and frontend (PWA version from _app/version.json),
matching the class pattern of the other stores.
* ui: Colocate context gauge popup state with its components
The gauge popup state is local UI state shared only by the
ChatFormContextGauge subtree, so it lives next to its consumers
instead of the app-scope stores barrel.
Kernel is a port of `ggml-cuda/fwht.cu`
(us/run, median):
```
m x n x k GEMM FWHT speedup
64 x 1 x 64 10.20 2.93 3.48x
64 x 2048 x 64 10.75 2.71 3.97x
128 x 1 x 128 10.33 2.88 3.59x
128 x 32 x 128 9.20 2.77 3.33x
128 x 2048 x 128 16.46 2.76 5.95x
256 x 1 x 256 10.19 2.77 3.68x
256 x 2048 x 256 16.69 3.41 4.89x
512 x 2048 x 512 54.16 12.89 4.20x
```
The collapsed \p{S} class was missing '~', which split " ~" into
separate pre-tokens and prevented the Ġ~ BPE merge used by DeepSeek V4.
This caused re-tokenized prompts to diverge from sampled tokens and
broke KV cache reuse.
Assisted-by: Codex
* update to ov-2026.3, update device drivers
* ci: skip nemotron-h rollback test on OpenVINO
The OpenVINO backend does not support SSM_SCAN, so the Nemotron-H recurrent state rollback graph is split and cannot preserve the recurrent cache output shape. Keep the test enabled for other backends and retain the qwen35 OpenVINO rollback coverage.
---------
Co-authored-by: ravi9 <ravi.panchumarthy@intel.com>
* xcframework : fix build
* mtmd : remove unused include path
* vendor : use vendor::hash alias target in cmake
CMake reserves "::" in target names for imported/alias targets, so the real
target keeps the name vendor-hash and a vendor::hash ALIAS target is added.
Consumers (mtmd, llama-gguf-hash) now link against the namespaced alias.
Assisted-by: pi:llama.cpp/Qwen3.8-27B
* vendor : add cmake targets for all vendored libs with vendor:: aliases
Add INTERFACE targets for the header-only vendor libs (miniaudio, nlohmann,
sheredom, stb) and ALIAS targets named vendor::<lib> for all of them,
including cpp-httplib and hash. Each exposes the vendor/ root so includes
are namespaced, e.g. <nlohmann/json.hpp>.
Consolidate the per-lib add_subdirectory calls into a single
add_subdirectory(vendor), keeping the cpp-httplib gate on LLAMA_BUILD_COMMON.
Consumers (llama-common, mtmd) now link the aliases instead of relying on
raw vendor/ include paths.
hash: consumers now include via "hash/hash.h"; the vendor/hash dir is kept
as a PRIVATE include so the synced upstream sources compile unmodified.
Assisted-by: pi:llama.cpp/Qwen3.8-27B
* readme : use foo/bar names in acknowledgements
Assisted-by: pi:llama.cpp/Qwen3.8-27B
* ocd : fix valign
* ci : create pre-release with change log and nightly link in make-release
After pushing the tag, create a pre-release using
ggml-org/action-create-release. The release description is generated by
scripts/make-release-desc.sh: the change log between the current and
previous version (one line per commit), a link to the corresponding
nightly build when it exists, and a note that semantic versioning is
still work in progress.
Assisted-by: pi:llama.cpp/Qwen3.8-27B
* cmake : bump version to 0.1.2
Assisted-by: pi:llama.cpp/Qwen3.8-27B
* ci : find the nightly tag by commit in make-release-desc.sh
The nightly release is guaranteed by the release checks to point at HEAD,
so instead of reconstructing its name (commit count, branch, hash) just
pick the b* tag pointing at HEAD. This also drops the RELEASE_BRANCH env
var from the workflow.
Assisted-by: pi:llama.cpp/Qwen3.8-27B
* ci : resolve the release commit from the version tag in make-release-desc.sh
The change log and nightly lookup now use the commit the version tag
points at (HEAD when the tag does not exist), instead of always HEAD.
This makes the script usable locally for older versions, e.g.
./scripts/make-release-desc.sh v0.1.1. The tag is resolved to a SHA
first, since --points-at does not peel annotated tags.
Assisted-by: pi:llama.cpp/Qwen3.8-27B
* ci : normalize the version argument in make-release-desc.sh
Accept the version with or without the leading v (0.1.1 == v0.1.1) and
reject anything else, instead of silently treating a bare version as a
non-existent tag name.
Assisted-by: pi:llama.cpp/Qwen3.8-27B
* cont : clean-up
* CUDA: MMVQ nwarps=8 for bs=1 for dense models on DGX Spark
Signed-off-by: ynankani <ynankani@nvidia.com>
* skip moe experts and allow others based on k geometry (allow only small idle tail)
Signed-off-by: ynankani <ynankani@nvidia.com>
* rename MMVQ DGX Spark params to GB10 and fix MSVC constexpr lambda capture
Signed-off-by: ynankani <ynankani@nvidia.com>
---------
Signed-off-by: ynankani <ynankani@nvidia.com>
* doc: document MCP stdio servers and CORS defaults in the server README
The MCP arguments were listed but nothing explained what an MCP server
is or how to declare one. Cover the stdio transport, the config keys,
the tool naming, and add a POSIX shell echo server as a minimal
example.
Also document the CORS behavior: the default reflected origin, the
switch to localhost once tools are enabled, and the recommended setting
per deployment.
* doc: drop the inline MCP shell example from the server README
The example parsed JSON-RPC by hand and sat in a page people copy paste
from, into servers spawned with the privileges of llama-server. Point to
the specification instead.
Link the pull request that introduced the feature, and keep a short
mcp.json snippet so the table of configuration keys has a declaration to
refer to.
Add a "Create and push git tag" step to the release job, right before
the "Create release" step. The tag is created with git tag and pushed
with the deploy key already configured by the Clone step, instead of
relying on the Releases API (action-create-release) to create it as a
side effect.
The tag is lightweight, matching all existing b<number> release tags.
The step is idempotent: if the tag already exists (e.g. on a re-run),
creation and push are skipped.
Assisted-by: pi:llama.cpp/Qwen3.8-27B
* ui: move get_datetime tool to frontend
* clarify docs
* server: drop the now unused ctime include
strftime() and gmtime_r() were the only users, both went away with the
get_datetime tool. Also make the renderer's catch inert: the browser
executor always emits JSON, so a non-JSON result is no longer a date to
display.
---------
Co-authored-by: Pascal <admin@serveurperso.com>
* ci : parallelize platform builds in build-xcframework.sh
The ios-xcode release job builds 7 platform/simulator configurations
sequentially, each with -j $(nproc). Run them with at most 3 concurrent
builds (the release runner has 3 cores), splitting the cores between the
builds (-j 1 each on the runner), so total CPU pressure is unchanged while
the build phase runs about 2.3x faster.
- convert the 7 build blocks into functions (flags unchanged)
- add a run_builds_parallel pool: 3-slot sliding window, per-build logs,
dumps the failing log and aborts on error (background job failures do
not trigger set -e)
- queue the 2-arch builds first so the slower builds occupy the slots early
Assisted-by: pi:llama.cpp/Qwen3.8-27B
* pi : add guideline for comments
* cont : disable 5/7 builds
* ci : make build-xcframework.sh builds configurable via CLI args
The script now takes an optional list of builds to run
(ios-sim ios-device macos visionos visionos-sim tvos-sim tvos-device);
with no arguments it builds all of them, as before. The per-build
lists for the build pool, framework setup, static library combining
and xcframework creation are now driven by a single build_spec
lookup instead of four hardcoded (partially commented-out) lists.
release.yml builds only macos and ios-device to cut the build time.
Assisted-by: pi:llama.cpp/Qwen3.8-27B
* dspark: support speculators-format checkpoints (SpecForge exports)
Speculators-format DSpark drafts (e.g. SpecForge exports for the
Gemma-4-26B-A4B target) differ from the dense DeepSpec checkpoints in
three ways:
- the config nests the backbone hparams under transformer_layer_config
and gives the extract layers as aux_hidden_state_layer_ids
- the block is the DFlash 1+N fill-in layout: the anchor slot is a bonus
token, not a prediction slot. Written as dflash.bonus_anchor; such
drafts build the block and read the mask positions exactly like
DFlash (n_max drafts from a 1+n_max block), only the Markov/confidence
sampling comes from DSpark
- the draft output vocab may be reduced (draft_vocab_size < vocab_size)
with a d2t remap table. The converter expands lm_head/markov_w2 back
to the full vocab and synthesizes an lm_head bias of -1e9 on the rows
the draft cannot produce, so the runtime needs no d2t remapping. Such
drafts ship their own (now optional) token_embd/output tensors instead
of sharing the target's
Verified against gemma4-26b-a4b-dspark: greedy outputs are byte-identical
with and without the draft; acceptance 0.46, mean draft len 3.7 (n_max 6).
Co-authored-by: desovo7 <942845546@qq.com>
Assisted-by: Claude Fable 5
* dspark: fold the speculators draft class into DSparkModel
One class now covers every DSpark variant. What used to pick the class is
a single flag, because the arch name turns out to be the only thing that
separates the two families: SpecForge also exports a flat schema that
carries no speculators_* fields yet still uses the 1+N bonus-anchor block,
so keying on those fields would silently mis-read its drafts.
Also rename i0 to i_first_pred in the draft read loop and the Markov head,
and give the head a real bonus_anchor bool instead of testing i0 > 0.
Converting the Qwen3-8B DeepSpec draft and both gemma-4 speculators drafts
produces byte-identical GGUFs. The one behaviour change is that the
markov_head_type check now also covers the DeepSpec checkpoints, which
previously skipped it.
Co-authored-by: desovo7 <942845546@qq.com>
Assisted-by: Claude Opus 5
* dspark: address review comments
- rename bonus_anchor to sample_from_anchor (GGUF key and code), matching
the checkpoint config field; absent key still means anchor-first
- rework the reduced draft vocab to match EAGLE3: d2t is written as I64
absolute target ids and the logits are scattered at runtime, instead of
expanding lm_head/markov_w2 and synthesizing an output bias at conversion
- move the t2d skip to modify_tensors, like EAGLE3
- drop _is_specforge: the arch name only picks the sample_from_anchor
default, embed/lm_head sharing is decided by the draft vocab size
- deduplicate the tok_embd create_tensor left behind by the rebase
Verified with the RedHat gemma-4-31b speculator draft: greedy output is
byte-identical with and without the draft; acceptance 0.26 (n_max 7).
Co-authored-by: desovo7 <942845546@qq.com>
Assisted-by: Claude Fable 5
* dspark: fold the sample_from_anchor read into the block_size block
* dspark: fix flake8 continuation indent
* clean up
* dspark: key the sample_from_anchor default off the export format
Co-authored-by: desovo7 <942845546@qq.com>
Assisted-by: Claude Fable
* dspark: drop t2d in filter_tensors
Co-authored-by: desovo7 <942845546@qq.com>
Assisted-by: Claude Fable
* dspark: map model.lm_head instead of bypassing the dflash prefix
Co-authored-by: desovo7 <942845546@qq.com>
Assisted-by: Claude Fable
---------
Co-authored-by: desovo7 <942845546@qq.com>
Co-authored-by: ruixiang63 <wangruixiang07@outlook.com>
* ci : allow make-release to target a specific commit
The make-release workflow now accepts an optional 'commit' input. When
set, that commit is checked out and the release checks verify that it
belongs to the branch selected in the "Run workflow" dialog and is not
older than 3 days from the branch tip. The check is part of
make-release-checks.sh (driven by the RELEASE_BRANCH env), so it follows
the same dry-run semantics as the other checks.
