ClaudeBot/llama.cpp
0
0
Fork 1
mirror of https://github.com/ggml-org/llama.cpp.git synced 2026-09-17 20:31:47 +02:00
Code Issues Packages Projects Releases Wiki Activity

model : skip gate_up_exps if TENSOR_SKIP is set (#29014)

Browse Source 
required for qwen35moe if MTP tensors are fused but not loaded
This commit is contained in:
David Friehs
2026-09-17 13:53:54 +02:00
committed by GitHub
parent 7076180486
commit b49650adb3
1 changed files with 9 additions and 0 deletions
Download Patch File Download Diff File Expand all files Collapse all files
src/llama-model.cpp
+9
View File
@@ -3251,6 +3251,15 @@ ggml_tensor * llama_model_base::create_tensor(const LLM_TN_IMPL & tn, const std:
}
void llama_model_base::create_tensor_gate_up_exps(llama_layer & layer, int bid, int64_t n_embd_, int64_t n_ff_, int64_t n_expert_, int flags) {
if (flags & TENSOR_SKIP) {
const int skip = TENSOR_NOT_REQUIRED | TENSOR_SKIP;
create_tensor(tn(LLM_TENSOR_FFN_GATE_UP_EXPS, "weight", bid), {n_embd_, n_ff_ * 2, n_expert_}, skip | TENSOR_SKIP_IF_VIRTUAL);
create_tensor(tn(LLM_TENSOR_FFN_GATE_EXPS, "weight", bid), {n_embd_, n_ff_, n_expert_}, skip);
create_tensor(tn(LLM_TENSOR_FFN_UP_EXPS, "weight", bid), {n_embd_, n_ff_, n_expert_}, skip);
return;
}
layer.ffn_gate_up_exps = create_tensor(tn(LLM_TENSOR_FFN_GATE_UP_EXPS, "weight", bid), {n_embd_, n_ff_ * 2, n_expert_}, TENSOR_NOT_REQUIRED);
if (layer.ffn_gate_up_exps == nullptr) {
layer.ffn_gate_exps = create_tensor(tn(LLM_TENSOR_FFN_GATE_EXPS, "weight", bid), {n_embd_, n_ff_, n_expert_}, flags);