CUDA: deduplicate mmq code (#7397 )

Tokenizer SPM fixes for phi-3 and llama-spm (bugfix) (#7425 )
* Update brute force test: add_special * Update brute force test: default values for add_bos_token and add_eos_token * Enable rtrim when pre-inserting BOS Co-authored-by: Georgi Gerganov <ggerganov@gmail.com> * Revert "server : fix test regexes"
2026-07-10 14:35:53 +02:00 · 2024-05-21 16:02:12 +02:00 · 2024-05-21 14:39:48 +02:00 · 2024-05-20 20:15:57 +02:00 · 2024-05-21 02:35:28 +10:00 · 2024-05-20 18:15:38 +02:00
10 changed files with 376 additions and 1464 deletions
@@ -107,7 +107,6 @@ Typically finetunes of the base models below are supported as well.
 - [X] [Aquila 1 & 2](https://huggingface.co/models?search=BAAI/Aquila)
 - [X] [Starcoder models](https://github.com/ggerganov/llama.cpp/pull/3187)
 - [X] [Refact](https://huggingface.co/smallcloudai/Refact-1_6B-fim)
- [X] [Persimmon 8B](https://github.com/ggerganov/llama.cpp/pull/3410)
 - [X] [MPT](https://github.com/ggerganov/llama.cpp/pull/3417)
 - [X] [Bloom](https://github.com/ggerganov/llama.cpp/pull/3553)
 - [x] [Yi models](https://huggingface.co/models?search=01-ai/Yi)
@@ -1148,45 +1148,6 @@ class RefactModel(Model):
        return tensors


-@Model.register("PersimmonForCausalLM")
-class PersimmonModel(Model):
-    model_arch = gguf.MODEL_ARCH.PERSIMMON
-
-    def set_gguf_parameters(self):
-        block_count = self.hparams.get("num_layers", self.hparams.get("num_hidden_layers"))
-        head_count = self.hparams["num_attention_heads"]
-        head_count_kv = head_count
-        hidden_size = self.hparams["hidden_size"]
-
-        self.gguf_writer.add_name('persimmon-8b-chat')
-        self.gguf_writer.add_context_length(self.hparams["max_position_embeddings"])
-        self.gguf_writer.add_embedding_length(hidden_size)
-        self.gguf_writer.add_block_count(block_count)
-        self.gguf_writer.add_feed_forward_length(self.hparams["intermediate_size"])
-
-        # NOTE: not sure about this change - why does the model not have a rope dimension count when it is smaller
-        #       than the head size?
-        #       ref: https://github.com/ggerganov/llama.cpp/pull/4889
-        # self.gguf_writer.add_rope_dimension_count(hidden_size // head_count)
-        self.gguf_writer.add_rope_dimension_count(hidden_size // head_count // 2)
-
-        self.gguf_writer.add_head_count(head_count)
-        self.gguf_writer.add_head_count_kv(head_count_kv)
-        self.gguf_writer.add_rope_freq_base(self.hparams["rope_theta"])
-        self.gguf_writer.add_layer_norm_eps(self.hparams["layer_norm_eps"])
-
-    def set_vocab(self):
-        self._set_vocab_sentencepiece()
-        # self.gguf_writer.add_bos_token_id(71013)
-        # self.gguf_writer.add_eos_token_id(71013)
-
-    def extra_f32_tensors(self, name: str, new_name: str, bid: int | None, n_dims: int) -> bool:
-        del name, new_name, bid, n_dims  # unused
-
-        # TODO: FP16 conversion produces garbage outputs. (Q8_0 does not, so..?)
-        return True
-
-
@Model.register("StableLmForCausalLM", "StableLMEpochForCausalLM", "LlavaStableLMEpochForCausalLM")
 class StableLMModel(Model):
    model_arch = gguf.MODEL_ARCH.STABLELM
@@ -1779,6 +1740,38 @@ class Phi3MiniModel(Model):
                    scores[token_id] = -1000.0
                    toktypes[token_id] = SentencePieceTokenTypes.USER_DEFINED