Assisted-by: pi:llama.cpp/Qwen3.8-27B
* cont : scan latest 100 relase workflow runs
* cont : do not check manually for release.yml success
* Adding support for bailingmoe3
* Adds speculative decoding support
* Make BailingMoE3 safe gate metadata optional
* bailingmoe3: apply trained SwiGLU clamps
* common: fix Bailing V3 tool argument parsing
* llama-model-saver, instantiate float vector metadata writer
* bailingmoe3: support Q-LoRA (Ling-3.0-tiny)
Ling-3.0-flash sets q_lora_rank: None and projects Q directly, so the current
implementation loads a single ATTN_Q tensor. Ling-3.0-tiny sets q_lora_rank: 256
and routes Q through a LoRA bottleneck instead:
q_a_proj -> q_a_layernorm -> q_b_proj
Conversion therefore failed with:
ValueError: Can not map tensor 'model.layers.3.attention.q_a_layernorm.weight'
Add the missing path, mirroring the existing deepseek2 MLA implementation:
* constants.py - add ATTN_Q_A / ATTN_Q_B / ATTN_Q_A_NORM to BAILINGMOE3
* tensor_mapping.py - map model.layers.{bid}.attention.q_{a,b}_proj and
q_a_layernorm
* conversion - emit attention.q_lora_rank when the config has it
* bailingmoe3.cpp - read n_lora_q; create the Q-LoRA tensors and build Q
through the bottleneck when q_lora_rank > 0
Everything is gated on q_lora_rank > 0. Ling-3.0-flash's config has no
q_lora_rank, the converter only emits the key when present, hparams.n_lora_q
defaults to 0, and get_key(..., required=false) leaves the target untouched when
the key is absent - so flash keeps taking the existing direct-Q branch.
The LoRA path produces the same shape as the direct projection, so the
nope/rope split, RoPE application and wk_b absorption downstream are unchanged.
* small mtp change
* bailingmoe3: support separate MTP GGUF and Q-LoRA MTP
* gguf: remove duplicate add_kda_gate_lower_bound definition
---------
Co-authored-by: bloomer <bloomer@booper.brushtail.me>
Co-authored-by: Dyluhn <dylanranejohnston1@gmail.com>
The log output does not append a newline, so the warning ran into the
next line printed on stdout, corrupting the benchmark table header.
Signed-off-by: Fathi Boudra <fathi.boudra@linaro.org>
Adjusts the thread/block count to be proportional to the size
of the quant, reducing under/over subscription.
Largest perf improvement is the q4_0 -> f32 path, with, on
a Arc 70, throughput goes from 20.21 GB/s to 158.19 GB/s
The rest of the quants are flat in performance uplift.
* ui: mask API Key field in settings and error splash to stop browser autofill
* ui: set autocomplete=new-password on private fields
The password input type makes browsers offer to save the API key
in the password manager and autofill saved site credentials into
the field. The new-password autocomplete value disables both.
---------
Co-authored-by: Pascal <admin@serveurperso.com>
* model: add Kimi-K3 text model
Hybrid KDA (linear) + MLA (full) attention as in Kimi-Linear-48B, plus five
things that architecture does not have:
1. cross-layer residual attention (attn_res_block_size)
2. latent MoE (routed experts run at n_expert_latent)
3. situ activation (replaces SwiGLU everywhere)
4. MLA output gate (sigmoid gate before o_proj)
5. full-rank KDA gate (single ssm_g instead of ssm_g_a/ssm_g_b)
K3's text_config reports KimiLinearForCausalLM - the older 48B architecture -
so get_model_architecture routes on the top-level name instead.
The KDA decay gate has two forms, selected by linear_attn_config's
gate_lower_bound. It is not a clamp: when set it swaps the activation entirely
(fla/ops/kda/gate.py), from -exp(A_log)*softplus(x) to
lower_bound*sigmoid(exp(A_log)*x). K3 sets it to -5.0; kimi-linear leaves it
unset, so that path is unchanged.
Cross-layer residuals reuse ggml_dsv4_hc_pre for the weighted sum. That op is
CPU + CUDA only, so Metal/Vulkan will fall back per-node until those kernels
exist.
The routed experts ship as compressed-tensors "mxfp4-pack-quantized". That is
bit-compatible with ggml's MXFP4 - same E2M1 code assignment, same E8M0 scale
byte, only the nibble positions within a block differ - so they are repacked
rather than dequantized, losslessly and without a ~5.5 TB bf16 round-trip.
The repack is built lazily because gguf_writer holds every added tensor until
the final write. DeepSeek-V4 was already doing the identical bit-shuffling, so
it now shares the helper.
Verified against Moonshot's own code path (transformers + fla's Triton KDA
kernels) on a tiny model exercising every K3-specific feature. Final-position
logits vs the fp32 reference: 6.7e-05 rel / corr 1.00000000 for both the
chunked and the recurrent delta-net path. MXFP4 blocks dequantize to the source
weights with 0.0e+00 error.
Assisted-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
* model: fix ty errors in the Kimi-K3 converter
- `_res_parts` buffers (kind, tensor) pairs, not bare tensors
- `get_tensors` must return an Iterator, matching ModelBase
- LazyBase's `func` takes one argument, so pass the expert loaders through
`args` instead of the closure
- borrowing KimiLinearModel.set_vocab from an unrelated TextModel is
deliberate and safe, but not expressible in the signature
No behaviour change: the MXFP4 repack still dequantizes to the source weights
with 0.0e+00 error and end-to-end logits are unchanged (8.386e-03 rel,
corr 0.99996630).
Assisted-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
* Update conversion/kimi_k3.py
Co-authored-by: Boris Dvorkin <b_dvorkin@niuitmo.ru>
* Increase LLAMA_MAX_EXPERTS from 512 to 1024
* tests : support for Kimi K3 in archs test
* chat : add Kimi K3 chat format (reasoning, content, typed tool calls)
K3's assistant output is an XTML-ish tagged format built by the template's
open_tag/close_tag macros. Two properties break generic parsing:
1. The generation prompt ends with open_tag('think'), so the completion
starts inside the think section with no opening marker in the output
(thinking_forced_open).
2. Only <|open|>/<|close|>/<|sep|>/<|end_of_msg|> are special tokens; tag
names ("think", "response", "message") are ordinary text tokens.
Adds common_chat_params_init_kimi_k3 (PEG_NATIVE) with detection on the
marker trio, reasoning extraction, response unwrapping, and tool-call
parsing of the tools/call/argument tag structure with argument types
taken from the tool schema. Includes the K3 chat template fixture and 9
test-chat cases derived from real generations of the full 2.8T model.
Verified end-to-end against Kimi-K3-Q2_K (GrEarl/Kimi-K3-GGUF) on 8x B200:
content, reasoning_content, streaming deltas, and tool_calls all correct;
finish_reason stop/tool_calls as appropriate.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
* chat : add message_delimiters for Kimi K3
Per-role message-start markers for token-level span splitting. User and
assistant messages carry only the role attribute, so their full opener
(through <|sep|>) is used; system and tool messages continue with more
attributes (type=/tool=/index=), so those delimiters stop after the
role's closing quote. Verified against the K3 tiktoken vocabulary that
the closing quote is always a standalone token across all attribute
variants, so the token-level prefix match stays exact.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
* fix: apply nits from @ngxson and text fixes from @danielhanchen
* tests : added missing hyperparameters and tensors for Kimi K3 in test-llama-archs
* chore : move overly verbose header file comments to Kimi K3 source file
* tests : re-enabled KIMI_K3 in test-llama-archs for WebGPU backend
* model-saver : emit kda_gate_lower_bound for Kimi K3
Quick fix. The Kimi K3 loader reads kda_gate_lower_bound and gates a graph branch on it (it scales the KDA gate when the bound is above -INFINITY), but the model
saver never wrote the key, so a save->load roundtrip silently dropped it back to the -INFINITY default and changed the model's output. The real K3 config sets gate_lower_bound = -5.0.
I propose to emit it from the saver, and set it to -5.0 in the test-llama-archs K3 case so the roundtrip check exercises it (the roundtrip fails without the saver line).
* Refactor conditional for model architecture check
* tests : re-enabled (again) KIMI_K3 and MINIMAX_M3 in test-llama-archs for WebGPU backend
* fix code comments
* add template on conversion
* move repack_mxfp4_blocks to model base
* nits
* add_value_length
* optimize res_stack construction
* nits
---------
Co-authored-by: Boris Dvorkin <b_dvorkin@niuitmo.ru>
Co-authored-by: Stanisław Szymczyk <sszymczy@gmail.com>
Co-authored-by: Deepankar Singh <singh.deepankar39@gmail.com>
Co-authored-by: Claude Fable 5 <noreply@anthropic.com>
Co-authored-by: Caleb DeLeeuw <caleb.deleeuw@gmail.com>
Co-authored-by: Xuan Son Nguyen <son@huggingface.co>
* common: support --models-dir loading MTP assistant models
* common: preset: check for MTP models with strict prefix
* common: preset: Take advantage of PR #27005
* handle other draft types
* drop eagle3
* clean up
---------
Co-authored-by: Xuan Son Nguyen <son@huggingface.co>
Replace the deprecated --mmap, --no-mmap, --mlock, and --direct-io flags with
the unified --load-mode argument across scripts, examples, and documentation.
Internal warning message and env var docs updated accordingly.
Signed-off-by: Fathi Boudra <fathi.boudra@linaro.org>
* llama : support for MiniMax-Text-01 model
* chore : renames to match the other MiniMax models
* model : add logits mask as MiniMax-Text-01 embeddings tensor has zero-valued embeddings for tokens >= 200032 that produce zero logits disrupting the token sampling process
* llama : replace hardcoded conditions with hparams.is_recr()
* model : used build_rs() for recurrent state management
* chore : code cleanup
* model : optimized MiniMax-Text-01 by removing the state tranpose operations
* chore : removed unnecessary ggml_cont() in MiniMax-Text-01 implementation
* llama : add generic logits mask graph input
* model : permuted diag_decay dimensions to avoid doing it inside MiniMax-Text-01 graph
* chore : code cleanup
* chore : code cleanup
* model : use token positions when calculating MiniMax-Text-01 decay tensors
* convert : add support for MiniMaxM1ForCausalLM as it seems to be the same as MiniMaxText01ForCausalLM
* chat : add jinja template for MiniMax-M1
Co-authored-by: QscQ <qscqesze@gmail.com>
* chore : code cleanup
* tests : MINIMAX_01-related fixes
* chore : silence Python lint errors
* vocab : remove unnecessary vocab type
* convert : update MiniMaxText01Model conversion to use yield when modifying tensors
* convert : suppress tokens with zero-valued embeddings during MiniMax-Text-01 conversion
* llama : removed logits mask - no longer necessary as token suppression is used instead
* model : use common functions to make MiniMax-Text-01 implementation more concise
Co-authored-by: Sigbjørn Skjæret <sigbjorn.skjaeret@huggingface.co>
* model : use common functions to make MiniMax-Text-01 implementation more concise
Co-authored-by: Sigbjørn Skjæret <sigbjorn.skjaeret@huggingface.co>
* convert : override non-working built-in chat template during conversion
* tests : skip arch MINIMAX_01 tests for WebGPU backend (it breaks again)
---------
Co-authored-by: Stanisław Szymczyk <sszymczy@gmail.com>
Co-authored-by: QscQ <qscqesze@gmail.com>
Co-authored-by: Sigbjørn Skjæret <sigbjorn.skjaeret@huggingface.co>
* chat: add reasoning_effort to common_chat_templates_inputs
Store OpenAI Chat Completions reasoning_effort and make it
available to jinja templates (with model specific translations
where required).
Assisted-by: llama.cpp:Muse-Glimmer-30B
* server : fixup reading reasoning effort from body
server_chat_convert_responses_to_chatcmpl already handles conversion of
Responses API reasoning.effort to reasoning_effort
* chat : expose reasoning effort
Assisted-by: Claude Opus 5
* chat : add reasoning_effort to generation_params
Assisted-by: Claude Opus 5
* chat : move reasoning_effort next to enable_thinking
Assisted-by: Claude Opus 5
* cont : mirror preserve_reasoning
* cont : pass context through analyze function
---------
Co-authored-by: Alde Rojas <hello@alde.dev>
* Initial changes for Recurrent state rollback for nemotron for cpu and cuda
* Removing CPU RS rollback. Will enable it in subsequent PRs
* addition of test case
* Removing assert and calling runtime API to check if op is supported
* removing extra API and updating the call sites for K
* replace static cuda detection to runtime fused_op api
* address review comments and fallback when SSM rollback not supprted
* Adding changes for supporting RS-rollback in CPU. Also added test-backend-ops for cpu and cuda
* removing memory manipulation as rs rollback is now supported in CPU
* removing the static probe which is not needed now
* correcting the format
* address review comments
* enabling test for all the backends, unsupported backends will fallback to CPU
* Apply suggestions from code review
Co-authored-by: Georgi Gerganov <ggerganov@gmail.com>
* choose different graph based on the result of fused_ssm_op is supported or not and also handled memory->n_rs_seq >1 case incase of op is not supported
* Support K > 1 in ssm_scan for all backends
* Fix CI Issues
---------
Co-authored-by: Georgi Gerganov <ggerganov@gmail.com>
Co-authored-by: Gaurav Garg <gaugarg@nvidia.com>
Scrub developer-specific /home/<user>/ paths from example docs and test
fixtures so they don't leak into the tree.
- examples/test-cmake/README.md: /home/danbev/... -> /path/to/llama.cpp/...
- tests/test-chat.cpp: /home/jarvis/... -> /home/user/... (input and
expected string kept identical so the parser test still passes)
Co-authored-by: Jim Wu <ywu@xilinx.com>
This commit tries to make the logging of target_layer_ids a bit clearer
and easier to read.
Currently the output generated looks like this:
```console
0.00.468.624 D load_arch_hparams: DFlash extract_layers = [0.00.468.626 D 2, 0.00.468.626 D 6, 0.00.468.626 D 20,
0.00.468.626 D 30, 0.00.468.627 D 42, 0.00.468.627 D 520.00.468.627 D ]
```
With the changes in the commit the output will be:
```console
0.00.522.765 D load_arch_hparams: DFlash extract_layers = [2, 6, 20, 30, 42, 52]
```
* OpenVINO backend: 1) enable gpt-oss moe on OV bk; 2) enable mxfp4 support
* OpenVINO backend: disable TOPK_MOE op test
* OpenVINO Backend: Add op FILL support
* OpenVINO backend: enable set rows with multi dims
* fix the name missmatch in setrow + view
* OpenVINO backend: enable op GGML_UNARY_OP_SIGMOID
* OpenVINO Backend: enable SQR & SQRT
* OpenVINO backend: 1) ensure unique node names for OpenVINO; 2) add org_src to recorde the src ggml tensor for OpenVINO dynamic shape infer
* OpenVINO backend: enable fallback for openVINO to CPU backend
* OpenVINO backend: fix accurace issue in gemma3n arch test
* fix mpt failed case
* OpenVINO backend: clean nodeinfo
* OpenVINO Backend: enable zero-size copy for view
* add concat ssm_conv in compute_dynamic_dim
enable qwen35
Fix after rebase
remove logging
* OpenVINO backend: disable EXP with FP32, which failed in op test. Root reason: the backend test initializes unary op inputs over a wide range, [-150, 150]. For FP32, exp(x) overflows around x ~= 88.7, so this test can randomly generate values right in or beyond the overflow region
* OpenVINO backend: fix CPY op test failed issue
* OpenVINO backend: fix GATED_DELTA_NET op test failed issue
* handle in-place op, handle qwen35 dynamic clearing of cache in cgraph
* handle qwen35 dynamic clearing of cache correctly
* Enable qwen35 dense multi seq
* Fix qwen35 9b gqa
* Fix after rebase
* Disable SOLVE_TRI
* openvino: fix NEOX RoPE accuracy on GPU stateful (mixed-rank Multiply)
In stateful mode the NEOX RoPE branch fed rank-3 data ([S, n_heads,
head_size]) into the Multiply against the rank-4 cos/sin tables
([1, S, 1, n_dims/2]). That mixed-rank broadcast is miscomputed by the
OpenVINO GPU plugin, corrupting the rotated Q/K and producing garbage
output (e.g. Phi-3-mini). Lift the data to rank-4 before the split/
Multiply so the operands are equal-rank, matching what the TYPE_NORMAL
branch already does. CPU and stateless paths are unaffected.
Phi-3-mini-Q4_K_M, wiki.test perplexity, GPU stateful:
before: PPL = 27120.43
after: PPL = 6.2263 (CPU reference: 6.2251)
* OpenVINO backend: 1) remove the unique name in llama.cpp; 2) add new ov name in ov bk; 3) fix issue in arch test & op test with latest code update
* OpenVINO Backenb: remove changes in llama.cpp
* Doc change (use x64 Native Tools Command Prompt for VS)
* Cleaner sentence
Co-authored-by: Copilot Autofix powered by AI <175728472+Copilot@users.noreply.github.com>
* OpenVINO Backend: cache key upgrade includes all src name
* OpenVINO Backend: enable llama arch test on ci
* OpenVINO Backend: move parameter node creating from decoder into translate
* OpenVINO Backend: create extra input ov node move from decoder to translate
* fix for op regression due to is_model_splitted
* openvino: fix CPY writeback for recurrent state rollback
Detect the rollback conv/gdn state writeback CPY nodes structurally
instead of by tensor name, since the rollback path in
build_conv_state does not call cb() and left the nodes unnamed. Add
per-node runtime offsets (rs_slot_begin_*, rs_src_begin_*) so the
cached IR handles any kv head, sequence count and snapshot slot for
both the conv state and the GDN state writeback.
Assisted-by: GitHub Copilot
* qwen35 moe
* optimize MoE expert aggregation with ReduceSum
* Skip GET_ROWS inaccurate test
* openvino: fallback dynamic MUL_MAT_ID shapes
* OpenVINO Backend: fix error in arch test model mpt
* fix error caused by cpy in arch test model kimi-linear
* OpenVINO Backend: fix error in arch test model minimax-m3
* openvino: fix GPU mul_mat_id op tests
* ggml-openvino: add GGML_OPENVINO_RELEASE_WEIGHTS to reclaim host weight RSS on GPU
The OpenVINO weight Constants are zero-copy views into host buffers
allocated by the backend (ggml_aligned_malloc, anonymous memory). On GPU
the plugin holds its own device copy after compile_model, so these host
pages are dead weight for inference. For a 1B Q4_K_M model this leaves
~850 MB of host RSS resident that the GPU path never reads again.
Add an opt-in GGML_OPENVINO_RELEASE_WEIGHTS mode that madvise(MADV_DONTNEED)s
the registered host weight buffers once the model is compiled, dropping
their resident pages while keeping the mappings valid (ggml still owns the
lifetime; tensors still point in). Measured steady-state RSS drops from
~1555 MB to ~710 MB on Llama-3.2-1B-Q4_K_M (Arc iGPU) with unchanged
throughput and correct output.
The GPU backend uses a single dynamic-shape model for both prefill and
decode, so a graph is compiled once and reused; the only event that forces
a recompile is clear_caches() on backend teardown. The change therefore:
- releases on the first cache-hit (model compiled, plugin has its copy);
- pins the compiled-model cache across backend teardown so a later
context reuses it instead of recompiling against the dropped pages;
- fails loud (GGML_ABORT) on a cache-miss recompile or on a second model
load, both of which would otherwise read zeroed weights or silently
reuse the wrong compiled graph.
Scope/limitations (all fail loud, never silently wrong): GPU only (the CPU
plugin reads the host Constants at inference time), one model per process,
and stable graph shapes. This reduces steady-state RSS, not the transient
compile-time peak. All changes are confined to the OpenVINO backend.
* ggml-openvino: stream weight requantization to cut the compile-time RSS peak
requantize_to_buffers() dequantized the entire tensor to a temporary
std::vector<float> of n_elements before requantizing. For token_embd.weight
(128256 x 2048) that transient is ~1 GB (1B model) / ~2 GB (8B), and it is
the single largest contributor to the OpenVINO compile-time memory peak --
it also fires twice for token_embd (once at load, once at graph build,
because token_embd is loaded via a CPU/mmap buffer and not cached as an OV
weight extra).
Stream the dequant instead: process a fixed window of complete rows
(CHUNK_ROWS=256) into a small scratch buffer and quantize/convert each chunk
straight into the output buffers. The transient F32 footprint is now
CHUNK_ROWS*ne0 floats regardless of tensor size.
quantize_q8_0/q8_1 gain an optional block_offset arg (default 0) so a chunk
writes its weights/scales/zp at the correct block. Streaming is applied to
the Q8_0_C / Q8_1_C / F16 targets (the large requant cases); the u4 (Q4_0)
path keeps the whole-array call because it packs two weights per byte with
running zp ORs, and a fallback handles any future target whose block size
does not divide a row.
Measured peak RSS (cold compile, GPU): 1B 2868 -> 1809 MB (-1.06 GB);
8B 11618 -> 9608 MB (-2.0 GB). Output verified unchanged
("capital of France is Paris"); throughput unchanged. Unlike
GGML_OPENVINO_RELEASE_WEIGHTS this reduces the transient peak, not just
steady-state, and needs no env flag. All changes confined to the OpenVINO
backend.
* ggml-openvino: avoid redundant token_embd requantization at compile
token_embd.weight is referenced twice in the graph path: as the GET_ROWS
embedding (a CPU/mmap-buffer tensor) it was re-extracted/re-requantized on
every weight-node build, and is_model_splitted() built a full (naive) set of
weight nodes just to test name membership — each requant is a ~1-2 GB F32
dequant of the 262M-element embedding.
Two changes:
- Add collect_weight_names(): a name-only collector for topology checks.
is_model_splitted() now uses it instead of create_weight_nodes(cgraph,
true), so the splitted-check no longer triggers any weight extraction.
- Memoize weight nodes built from non-OpenVINO buffers in a process-lifetime
cache keyed by tensor->data. These tensors have no OV buffer context to own
a cached extra, so without this they were rebuilt on every (re)compile;
prefill and decode graphs now share one build (verified: 2nd graph hits the
cache instead of re-requantizing).
Peak RSS is unchanged (the streaming-requant commit already removed the F32
transient); this removes redundant compile-time work. Output verified
unchanged ("capital of France is Paris"). Confined to the OpenVINO backend.
* ggml-openvino: gate compile-memory optimizations behind GGML_OPENVINO_REDUCE_COMPILE_MEM
The streaming requantization and the non-OpenVINO-buffer weight-node cache
(plus the name-only is_model_splitted path that pairs with it) are now opt-in
via GGML_OPENVINO_REDUCE_COMPILE_MEM. When unset, requantize_to_buffers()
fully materializes the F32 buffer and weights are rebuilt per compile exactly
as before; when set, the streaming path and the cross-compile weight cache
are used.
Default off keeps behavior identical to upstream unless explicitly enabled.
Verified: flag off -> peak RSS 2800 MB (original), flag on -> 1810 MB; output
"capital of France is Paris" in both modes. (GGML_OPENVINO_RELEASE_WEIGHTS,
added earlier, remains a separate opt-in for the steady-state release.)