+        tokenizer_config_file = self.dir_model / 'tokenizer_config.json'
+        if tokenizer_config_file.is_file():
+            with open(tokenizer_config_file, "r", encoding="utf-8") as f:
+                tokenizer_config_json = json.load(f)
+                added_tokens_decoder = tokenizer_config_json.get("added_tokens_decoder", {})
+                for token_id, foken_data in added_tokens_decoder.items():
+                    token_id = int(token_id)
+                    token = foken_data["content"].encode("utf-8")
+                    if toktypes[token_id] != SentencePieceTokenTypes.UNKNOWN:
+                        assert tokens[token_id] == token
+                    tokens[token_id] = token
+                    scores[token_id] = -1000.0
+                    toktypes[token_id] = SentencePieceTokenTypes.USER_DEFINED
+                    if foken_data.get("special"):
+                        toktypes[token_id] = SentencePieceTokenTypes.CONTROL
+
+        tokenizer_file = self.dir_model / 'tokenizer.json'
+        if tokenizer_file.is_file():
+            with open(tokenizer_file, "r", encoding="utf-8") as f:
+                tokenizer_json = json.load(f)
+                added_tokens = tokenizer_json.get("added_tokens", [])
+                for foken_data in added_tokens:
+                    token_id = int(foken_data["id"])
+                    token = foken_data["content"].encode("utf-8")
+                    if toktypes[token_id] != SentencePieceTokenTypes.UNKNOWN:
+                        assert tokens[token_id] == token
+                    tokens[token_id] = token
+                    scores[token_id] = -1000.0
+                    toktypes[token_id] = SentencePieceTokenTypes.USER_DEFINED
+                    if foken_data.get("special"):
+                        toktypes[token_id] = SentencePieceTokenTypes.CONTROL
+
        self.gguf_writer.add_tokenizer_model("llama")
        self.gguf_writer.add_tokenizer_pre("default")
        self.gguf_writer.add_token_list(tokens)
@@ -1,143 +0,0 @@
-#!/usr/bin/env python3
-from __future__ import annotations
-
-import logging
-import argparse
-import os
-import sys
-from pathlib import Path
-from pprint import pprint
-
-import torch
-from sentencepiece import SentencePieceProcessor
-
-if 'NO_LOCAL_GGUF' not in os.environ:
-    sys.path.insert(1, str(Path(__file__).parent / 'gguf-py'))
-import gguf
-
-logger = logging.getLogger("persimmon-to-gguf")
-
-
-def _flatten_dict(dct, tensors, prefix=None):
-    assert isinstance(dct, dict)
-    for key in dct.keys():
-        new_prefix = prefix + '.' + key if prefix is not None else key
-        if isinstance(dct[key], torch.Tensor):
-            tensors[new_prefix] = dct[key]
-        elif isinstance(dct[key], dict):
-            _flatten_dict(dct[key], tensors, new_prefix)
-        else:
-            raise ValueError(type(dct[key]))
-    return None
-
-
-def _get_sentencepiece_tokenizer_info(dir_model: Path):
-    tokenizer_path = dir_model / 'adept_vocab.model'
-    logger.info('getting sentencepiece tokenizer from', tokenizer_path)
-    tokenizer = SentencePieceProcessor(str(tokenizer_path))
-    logger.info('adding tokens')
-    tokens: list[bytes] = []
-    scores: list[float] = []
-    toktypes: list[int] = []
-
-    for i in range(tokenizer.vocab_size()):
-        text: bytes
-        score: float
-
-        piece = tokenizer.id_to_piece(i)
-        text = piece.encode("utf-8")
-        score = tokenizer.get_score(i)
-
-        toktype = 1
-        if tokenizer.is_unknown(i):
-            toktype = 2
-        if tokenizer.is_control(i):
-            toktype = 3
-        if tokenizer.is_unused(i):
-            toktype = 5
-        if tokenizer.is_byte(i):
-            toktype = 6
-
-        tokens.append(text)
-        scores.append(score)
-        toktypes.append(toktype)
-        pass
-    return tokens, scores, toktypes
-
-
-def main():
-    parser = argparse.ArgumentParser(description="Convert a Persimmon model from Adept (e.g. Persimmon 8b chat) to a GGML compatible file")
-    parser.add_argument("--outfile",             type=Path, help="path to write to; default: based on input")
-    parser.add_argument("--ckpt-path",           type=Path, help="path to persimmon checkpoint .pt file")
-    parser.add_argument("--model-dir",           type=Path, help="directory containing model e.g. 8b_chat_model_release")
-    parser.add_argument("--adept-inference-dir", type=str,  help="path to adept-inference code directory")
-    parser.add_argument("--verbose",  action="store_true",  help="increase output verbosity")
-    args = parser.parse_args()
-    logging.basicConfig(level=logging.DEBUG if args.verbose else logging.INFO)
-    sys.path.append(str(args.adept_inference_dir))
-    persimmon_model = torch.load(args.ckpt_path)
-    hparams = persimmon_model['args']
-    pprint(hparams)
-    tensors: dict[str, torch.Tensor] = {}
-    _flatten_dict(persimmon_model['model'], tensors, None)
-
-    arch = gguf.MODEL_ARCH.PERSIMMON
-    gguf_writer = gguf.GGUFWriter(args.outfile, gguf.MODEL_ARCH_NAMES[arch])
-
-    block_count = hparams.num_layers
-    head_count = hparams.num_attention_heads
-    head_count_kv = head_count
-    ctx_length = hparams.seq_length
-    hidden_size = hparams.hidden_size
-
-    gguf_writer.add_name('persimmon-8b-chat')
-    gguf_writer.add_context_length(ctx_length)
-    gguf_writer.add_embedding_length(hidden_size)
-    gguf_writer.add_block_count(block_count)
-    gguf_writer.add_feed_forward_length(hparams.ffn_hidden_size)
-    # ref: https://github.com/ggerganov/llama.cpp/pull/4889/commits/eea19039fc52ea2dbd1aab45b59ab4e3e29a3443
-    gguf_writer.add_rope_dimension_count(hidden_size // head_count // 2)
-    gguf_writer.add_head_count(head_count)
-    gguf_writer.add_head_count_kv(head_count_kv)
-    gguf_writer.add_rope_freq_base(hparams.rotary_emb_base)
-    gguf_writer.add_layer_norm_eps(hparams.layernorm_epsilon)
-
-    tokens, scores, toktypes = _get_sentencepiece_tokenizer_info(args.model_dir)
-    gguf_writer.add_tokenizer_model('llama')
-    gguf_writer.add_tokenizer_pre('default')
-    gguf_writer.add_token_list(tokens)
-    gguf_writer.add_token_scores(scores)
-    gguf_writer.add_token_types(toktypes)
-    gguf_writer.add_bos_token_id(71013)
-    gguf_writer.add_eos_token_id(71013)
-
-    tensor_map = gguf.get_tensor_name_map(arch, block_count)
-    logger.info(tensor_map)
-    for name in tensors.keys():
-        data_torch = tensors[name]
-        if name.endswith(".self_attention.rotary_emb.inv_freq"):
-            continue
-        old_dtype = data_torch.dtype
-        # TODO: FP16 conversion produces garbage outputs. (Q8_0 does not, so..?)
-        data = data_torch.to(torch.float32).squeeze().numpy()
-        new_name = tensor_map.get_name(name, try_suffixes = (".weight", ".bias"))
-        if new_name is None:
-            raise ValueError(f"Can not map tensor '{name}'")
-
-        n_dims = len(data.shape)
-        logger.debug(f"{new_name}, n_dims = {str(n_dims)}, {str(old_dtype)} --> {str(data.dtype)}")
-        gguf_writer.add_tensor(new_name, data)
-    logger.info("gguf: write header")
-    gguf_writer.write_header_to_file()
-    logger.info("gguf: write metadata")
-    gguf_writer.write_kv_data_to_file()
-    logger.info("gguf: write tensors")
-    gguf_writer.write_tensors_to_file()
-
-    gguf_writer.close()
-
-    logger.info(f"gguf: model successfully exported to '{args.outfile}'")
-
-
-if __name__ == '__main__':
-    main()
@@ -42,10 +42,13 @@ In addition to the KL divergence the following statistics are calculated with `-