* ggml-openvino: add frontend model cache (GGML_OPENVINO_MODEL_CACHE_DIR)
The plugin-level ov::cache_dir caches the compiled blob keyed by the OV
model, but producing that model still runs the full frontend every time:
weight requantization (incl. the large token_embd F32 transient) and the
ggml->OV graph conversion. This adds an opt-in frontend cache keyed off a
fingerprint computed directly from the ggml cgraph, so a hit imports a
previously exported CompiledModel and skips requant + convert + compile
entirely.
Key (model-cache.{h,cpp}) = 64-bit FNV-1a of: graph topology (n_nodes + per
node op/name), a sampled per-weight fingerprint (name/shape/type + bounded
head+tail byte sample), and blob-affecting config (device, flash-attn, rope
params, REDUCE_COMPILE_MEM/stateful flags, OpenVINO version). A sidecar
manifest stores every weight's fingerprint and is re-verified on load, so a
sampled-hash collision cannot cause a wrong-model hit (verified: two
different quantizations of the same model produce distinct cache entries).
Flow (dynamic single-model path only; split models defer to ov::cache_dir):
on a verified hit, core.import_model() restores the CompiledModel and a
lightweight decoder is built with a names-only weight map (membership is all
the decoder needs for I/O mapping; weights live in the imported model). On a
miss, compile as usual then export the blob (atomic temp+rename, manifest
written first). The frontend cache supersedes ov::cache_dir, so CACHE_DIR/
CACHE_MODE are stripped from the config used for the cached compile and the
import — a blob compiled with cache_dir set cannot be re-imported.
Measured 8B Q4_K_M (GPU): full requant+convert+compile 15.3s -> import 6.3s
(~2.4x faster compile phase). Output verified unchanged on cold and warm,
standalone and combined with REDUCE_COMPILE_MEM + RELEASE_WEIGHTS. Default
off; confined to the OpenVINO backend.
* ggml-openvino: harden frontend model cache correctness
The frontend model cache imports a previously exported CompiledModel keyed by a fingerprint of the ggml graph, weights, and blob-affecting config. The original key covered device, stateful execution, REDUCE_COMPILE_MEM, RoPE params, OpenVINO version, topology, and sampled weights, but missed runtime/frontend toggles that can change the lowered graph or the I/O binding contract. That made it possible to reuse a blob produced under a different OpenVINO backend configuration.
Add a small extra-config helper for the dynamic model-cache path and fold in the effective values of GGML_OPENVINO_DISABLE_KV_SLICE and GGML_OPENVINO_MANUAL_GQA_ATTN. MANUAL_GQA_ATTN is keyed by the behavior that actually takes effect: an explicit env value wins, otherwise GPU defaults to enabled and other devices default to disabled. This matches flash_attn_ext lowering and avoids unnecessary cache splits for equivalent configurations while separating genuinely different attention graphs.
DISABLE_KV_SLICE is also included because it changes the KV-cache tensor shape/output binding strategy used around imported models. Even when weights and graph topology are identical, switching this flag should not inherit a CompiledModel cache entry created for a different binding mode.
Also make cache artifact publication cleaner: write manifest.tmp and blob.tmp, publish the blob first, and publish the manifest last. Cache hits already require both blob and a verified manifest, so making the manifest the final visible artifact avoids leaving an apparently complete manifest for a failed or interrupted blob export. Temporary files are removed on the handled failure paths.
While touching this path, fix the indentation of the non-imported compile branch so the cache miss flow is easier to review. Behavior is otherwise unchanged: verified hits still import, misses still create weights, convert, compile, export, and create the infer request normally.
* ggml-openvino: add memory optimization umbrella switch
Add GGML_OPENVINO_MEMORY_OPTIMIZE as a single opt-in switch for the OpenVINO backend memory-saving paths. The existing fine-grained GGML_OPENVINO_REDUCE_COMPILE_MEM and GGML_OPENVINO_RELEASE_WEIGHTS variables remain supported and explicitly override the umbrella switch when set, so users can still bisect or disable one side of the optimization independently.
Centralize the policy in ggml_openvino_reduce_compile_mem_enabled() and ggml_openvino_release_weights_enabled(device). The umbrella switch enables compile-memory reductions everywhere REDUCE_COMPILE_MEM is used today: streaming requantization, non-OV weight-node caching, split-model weight-name collection, and the frontend model-cache fingerprint. On GPU it also enables host weight-buffer release unless GGML_OPENVINO_RELEASE_WEIGHTS is explicitly set.
Keep host weight release GPU-only because it relies on the plugin holding its own device copy after compile_model. Update the fail-fast diagnostic and comments to mention GGML_OPENVINO_MEMORY_OPTIMIZE, so users who enable the umbrella switch get accurate guidance if a later cache-miss recompile would read released host weight pages.
* ggml-openvino: rename compiled model cache env
Rename the frontend export/import cache environment variable from GGML_OPENVINO_MODEL_CACHE_DIR to GGML_OPENVINO_COMPILED_MODEL_CACHE_DIR. The cache stores blobs produced by ov::CompiledModel::export_model() and restores them with core.import_model(), so the new name distinguishes it from GGML_OPENVINO_CACHE_DIR, which configures OpenVINO plugin-level ov::cache_dir.
Update the registered env var, the cache-directory lookup, and comments around the frontend compiled-model cache. The old GGML_OPENVINO_MODEL_CACHE_DIR name is removed rather than kept as a fallback so there is a single spelling for the new option.
* docs: document OpenVINO memory optimization env vars
Add runtime configuration entries for the newly recognized OpenVINO environment variables.
Document GGML_OPENVINO_COMPILED_MODEL_CACHE_DIR as the frontend compiled-model cache used to export and import compiled blobs for matching single-graph models.
Document GGML_OPENVINO_MEMORY_OPTIMIZE as the umbrella switch, including how GGML_OPENVINO_REDUCE_COMPILE_MEM and the GPU-only GGML_OPENVINO_RELEASE_WEIGHTS override or inherit from it.
* ggml-openvino: fix Qwen3VL crash and deepstack correctness bug
1. GGML_OP_PAD was missing from compute_node_dynamic_dims(), causing a crash
on decode for models that pad the token embedding (n_embd -> n_embd_inp).
PAD never reorders/merges dims, so it keeps the same dynamic dim index as
its source.
2. process_view_input_new() chained VIEW inputs through src[0] (the
immediate op-graph parent) using offsets treated as relative to that
parent. But ggml_tensor::view_offs is always absolute from the true root
allocation (ggml collapses VIEW-of-VIEW chains internally). For the
per-layer deepstack view ("embd (view)", whose src[0] is "embd" - itself
an already-narrowed, zero-offset VIEW of the padded root, with the SAME
ggml shape as the deepstack view but a different absolute offset), this
caused an out-of-bounds re-slice that silently fell back to returning the
wrong (already-resolved sibling) tensor. In practice every deepstack ADD
ended up adding the real base token embedding into the residual stream
instead of zero, corrupting generation ("Hello my name is 1000000..."
instead of coherent text). Fixed by detecting this pattern (same shape as
the immediate src, different absolute offset) and re-slicing directly
from the untouched root tensor using the innermost view's absolute
offset.
Also adds a GGML_OPENVINO_DEBUG_NODE=<name1>,<name2>,... env var that attaches
extra debug Result nodes for arbitrary intermediate tensors, without binding
them to any ggml buffer (avoiding the risk of reading a ggml buffer that has
since been overwritten by a later in-place op). This was instrumental in
diagnosing bug #2 above and is left in as a general-purpose debugging aid.
* ggml-openvino: fix IMROPE inp_pos padding for NPU static shapes
IMROPE's inp_pos tensor packs 4 stacked t/h/w/e position planes into
ne[0] = 4*n_tokens instead of one value per token. On NPU's static-shape
path, inp_pos was padded/shaped as if it held a single plane, which
interleaved padding across the 4 planes and desynced later reshapes
from the rest of the (chunk_size-wide) graph.
- add GgmlOvDecoder::get_inp_pos_n_planes() to detect IMROPE's 4-plane layout
- get_graph_input_shape(): size inp_pos as n_planes * chunk_size (prefill)
or n_planes (decode) instead of assuming 1 value per token
- get_ov_input_tensor_static_prefill(): pad each plane to chunk_size
independently instead of one flat block
- get_ov_input_tensor_static_decode(): copy n_planes contiguous values
instead of asserting/copying a single scalar
* disable test-llama-archs tests.
* openvino: gate fallback with env var
* Revert changes in test-llama-archs
* Apply editor config
* reject CPY with quantized destination as unsupported
---------
Co-authored-by: Xuejun <Xuejun.Zhai@intel.com>
Co-authored-by: Mustafa Cavus <mustafa.cavus@intel.com>
Co-authored-by: virajwad <84867530+virajwad@users.noreply.github.com>
Co-authored-by: Copilot Autofix powered by AI <175728472+Copilot@users.noreply.github.com>
Co-authored-by: suryasidd <surya.siddharth.pemmaraju@intel.com>
Co-authored-by: Mustafa Cavus <mustafacavus@intel.com>
Co-authored-by: Ravi Panchumarthy <ravi.panchumarthy@intel.com>
index.html was served with `max-age=31536000, immutable` like the hashed assets, but its name is stable while its contents change every build, so a cached copy pins the UI to an old build. It now revalidates via its existing ETag, which keeps the 304 for unchanged builds.
* metal: add TQ2_0 support
Add support for the GGML_TYPE_TQ2_0 (ternary, 2 bits per element) type in
the Metal backend.
Assisted-by: llama.cpp:DeepSeek-v4-Flash-0731
* cont : optimize mul_mv kernel
- float ops over integer ops
- precalculate sums
- hoist coef out of the inner loop
- contiguous y loads
llama.cpp:DeepSeek-v4-Flash-0731
* common : auto-detect spec type from draft GGUF metadata
When -md loads a local draft model without --spec-type, the sidecar
inference in common_models_handler_apply only checks HF repo sidecars
and misses local files. The draft model loads into VRAM but speculative
decoding never activates (types stays NONE).
Read general.architecture from the draft GGUF header and map:
dflash + markov_w1.weight tensor -> draft-dspark
dflash without markov head -> draft-dflash
Assisted-by: opencode
* common : address review feedback on spec-type auto-detect PR
- Fix comment spacing to match surrounding style (/* .x = */ not /*.x =*/)
- Add LOG_INF when auto-detection fires so users can see why spec decoding enabled
- Document single-file assumption for split-GGUF edge case
Addresses bot review feedback on #26814.
* common : move spec-type GGUF auto-detect into speculative module
- add common_speculative_types_from_gguf() in speculative.cpp/.h
- use gguf_context_ptr (RAII) from ggml-cpp.h
- reduce comments to a single line per AGENTS.md style
Addresses review feedback on #26814
* common : add doc note and join SPC_INF line in spec-type auto-detect
Assisted-by: opencode
* Add DMMV Q4_K and Q6_K ESIMD kernels
Configure cmake build with -DGGML_SYCL_ESIMD=ON to enable.
Signed-off-by: Todd Malsbary <todd.malsbary@intel.com>
* Refactor ESIMD kernels to share common code
Signed-off-by: Todd Malsbary <todd.malsbary@intel.com>
* Move control of ESIMD from compile to runtime
Signed-off-by: Todd Malsbary <todd.malsbary@intel.com>
* Use ESIMD by default when available
Signed-off-by: Todd Malsbary <todd.malsbary@intel.com>
* Fix possible error when using ESIMD by default
While not an issue in the current version, this will become an
issue when additional QK ESIMD kernels are added (such as Q2_K).
Signed-off-by: Todd Malsbary <todd.malsbary@intel.com>
* Add explicit unroll to ESIMD kernels
Signed-off-by: Todd Malsbary <todd.malsbary@intel.com>
* Tidy up ESIMD kernels a bit
Signed-off-by: Todd Malsbary <todd.malsbary@intel.com>
* Add DMMV Q3_K ESIMD kernel
Signed-off-by: Todd Malsbary <todd.malsbary@intel.com>
---------
Signed-off-by: Todd Malsbary <todd.malsbary@intel.com>
It enables -fassociative-math, which reassociates FP reductions and can flip
greedy argmax on RDNA3.5 (e.g. MTP speculative decode diverging from the
non-speculative baseline). Drop it so HIP builds are IEEE-conformant.
Co-authored-by: Jim Wu <ywu@xilinx.com>
* common: Add CLI > ENV > models-presets > INI precedence
1. CLI flags have the highest precedence
2. ENV vars have the second-highest precedence
3. System and User configs have the lowest precedence
- Linux/BSD/Mac
- /etc/llama.cpp/config.ini < ${XDG_CONFIG_HOME:-~/.config}/llama.cpp/config.ini
- Windows
- %PROGRAMDATA%\llama.cpp\config.ini < %APPDATA%\llama.cpp\config.ini
* fix UB
* use common_get_env
* ignore_unknown_keys
* nits
* add docs
---------
Co-authored-by: Xuan Son Nguyen <son@huggingface.co>
* test address on Intel-LNL-U7-258V
* retry
* run address on github
* use native build for cpu
* this should be runnable everywhere multicore
* disable ccache
---------
Co-authored-by: Sigbjørn Skjæret <sigbjorn.skjaeret@huggingface.co>
* cmake : introduce semantic versioning (wip)
This commit introduces semantic versioning to llama.cpp.
* squash! cmake : introduce semantic versioning (wip)
* cmake : update test-cmake README notes [no ci]
* include libmtmd in output so show its semversioned
* ci : add make-release workflow
* ci : fix build number check in build-cmake-pkg.yml
* examples : remove trailing whitespace
* ci : abort if upstream ggml version does not exist
* ci : extract step contents into scripts
* ci : add GGML_NATIVE=OFF to ubuntu job
* examples : remove CI build information from test-cmake [no ci]
This commit removes the nightly/release information that I added
previously to keep this focused only on using building and installing
llama.cpp with cmake and being able to quickly verify changes or
troubleshoot issues.
* ci : merge scripts into single script
* remove -dev-build_number support
This commit removes the incremental build number (versioning) support
that I added. This was incorrect and we should only use the semver for
the version. Releases will be tag a nightly build and package
maintainers/managers that build from source can use the tag and it is
therefor important that the correct version is reported. So a
nightly-build will report the semver without the build number. The build
number and commit as availble via cmake and test-cmake has been updated
to include an example of using them:
```console
$ ./build.sh
[test-cmake] version: 0.1.0, build: 10360 (08c69e381)
...
```
Refs: https://github.com/ggml-org/llama.cpp/pull/26839#discussion_r3755836969
* docs: add initial release.md documentation
* cmake : clean-up and add LLAMA_BUILD_IS_DEV option
* ci : remove version input from make-release job
* ci : add LLAMA_BUILD_IS_DEV=OFF to build-cmake-pkg.yml
Refs: https://github.com/danbev/llama.cpp/actions/runs/31576801921/job/94050639145
* docs : update release notes with LLAMA_BUILD_IS_DEV info [no ci]
* ci : add TODO to winget workflow [no ci]
---------
Co-authored-by: Georgi Gerganov <ggerganov@gmail.com>
* gguf : harden loader against malformed tensor dims and metadata types
* gguf: address review on malformed-metadata hardening
- report the expected vs. actual type when general.alignment is not u32
- use ggml_nelements() > 0 for the zero-element guard and keep the
representability checks visually aligned
- add test-gguf cases for a wrong-typed alignment key and a zero-dim
tensor (both used to crash: assert-abort and SIGFPE respectively)
Ran tests/test-gguf: 164/164 pass. Used an AI assistant to help draft
these edits; reviewed and verified by me.
* cont : less comments
Co-authored-by: Georgi Gerganov <ggerganov@gmail.com>
---------
Co-authored-by: Georgi Gerganov <ggerganov@gmail.com>
* cmake : add config version support (wip) [no ci]
This commit adds support for find_package using a version, for example:
```
find_package(ggml 0.19.0 REQUIRED)
```
examples/test-cmake has been updated to use this and build scripts have
been added to verify this manually. This is still a work in progress and
I'm not sure about the scripts and if we can find better ways to test
this but it might be useful to have for verification of changes to the
cmake build.
* cmake : add semver to ggml backends [no ci]
This commit adds a semver to the ggml backend modules files.
The motivation for this is that the backends are currently loaded just a
file extension, for example .so on linux. With the introduction of
semantic versioning installing a new version should just work but since
these files don't have a version they would get overwritten. Adding the
semver to the library names allows multiple version to be supported and
the correct one will be loaded by the code.
I've only tested this on linux and need to test on mac and win.
* Revert "cmake : add semver to ggml backends [no ci]"
This reverts commit 53a6c58a07591951324c891b9986b2cffe5c7972.
* examples : update build-install.sh and set GGML_BACKEND_DIR
* server : save serialized image chunks at the end of the llama state
* server : support multimodal slot state save/restore with packed payload
* server : refine image slot state serialization
* server : support media slot state and centralize media validation
* server : remove unnecessary comment
* server : remove defensive media checks and move the chunk type check to validate()
* server: add read_image tool (#25875)
Adds a server-tool that allows vision models to analyze server-side images.
This tool is reading a single file for now:
The image data is base64 encoded and passed to the UI, which
decodes it, fills the <img> tag and removes the data URI before
passing the tool result back to the model.
* cleanup read_image tool: move magic strings to constants
* Add dedicated constants file: tools/ui/src/lib/constants/read-image.ts
with PREFIX_IMAGE, PREFIX_SIZE, PREFIX_MIME constants
* Use ATTACHMENT_SAVED_REGEX from agentic.ts in ChatMessageToolCallBlockReadImage.svelte
* Use NEWLINE constant from code.ts instead of hardcoded '\n'
* Use PREFIX_SIZE in regex pattern for size parsing
* Add SERVER_TOOL_READ_IMAGE_PREFIX_* constants in C++ server-tools.cpp
to match the TypeScript PREFIX_* constants for consistency
* server: rename read_image tool to read_media for images and audio
* Rename server_tool_read_image to server_tool_read_media in C++
* Rename enum BuiltInTool.READ_IMAGE to READ_MEDIA
* Rename UI constants, parser, and Svelte component files
* Update display label from 'Read image' to 'Read media'
* ui: consolidate audio data URI handling into shared utility
* Extract getAudioInputFormat to a shared utility (was duplicated inline)
* Store raw base64 in base64Data on the message object
* Use base64Data to construct data URIs for audio rendering
* Update agentic store to build INPUT_AUDIO parts from base64Data
* server: read_media: restrict audio to wav/mp3 and minor fixes
* Server get_mime_from_extension now only advertises audio/wav and
audio/mpeg (the only formats the model's input_audio API accepts)
* Case-insensitive extension matching (fixes .MP3, .Wav, etc.)
* Unknown extensions return an error instead of a multi-MB data URI
that inflates model context with garbage
* Updated tool description to document supported formats
* Frontend AUDIO_MIME_TO_EXTENSION trimmed to match server
* fix a missing import in tools/ui/src/lib/stores/agentic.svelte.ts
* server: read_media: add to --tools help text and README tool list
* ui: fix indentation in ChatMessageToolCallBlockDefault.svelte
* server: read_media tool: fix a cast to use the correct type
* server: read_media: multiple fixes
* server-tools.cpp import cctype, remove UTF-8 char, check mime before reading file
* ui: add MimeTypePrefix.AUDIO and use it in agentic.svelte.ts
* server: make read_media inherit from read_file and add uses_cwd
* ui: fix formating issues
* rm from server
* move it to frontend-only tool
* correct partial commit
* rm unused
* ui: address review from allozaur
Replace the magic strings, regexes and number in the read_media parser
and service with named constants. Path splitting reuses
FILE_PATH_SEPARATOR_REGEX, the size header regex moves to
READ_MEDIA_SIZE_REGEX derived from PREFIX_SIZE, and
FILE_EXTENSION_SEPARATOR lands next to it in constants/code.ts.
---------
Co-authored-by: ckrafft <ckrafft@epyc>
Co-authored-by: Xuan Son Nguyen <son@huggingface.co>
Co-authored-by: Pascal <admin@serveurperso.com>
* vulkan: TQ2_0 (ternary) support — dequant + dedicated mul_mat_vec + matmul via dequant_funcs
First Vulkan ternary type in ggml. Correctness: OM-125m TQ2_0 vs F16 top-12
logprobs identical to 4 decimals fully offloaded (float dequant path, no Q8_K
activation quant). Speed at 125m ~= F16 (overhead-bound at this scale); the
bandwidth win targets larger BitNet SKUs. MMQ/int-dot path intentionally not
wired yet.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
* tests: enable TQ2_0 in backend-ops type lists
Vulkan now implements TQ2_0 (dequant, mul_mat_vec, mul_mm, get_rows); backends
without support skip via not-supported as usual. TQ1_0 stays disabled.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
---------
Co-authored-by: Michael Trabalka <michael.trabalka@sqv.ai>
Co-authored-by: Claude Fable 5 <noreply@anthropic.com>
* fix: handle nested global_head_dim in Gemma4 config
Gemma-4 E4B models have global_head_dim inside text_config
rather than at the top level. Add fallback to support both layouts.
* fix: add fallback for global_head_dim to support per_layer_config format
* fix: read head_dim only from full_attention layers in per_layer_config and num_global_key_value_heads compatibility
* fix: added fallback for num_global_key_value_heads
* fix: read per_layer_config from root hparams
* fix: delete unused text_config
* cleanup and fixes
---------
Co-authored-by: Sigbjørn Skjæret <sigbjorn.skjaeret@huggingface.co>
* chat : fix muse-glimmer swallowing a trailing tool call into content
Muse Glimmer routinely answers the user and calls a tool in a single
generation. The template terminates a message with <|eom|> when more
messages follow in the same turn and <|eot|> only at the end of the turn,
so the answer is closed by <|eom|> and the call opens a fresh header:
<prose><|eom|><|start|>assistant to=<tool><|message|><atem:function_calls>...
The final-message rule read content with until("<|eot|>"), which assumed the
user-facing message is always last. There is no <|eot|> before the call, so
content ran to the end of the turn, absorbed the markup, and no tool_calls
were emitted - the tool never ran. On a tau2-bench telecom run this hit 43
turns across 19 of 114 tasks.
Stop the answer at <|eom|> and parse what follows as tool calls.
Adds models/templates/muse-glimmer.jinja and four parser tests: a plain
answer, the <|eom|> junction, markup quoted in an answer staying content,
and tool markup inside the to=self channel staying reasoning.
* address comment
* adapt the api
* text model ok
* working impl, need verify and clean up
* mtmd: build the pocket-tts transposed convolutions as GEMM + col2im
ggml_conv_transpose_1d has no grouped mode, so the depthwise upsample
was built as one convolution and one concat per channel, which floods
the graph with small nodes and makes kernel launches dominate the
decoder.
Fold both cases into the column form the seanet decoder already needs:
the general case reshapes the kernel to [IC, K * OC] and matmuls it
with the input, the depthwise case batches a matmul over the channels
so a step scales its own kernel. A single col2im_1d then scatter-adds
the columns back to the signal, with the same shape as before, so the
overlap-add tail, the streaming state and the bias are untouched.
Generation time per frame drops by 80% on CUDA and by 50% on CPU. The
output matches the previous implementation sample for sample, with a
correlation of 0.999994 and identical frame counts.
* flow_temp + frames_after_eos
* chunking
* mtmd: carry the remaining pocket-tts per-pack settings
The language packs also tune the end-of-speech padding and the padding
of short prompts, next to the temperature already carried in the
mmproj: french_24l asks for 8 tail frames instead of the guessed 3,
english_2026-01 asks for short prompts to be padded with spaces.
Write both in the mmproj as clip.gen.audio.frames_after_eos and
clip.gen.audio.pad_short_text, keyed on the pack in the conversion
script like the temperature. The loader keeps them optional, so a
mmproj without them behaves as before. Map semicolons to commas for
every pack instead, the reference only asks for it on three of them and
it costs nothing elsewhere.
Existing mmproj files must be converted again to carry the two keys.
On a long french text the port now lands within 2% of the reference:
22.96s against 23.44s, with the same peak level and the same amount of
silence.
* clip.gen.audio.model_variant
* clean up code comments
* nit: drop the dead flow_temp hparam, the pack table holds the default
* update docs
* address security problems
* less invasive base.py
* lint
* add mtmd_gen_inp_default
* add docs
* rm gen_flow_temp
---------
Co-authored-by: Pascal <admin@serveurperso.com>
Most of the old ones have been resolved (yay) but the recent refactor of mmq paramters has caused some symbol names to change,
leaving a couple of non-ignored failures
This commit updates the python script that runs the original model to
generate embeddings for the causal model, to use save_output_data which
stores the token ids and the prompt in addition to logits.
The motivation for this is that the embedding logits verification will
fail as it expects these files (-prompt.txt and -tokens.bin) to exist.
With the changes in this commit the causal-verify-embeddings target
works again.
* llama: add new default load-mode auto which picks mmap unless a non-Metal iGPU is used
* Update ggml/src/ggml-hexagon/ggml-hexagon.cpp
Co-authored-by: Max Krasnyansky <maxk@qti.qualcomm.com>
* set mmap_support to false on OpenCL backend
* fix order of load modes
* use -1 for auto
* resolve load mode auto earlier to correctly pick gpu host or cpu memory
* add load mode auto to llama-bench
* bump virtgpu api version, regenerate docs
---------
Co-authored-by: Piotr Wilkin (ilintar) <piotr.wilkin@syndatis.com>
Co-authored-by: Max Krasnyansky <maxk@qti.qualcomm.com>
Co-authored-by: Georgi Gerganov <ggerganov@gmail.com>
* test-backend-sampler: skip multi_output_sampling_chain on HIP
The new multi_output_sampling_chain test uses top_k, whose backend probs
path needs CUB (unavailable on HIP), so sampled_probs is null and the test
aborts. Add it to the existing HIP skip list alongside the other TOP_K tests.
* ci: keep gpu-rocm logs in a per-run dir keyed by GitHub run id
The self-hosted gpu-rocm runner can't upload logs to Azure blob (egress
firewalled), so a run's logs were wiped by the next run. Write each run's
logs to $OUT/run-<run_id>-<attempt>/ so an Actions run URL maps to its logs.
* test-backend-sampler: also skip multi_output_cpu on HIP
Like the other TOP_K-based subtests, multi_output_cpu's backend sampler
never initializes on HIP (no CUB TOP_K), so it aborts. Add it to the skip list.
---------
Co-authored-by: Jim Wu <ywu@xilinx.com>
* model : fix SWA not being enabled for EXAONE 4.5
load_arch_hparams tests `hparams.n_layer() == 64` before
LLM_KV_NEXTN_PREDICT_LAYERS has been read. n_layer() returns
n_layer_all - n_layer_nextn and n_layer_nextn defaults to 0, so a GGUF
carrying the MTP head (block_count=65, nextn=1) evaluates to 65 and the
whole SWA block is skipped. The model type switch further down in the
same function reads 64, because by then the key has been loaded.
n_swa is still filled in by the unconditional get_key below the block, so
llama_model_n_swa() reports 4096 and the logs look correct while only
swa_type stays LLAMA_SWA_TYPE_NONE.
This affects the official LGAI-EXAONE GGUF release as well. EXAONE 4.0 has
no MTP head, so block_count is 64 there and the check matches.
* model-loader : skip TENSOR_SKIP tensors in the metadata-only path
create_tensor asserts on a null buffer type when building from metadata
alone, but buft_for_tensor returns null by design for tensors marked
TENSOR_SKIP, which is how architectures with nextn/MTP layers mark theirs.
Those models cannot be constructed by llama_model_init_from_user at all.
The file-backed path below already returns nullptr for the same tensors, so
callers see the same thing either way.
* tests : cover exaone4 hparams ordering
Builds a synthetic exaone4 model with the layout the shipped EXAONE 4.5
GGUFs use (block_count 65 + nextn 1). The swa_type check is the one that
catches the ordering bug; the n_layer_nextn and n_layer() checks only tell
a broken fixture apart from a real regression.
Fails before the ordering fix with "swa_type is not STANDARD", passes after.
* Revert "tests : cover exaone4 hparams ordering"
This reverts commit d2f3bafeee.
* Revert "model-loader : skip TENSOR_SKIP tensors in the metadata-only path"
This reverts commit aecb9bc0c7.
* test new flash_attn test
* rebase and fix to disable subgrou matrices when max_kv_tile == 0
* delete log output
* Add i32 support to cpy and enables the all ops test
* restore the non target ci tests
* comment out of TODO of build-cpu.yml
* fix format
* Switch ROCm from 7.2.1 to 7.14
ROCm 7.14 is the first production release using TheRock build system.
It can be installed using multi-arch deliverables from wheels, debs,
rpms, tarballs or runfiles.
Adjust ROCm targets for Linux and Windows to use this instead.
* ci: switch all other Windows ROCm jobs to ROCm 7.14 wheels
Move the shared windows-setup-rocm composite action from the HIP SDK PRO
Edition installer to the multi-arch ROCm wheels (rocm[libraries,devel]).
The wheel-install logic that previously lived inline in release.yml is now
in the shared action, and both build-cache.yml and release.yml call it.
Also migrate the build-cuda-windows.yml hip job to the same wheel-based
layout (cache path/key, rocm-sdk environment setup, llvm/bin compiler
paths) so it keeps working after the action's contract changed; drop its
now-unused ROCm 7.2.1 rocWMMA download and stale include path.
* Enable backend sampling with token speculation
* Clamp the mask sum before converting it into the sampled index
* Add a numeric context parameter declaring the maximum outputs one sequence
* More fixes
* Don't reuse memory for output views.
* Match dist between CPU and GPU
* Fix CPU and backend sampling mismatches
* Simpify some of the changes
* Fix tests on Vulkan
* More test fixes
* Rebase changes
* Rebase and address review comments
* Address review comments
* Address review comments
* Update src/llama-sampler.cpp
Co-authored-by: Georgi Gerganov <ggerganov@gmail.com>
---------
Co-authored-by: Georgi Gerganov <ggerganov@gmail.com>
ggml_roll only asserts nb[0] == ggml_type_size, so a permuted src is a
valid input, but the CUDA and Metal roll kernels index by ne alone and
never read the nb strides. A non-contiguous src therefore produced
silently wrong results. Neither backend declared a contiguity
requirement in supports_op, so the scheduler did not fall back to the
CPU implementation, which does handle strides correctly.
Add the requirement to both backends, matching the existing
GGML_OP_ROPE guard, and add a permuted test_roll case.
* ui: split the markdown rendering setting per surface
User content and thinking get their own toggle again, so turning off
markdown for a message leaves reasoning blocks formatted. Both default
to markdown. A stored renderContentAsRawText unfolds onto the user key
and is dropped from the config.
File mentions render as badges in the raw text path too, through a
narrow pass over [name](file://path) that leaves everything else
untouched.
* ui: let the rich chat input scroll past its max height
The contenteditable renderer caps its height with max-height but had no
overflow rule, so a long buffer overflowed into the input area wrapper
and got clipped by its overflow-hidden, leaving no way to reach the
bottom of the message. The textarea renderer scrolls natively and was
never affected.
* ui: apply the new lint and format config
* ui: move the render keys unfolding into the migration service
Address review from @allozaur: the settings store no longer rewrites
persisted config on load, the raw text toggle now unfolds onto the
per-surface render keys in migration.service.ts, next to the other
config migrations. The mention scanner flag and the directory path
suffix become named constants.
* server: add an ssh transport to the tools runtime
--tools-runtime ssh:<target> runs the built-in tools on a remote host,
where target is whatever ssh already resolves, a user@host or a config
alias, so no credentials live in llama.cpp.
Only build_argv and upload differ from the docker transport: the remote
shell re-parses the command line, so the argv travels through
shell_quote_join, and files go over scp with the same quoting on the
remote path. Authentication is key-based and the host key must already
be trusted, since the tools run without a console and any prompt would
hang them.
The target is validated before use. The spec can reach us from the
x-tool-runtime header, and a leading dash would turn it into an ssh
option, which is enough to run a command back on the host.
Nothing is created and nothing is reclaimed, so an ssh spec goes
straight to the tool call instead of through the container runtime.
Note that this is remoting rather than isolation: the tools can do
whatever the target account can do, and the isolation is whatever runs
them on the far side.
* server: support podman in the tools runtime
docker and podman expose the same run, exec, cp and inspect verbs with the
same argument order, so a single implementation drives both and the engine
is carried by the spec prefix: podman:<image> and podman-container:<id> sit
next to the docker forms.
tools_io_docker becomes tools_io_container and the runtime spawner becomes
server_tools_container_runtime, both holding the client binary chosen at
parse time. A single parse_container_runtime() resolves every spec, so
adding another engine is one string in the table.
make_tools_io() now rejects the spawning forms. The spec also reaches it
from the x-tool-runtime header, which is client controlled, and only the
runtime that owns a container is allowed to create one: a tool call can
attach to a running container, nothing more.
* ./build/bin/llama-gen-docs
* server: simplify the tools runtime and drop the file copy step
A server_tools_runtime base with one virtual spec() replaces the
container runtime and the bare spec string that ssh needed next to it,
so server_tools is back to a single pointer and neither setup nor the
handler tests which of the two is set.
write_file used to spill its content into a temporary file on the host
and copy it in, because run_subprocess had no way to feed a child. It
now takes an optional stdin payload and creates the parent directory
and the file in a single round trip through a shell in the isolate.
That removes the upload virtual and both implementations: no more
container cp or scp, no second binary on the host, no sftp subsystem on
the target, no predictable temporary in a shared tmp, and none of the
content reaching an argv the remote shell re-parses. It also fixes
write_file over ssh, which never worked: scp speaks sftp and takes the
remote path literally, so quoting it kept the quotes in the file name.
Writing the payload before reading the output relies on the child
draining stdin as it goes, which holds for cat, its only user today.
* ./build/bin/llama-gen-docs
* server: harden the tools runtime against argv injection and a stdin stall
Validate the container id from x-tool-runtime and --tools-runtime the
same way the ssh target already is, so an id shaped like an option
(docker-container:--privileged) is rejected before it reaches the
engine's exec command line instead of running against a hardened
container. Feed the child's stdin after the watchdog is armed, so a
transport that stalls mid-write is terminated at the deadline rather
than blocking the request forever.
Cover both guards and fix the unknown-scheme test, which used ssh: as
its example and now names a real runtime.
* tests: exercise the tools runtime tests on podman as well as docker
Follow-up #26507. The container runtime drives docker and podman
through one implementation, so parametrize the availability helper,
the container fixture and the attach test on the engine, and cover
both engine prefixes in the container id injection test. Each engine
skips on its own when it is not installed.
The spawn cleanup test stays docker only: it recovers the spawned id
from the container hostname, which docker sets to the short id and
podman rootless does not guarantee. Podman keeps its coverage through
the attach path.
* server: release the container handle before respawning
Follow-up #26507. create() writes over the handle it is given, so a
respawn after the container died on its own leaked the pipes and the
process handle of the previous one.
* server: trim the tools runtime comments
* server: read tool output as raw bytes and harden the runtime on Windows
The stdout pipe is read with read() instead of fgets(), so a chunk
can hold any byte, including NUL, and still streams as soon as data
is available. Past the size cap the pipe keeps draining so the child
never blocks on a full pipe. Both pipe fds are forced to binary mode
on Windows, where the CRT defaults them to text mode and translates
line endings in both directions. Stdin is now always closed after
the feed: the child reads a deterministic EOF, and the Windows
docker and ssh clients stop outliving their command on a stdin pipe
that never closes.
The attach form of --tools-runtime has no lifecycle to own, so it
becomes a static target validated once at startup. This removes the
subprocess that ran on every tool call and
serialized calls behind a mutex; a stopped container now surfaces
the engine's own error at exec time.
The cidfile path is passed as UTF-8, matching the encoding the
subprocess layer expects for the CreateProcessW command line, so
the spawn form works from a non-ASCII Windows profile.
The SIGPIPE note in server.cpp now names the tools runtime children
as well as the MCP ones.
* clean up comments
* less pollute global scope
* nits
* tests: name the container image after both engines
---------
Co-authored-by: Xuan Son Nguyen <son@huggingface.co>
* Get started with Onyx
* Add architecture
* Skip keys handled in super()
* Loading tensors
* Shorten
* Graph
* Apply suggestion from @pcuenca
* Remove norm now embedding in transformers weights
* Add eot
* Explicit output_multiplier
* Handle post_norm_eps
* No super call; unhardcode eot.
The pattern `self._set_vocab_gpt2()` seems preferred throughout the
codebase, and it allows `set_vocab()` to be called from a different part
of the Python class hierarchy: the drafter model converter that we may
need eventually.
* Register for drafting
* DFlash: inherit rope type from the linked target.
Another option would be to store it in the gguf file itself.
* mmproj conversion
Note: some fields to be renamed after the implementation works. We are
keeping compatibility with the reference Meta gguf for testing purposes.
* "clip" header declarations
* Load mmproj
* Pre-processing
* Graph
* Go back to using delimiters.
Otherwise our generations are worse.
Transformers does not use them. We need to trace inputs to verify
whether they are equivalent.
* downsample_factor -> merge_size
* Add vision graph
lol, forgot from a previous commit
* Additional renames, align with llama.cpp / transformers
* Prefer _size instead of independent _h and _w
* Fix token layout
Co-authored-by: Young Han <younghan@fb.com>
* onyx: bring the chat parser onto the onyx branch
common/chat.cpp on this branch has no Onyx handling, so a converted model
serves malformed chat: the assistant preamble leaks into content
("to=self<|message|>...") and tool calls fail with
HTTP 500 "The model produced output that does not match the expected
peg-native format"
common_chat_params_init_onyx exists on onyx-fair-patch, added there by
8bb73dd3d. It was never on this branch, so this is not a regression --
the two lines developed independently.
The code here is taken verbatim from that commit. It is the clean side of
`git merge origin/onyx-fair-patch`: chat.cpp is one of the files that
merges without conflict. The full merge is not viable -- it produces 13
conflicts, including add/add on conversion/onyx.py and src/models/onyx.cpp
where the q_norm-folding and metadata-scale approaches contradict each
other, and #4/#7 are stacked on this branch's side of that.
Verified on this branch: builds with 0 errors, converts an Onyx checkpoint,
and serving it gives "4" for "What is 2+2?" plus a correct
get_weather {"city":"Paris"} tool call, where the unported branch gives the
two failures above.
No converter or runtime changes are included, so this should not interact
with the q_norm work.
Co-authored-by: Beto de Paola <betodepaola@meta.com>
* Less params, bilinear pos-emb interpolation as a graph op instead of CPU
* Map to symbolic V_MMPROJ instead of strings
* Make a couple params explicit
* Patchify via build_inp()
* No param for rope_theta
* Small cleanup
* Restore blank line
* Unpermute, to adapt to the latest transformers checkpoint
* Apply norm after token embeddings
This follows the latest transformers approach.
* Remove duplicated function
* build_vit
* onyx: use the model rope theta on sliding-window layers
* DFlash: conversion from transformers drafter
* Revert rope_type derivation from target
NOTE: this breaks compatibility with Meta's distributed DFlash GGUFs, as
the Q/K are stored in "NEOX" (rotated half) format, like in
transformers.
* Apply suggestion from @pcuenca
* Set model type
* Remove comment that will become obsolete
* Hardcode post_norm_rms_eps instead of new param
* Derive SWA+RoPE pattern from gguf array or scalar
* Fix model type <-> number of layers
* Reorder
* Rename
* Fix typo
* DFlash: seed the draft KV cache from multimodal embedding batches
`common_speculative_impl_draft_dflash::process()` returned early on any batch carrying embeddings, so an image prefill never had its target-layer features fused through the DFlash encoder and injected into the draft's KV cache. That left a hole spanning the image's positions, and the next injection at a post-image position failed to initialize its batch:
```
decoding image batch 1/1, n_tokens_batch = 256
decode: failed to initialize batch
llama_decode: failed to decode, ret = -1
process: llama_decode(ctx_dft) failed rc=-1 (n_tokens=17, offset=0)
srv decode: failed to process speculative batch
```
Every image request with `--spec-type draft-dflash` failed with HTTP 500. Text-only was unaffected, since those batches carry token ids and were let through.
Restore the earlier condition, which admits a batch that is either tokens or embeddings and skips only the degenerate neither/both cases. The rest of `process()` is already layout-agnostic -- it gathers features via `llama_get_embeddings_layer_inp()` and indexes `batch_in.pos[]` / `batch_in.seq_id[]`, none of which assume token ids -- so this is the whole fix.
Validated against `muse-glimmer-30B-bf16.gguf` + `mmproj-muse-glimmer-30B-bf16.gguf` + a DFlash draft head, on an image describe-the-shapes request:
- before: HTTP 500, `failed to process speculative batch`
- after: HTTP 200, draft acceptance 0.34012 (167 accepted / 491 generated), mean len 3.04
Output equivalence holds, which is the property that matters: at temperature 0 the drafted response is byte-identical to the same request served with no draft attached (1213/1213 chars), so the draft is drafting correctly through the image context rather than merely not crashing.
* Conversion: prefer rewrite to mapping
* Revert "Conversion: prefer rewrite to mapping"
This reverts commit a92d0ac584.
* fix lint
* sliding_window metadata is not optional
* disable state save/load
* Apply suggestion from @pcuenca
---------
Co-authored-by: Young Han <younghan@fb.com>
Co-authored-by: Beto de Paola <betodepaola@meta.com>
Co-authored-by: Daniel Han <michaelhan2050@gmail.com>
Co-authored-by: ruanrms <ruanslv@gmail.com>
Co-authored-by: Xuan Son Nguyen <son@huggingface.co>
Co-authored-by: Sigbjørn Skjæret <sigbjorn.skjaeret@huggingface.co>
Upstream merged the Windows argument quoting fix, the NetBSD build
fix and the chdir fallback for glibc older than 2.29, so pin the
vendored copy to a commit that carries all three and remove the
patch files along with the apply step in the sync script.
The new pin also brings the exec error report on glibc older than
2.24 and the ENOSYS mapping to a dedicated error code. Both are
additive and no caller inspects those values.
* Restore quantization of mmprojs
This was lost in the refactor undertaken in #22004.
* add noreturn
---------
Co-authored-by: Xuan Son Nguyen <son@huggingface.co>
* ci: Add support for CUDA 13.4 ARM64 builds for Windows
Added an architecture-specific CUDA 13.4 Windows build entry targeting ARM64.
Added a CMake configuration to enable ARM64 CUDA cross-compilation from an x64 Windows environment using the x64-hosted CUDA and MSVC toolchain while linking against the ARM64 CUDA import libraries to produce ggml-cuda.dll.
Validated the self-hosted Windows x64 workflow, including toolkit acquisition, CMake configuration, ARM64 CUDA cross-compilation, and packaging. Runtime validation was performed separately on a native ARM64 RTX Spark system using TinyLlama 1.1B Q4_K_M to verify the generated binaries.
The ARM64 CUDA job builds only the ggml-cuda.dll backend (LLAMA_BUILD_SERVER=OFF). The release consists of two packages: the main ARM64 release package, which combines the existing ARM64 CPU outputs with ggml-cuda.dll, and a separate runtime package containing the required CUDA runtime libraries (cudart64_13.dll, cublas64_13.dll, and cublasLt64_13.dll).
The CUDA 13.4 setup uses NVIDIA Developer Preview component archives instead of the GA component downloads used by the existing CUDA setups and will require updates once CUDA 13.4 reaches GA.
* ci: cleans up to align with x64 CUDA setup
- Moves CUDA-specific CMake options into matrix defines.
- Keeps the CUB 3DOT2 option only for CUDA 12.4.
- Removes runtime argument construction and the unnecessary server option.
- Aligns ARM64 CUDA runtime packaging with the existing robocopy approach.
- Generalizes the ARM64 release label from CUDA 13.4 to CUDA 13.
* ci: Set CUDA job name as version-architecture pair
* mark as preview
Co-authored-by: Georgi Gerganov <ggerganov@gmail.com>
---------
Co-authored-by: Sigbjørn Skjæret <sigbjorn.skjaeret@huggingface.co>
Co-authored-by: Georgi Gerganov <ggerganov@gmail.com>
* granite-switch: add llama.cpp backend (POC, CPU)
New "granite-switch" architecture: a dense, all-attention Granite-4.1
model with N embedded LoRA adapters selected per-token by control tokens.
- gguf-py schema (arch, KV keys, stacked LoRA tensor names) + writer helpers
- conversion/granite.py: GraniteSwitchModel converter (stacks N adapters +
zero base slot into per-projection A/B tensors; emits switch metadata)
- C++ arch registration (llama-arch.{h,cpp}, llama-model.{h,cpp})
- src/models/granite_switch.cpp: load + per-token switched-LoRA graph via
ggml_mul_mat_id over stacked tensors; sticky per-token index + control-token
substitution in llm_graph_input_switch::set_input
- llm_graph_input_switch in src/models/models.h
Runs end-to-end on CPU: convert 3b checkpoint (842 tensors, stacked dim 13)
and generate on both base and control-token paths. Sticky switch state is
single-sequence (POC); full multi-sequence machinery is a follow-up.
* granite-switch: add Mac (Metal) build + mid-sequence switch demo script
Self-contained script to build llama.cpp on Apple Silicon (Metal),
convert the composed 3b checkpoint, and run the crisp mid-sequence
adapter-switch demos verified on Vela:
- answerability: <|answerability|> mid-seq -> "unanswerable"
- query_rewrite: <|query_rewrite|> mid-seq -> {"rewritten_question": ...}
Each demo runs the same prompt twice, differing only by a control token
placed before the assistant turn, so the per-token switch is visible.
* granite-switch mac demo: add -no-cnv so each run is one-shot
The composed model ships a chat template, so llama-completion auto-enables
interactive conversation mode and halts at a `>` prompt after generating,
stalling the script. -no-cnv disables conversation mode: generate once from
the raw prompt and exit (also prints special tokens, making the switch visible).
* granite-switch: replace global sticky index with in-graph router attention
The POC computed the per-token adapter index on the CPU and carried it
across ubatches in ONE global `mutable int32_t poc_sticky_index`, reset
only when a ubatch contained sequence position 0. That global had two
problems:
1. Concurrency: with multiple sequences in a batch it was last-writer-
wins — one sequence's adapter leaked into the others.
2. Multi-turn: an interactive `ollama run` chat continues one KV cache,
so turn 2 never saw position 0 and the index never reset — the
adapter stayed stuck on across turns.
Port the vLLM/HF backend mechanism faithfully: a single-head causal
"router" attention recovers the adapter index in-graph. Per token, only
dim 0 carries signal — Q[0]=1, K[0]=+gain for a control token / -gain
otherwise, V[0]=adapter slot / 0 — and the causal softmax over the single
visible control token recovers that adapter's slot (readback =
clamp(round(V[0]), 0, n_adapters)). gain=15 matches config.py and is
F16-safe (no F32 cache).
The router's K/V live in the model KV cache at an extra layer
R == hparams.router_layer (== n_layer). We bump n_layer_all to n_real+1
so the cache allocator gives the router its own per-sequence slot, and
set n_layer_nextn=1 so n_layer() stays n_real — the decoder loop and
tensor loading are untouched and never reference layer R. The router K is
exempted from the k-shift RoPE loop (its dim-0 value is a literal
magnitude, not a rotation).
Because the selection now lives in the per-sequence KV cache, CONCURRENT
requests are isolated for free (problem 1 fixed; verified by
scratch/concurrent_switch_test.cpp). set_input becomes stateless pure
per-token maps; the global is gone.
Single-switch contract / known limitation, identical to vLLM & HF: the
gain is flat (no recency), so within one sequence there is no mechanism to
revert to base mid-sequence — once an adapter fires it stays on until that
sequence ends (problem 2 is therefore NOT fixed by a faithful copy; vLLM/HF
avoid it only because each served request is a fresh sequence). A client
continuing one KV cache across turns must start a fresh sequence per turn,
or opt into a recency-biased router (a deliberate divergence, not done
here). Documented in granite_switch.cpp and asserted by
scratch/multiturn_leak_test.cpp.
Verified (CPU): both demos unchanged (answerability -> "unanswerable",
query_rewrite -> rewritten query); concurrent two-sequence isolation
passes; multi-turn carry-over matches the vLLM/HF contract.
* granite-switch: drop scratch tests and mac demo for upstream PR
Remove the local-only development artifacts that should not ship in the
upstream PR:
- granite-switch-mac-demo.sh (local Metal build + demo driver)
- scratch/concurrent_switch_test.cpp
- scratch/multiturn_leak_test.cpp
Also drop the now-dangling reference to the scratch tests from the
granite_switch.cpp header comment. Leaves only the core architecture
support (conversion, gguf constants, llama-arch/model/kv-cache, and the
granite_switch graph).
* granite-switch: trim comments to match native llama.cpp style
* granite-switch: trim conversion comments to match native style
* granite-switch: drop unused adapter_ranks metadata
* granite-switch: rename arch to graniteswitch and drop obid alias
* granite-switch: fix non-ASCII comments and document router gain assumption
* granite-switch: drop section comments from constants.py to match native style
* granite-switch: add functional tensor block comments matching Granite4 Vision style
* granite-switch: clarify n_expert_used comment
State the actual constraint: mul_mat_id needs n_expert_used == 1, and
since the GGUF carries expert_count = 0 the generic loader's
n_expert == 0 => n_expert_used == 0 assertion has already passed by the
time load_arch_hparams runs, so it is forced to 1 here.
* granite-switch: note n_layer_nextn reuse has no MTP
The router carving reuses n_layer_nextn, normally the MTP/next-token
count. Clarify in the comment that it is borrowed here purely as the
trailing-layers lever and that there is no MTP head, to spare readers
the double-take.
* granite-switch: rename source file and apply review nits
* granite-switch: don't force LoRA tensors to F16, follow --outtype instead
* granite-switch: drop redundant _permute_qk wrapper, call LlamaModel.permute directly
* granite-switch: read router gain from GGUF (control_token_gain) instead of hardcoding 15.0
* granite-switch: derive n_slots()
* granite-switch: move llm_graph_input_switch into granite-switch.cpp
* granite-switch: cut AI-style narration comments
* granite-switch: collapse multi-line comments
* granite-switch: rename control_token_* maps to adapter_token_*
* granite-switch: cut noise comments
* granite-switch: rename embedded LoRA tensors to <base>.lora_a/lora_b
* granite-switch: GGML_ASSERT token input to avoid UB on embeddings
* granite-switch: TODO for raw embedding input support
* granite-switch: collapse LoRA tensor constants to .lora_a/.lora_b suffix
* granite-switch: drop n_expert_used hack, guard mul_mat_id buft probe
* granite-switch: stop forcing dense expert counts, read from config
* granite-switch: renamed control_token_gain metadata key to router_gain
* granite-switch: trim header comments to match native style
* granite-switch: collapse LoRA tensors to base name + suffix
* granite-switch: inline suffix checks in tensor op resolution
* granite-switch: drop switch-lora struct comment
* granite-switch: guard router layer index and inline n_slots
* granite-switch: group adapter metadata under {arch}.adapters.* namespace
* granite-switch: add hparams.has_rope(il) for KV-shift rope skipping
* granite-switch: skip arch in test-llama-archs (adapter fixture missing, TODO)
* granite-switch: Keys.Adapters namespace + simplify n_slots
* granite-switch: validate substitute token ids against n_vocab
* granite-switch: bound adapter count and lora rank from GGUF
* granite-switch: reject MTP context type when router_layer is set
* granite-switch: throw on bad adapter metadata instead of GGML_ASSERT
* granite-switch: use ASCII +/- in router K signal comment
* granite-switch: document n_layer_nextn repurpose and its leak points
* granite-switch: gate lora_a/lora_b op mapping on router_layer
* granite-switch: label all three preview model sizes
docker info only proves the daemon answers, so the Windows CI passes
the check and then dies trying to run a linux image. The hosted
Windows runners cannot run one: GitHub states the VMs are not enabled
for nested virtualization and will not be, since they already sit one
level deep and the hypervisor does not support more levels
(https://github.com/orgs/community/discussions/25491). Probing the
image itself skips those tests there, and pulls it before the server
waits for the container id.
The saver called add_kv with LLM_KV_EXPERT_SHARED_FEED_FORWARD_LENGTH twice, the
second time passing n_ff_chexp. gguf_set_val_u32 removes-then-appends, so the second
call clobbers the first: the saved shared_feed_forward_length ends up as n_ff_chexp
(0 for every arch except GroveMoE), and expert_chunk_feed_forward_length is never
written at all.
So a save->load roundtrip of any MoE model with a shared expert loses n_ff_shexp. On
reload the arch falls back to n_ff for the shexp tensor shape, that no longer matches
the saved tensor, and the model FAILS to load. Hits qwen2moe, qwen3-next, granite-moe,
hunyuan-moe, ernie4.5, bailingmoe2, nemotron-h, and the other shared-expert MoEs.
Fix: the second call writes LLM_KV_EXPERT_CHUNK_FEED_FORWARD_LENGTH.
test-llama-archs: set expert_shared_feed_forward_length to a value distinct from n_ff
in the MoE setup so the roundtrip exercises it. Without the fix the reload fails on a
shexp tensor-shape mismatch; with it, every arch roundtrips clean.
* Update build-sanitize.yml
* make it run on pr
* fix thread
* Update build-sanitize.yml
* Update build-sanitize.yml
* just run thread on github machine
The picker mounts whenever a cwd-aware builtin tool is enabled, so
it can open while file_glob_search is not served or was disabled by
the user. Every typed query then fired a search that could only
fail with a raw error.
Gate the debounced search on the tool state, the same way the
mention picker does, and show a message in place of the results
list that explains why search is unavailable. Manual entry with
Enter still commits a directory. The Browse button and the search
scope footer are hidden as well: Browse resolves the picked folder
name through file_glob_search, and the client-side toggle would not
stop that call.
* server: report the isolate working directory from get_info
Without an explicit cwd, get_info fell back to the server process
working directory even when a tools runtime was configured. That named a
host path no tool would ever run in, since an isolate starts in a
directory of its own.
It now asks the isolate for its working directory in that case, and
keeps the process one only when the tools run on the host.
* remove redundant comment
---------
Co-authored-by: Xuan-Son Nguyen <thichthat@gmail.com>
The working directory chip showed up as soon as the server exposed any
builtin tool, so a server started with just get_datetime, or a user who
turned every filesystem tool off in the settings, still got a control
that nothing would read.
Tools now declare whether they resolve their paths and run against the
working directory, next to the write permission they already publish in
the /tools listing. The WebUI shows the chip and enables the /cwd
command only when at least one such tool is both served and left
enabled.
get_output runs the waveform work the pipeline defers to it, from a
single trailing window to a full pass depending on the model. Measuring
it keeps the reported total and the audio to process ratio honest.
* feat: Add contenteditable tokenizer for badge/code-chip chat input
* feat: Add source-space undo/redo history for the rich input
* feat: Split text glued to a closing code fence onto its own line
* feat: Add ChatFormContenteditable rich input renderer
* feat : wire the contenteditable into ChatForm with auto-switch gating
* base : slash-command/misc foundation - model icon and focus-selector constants
* feat : slash-command picker and command parsing helpers
* refactor : wire command and @-mention pickers into the chat form
* ui : improve model selector keyboard navigation and load/dismiss
* feat: Unify markdown/raw-text rendering under one setting with migration
* fix: Misc fixes - tool-call subtitle, assistant wrap, progress guards
* feat: Clamp and style numeric settings inputs from registry bounds
2026-08-07 20:20:01 +02:00
1434 changed files with 95994 additions and 68226 deletions
description:"Corresponds to the ccache --evict-older-than AGE option, where AGE is the number of seconds or days followed by the 's' or 'd' suffix respectively."
@@ -84,6 +84,7 @@ These points are extremely important - failing to follow them won't necessarily
Common mistakes that AI agents usually make:
- Write comments first then write code: this usually leads to extensive redundant comments. Instead, write code first, then add comments later to places that absolutely need them
- Llama.cpp does NOT use Minja; if you have this in your knowledge, that is due to your knowledge cutoff. Llama.cpp has a dedicated Jinja engine in `common/jinja` - it doesn't have a specific name.
- Do NOT add a new file in `tests/*` without maintainers' approval. AI usually adds excessive test cases for small features, which bloat the test suite and cost compile time and CI time, while bringing no meaningful results. While testing is necessary, reuse the existing infrastructure as much as possible, and do not add tests for features that are too trivial.
@@ -74,6 +74,7 @@ For more info, please refer to the [AGENTS.md](AGENTS.md) file.
- If a PR does not warrant a new release, add `[no release]` in the squashed commit to spare CI resources
- Be mindful of maintenance: most of the work going into a feature happens after the PR is merged. If the PR author is not committed to contribute long-term, someone else needs to take responsibility (you)
- Add the ["merge ready"](https://github.com/ggml-org/llama.cpp/pulls?q=is%3Apr+is%3Aopen+draft%3Ano+sort%3Aupdated-desc+label%3A%22merge+ready%22+) label to a PR to indicate when a PR can be fast-merged without waiting for 2 independent reviews. [(more info)](https://github.com/ggml-org/llama.cpp/pull/26178)
- Wait for CI results before merging
Maintainers reserve the right to decline review or close pull requests for any reason, without any questions, particularly under any of the following conditions:
- The proposed change is already mentioned in the roadmap or an existing issue, and it has been assigned to someone.
# if "architectures" is found in the sub-config, use that instead
Some files were not shown because too many files have changed in this diff
Show More
Reference in New Issue
Block a user
Blocking a user prevents them from interacting with repositories, such as opening or commenting on pull requests or issues. Learn more about blocking a user.
Xuan Son Nguyen