 Results were generated using the CUDA backend and are sorted by Kullback-Leibler divergence relative to FP16.
 The "WT" importance matrices were created using varying numbers of Wikitext tokens and can be found [here](https://huggingface.co/JohannesGaessler/llama.cpp_importance_matrices/blob/main/imatrix-llama_3-8b-f16-2.7m_tokens.dat).
+Note: the FP16 logits used for the calculation of all metrics other than perplexity are stored in a binary file between runs.
+In order to save space this file does **not** contain the exact same FP32 logits but instead casts them to 16 bit unsigned integers (with some scaling).
+So the "f16" results are to be understood as the difference resulting only from this downcast.

 | Quantization | imatrix | Model size [GiB] | PPL                    | ΔPPL                   | KLD                   | Mean Δp           | RMS Δp           |
 |--------------|---------|------------------|------------------------|------------------------|-----------------------|-------------------|------------------|
-| f16          | None    |            14.97 | 6.233160 ±   0.037828  | -                      | -                     | -                 | -                |
+| f16          | None    |            14.97 | 6.233160 ±   0.037828  | 0.001524 ±   0.000755  | 0.000551 ±   0.000002 |  0.001 ± 0.002 %  | 0.787 ± 0.004 %  |
 | q8_0         | None    |             7.96 | 6.234284 ±   0.037878  | 0.002650 ±   0.001006  | 0.001355 ±   0.000006 | -0.019 ± 0.003 %  | 1.198 ± 0.007 %  |
 | q6_K         | None    |             6.14 | 6.253382 ±   0.038078  | 0.021748 ±   0.001852  | 0.005452 ±   0.000035 | -0.007 ± 0.006 %  | 2.295 ± 0.019 %  |
 | q5_K_M       | None    |             5.33 | 6.288607 ±   0.038338  | 0.056974 ±   0.002598  | 0.010762 ±   0.000079 | -0.114 ± 0.008 %  | 3.160 ± 0.031 %  |
@@ -115,7 +115,6 @@ class MODEL_ARCH(IntEnum):
    GPTNEOX    = auto()
    MPT        = auto()
    STARCODER  = auto()
-    PERSIMMON  = auto()
    REFACT     = auto()
    BERT       = auto()
    NOMIC_BERT = auto()
@@ -193,7 +192,6 @@ MODEL_ARCH_NAMES: dict[MODEL_ARCH, str] = {
    MODEL_ARCH.GPTNEOX:        "gptneox",
    MODEL_ARCH.MPT:            "mpt",
    MODEL_ARCH.STARCODER:      "starcoder",
-    MODEL_ARCH.PERSIMMON:      "persimmon",
    MODEL_ARCH.REFACT:         "refact",
    MODEL_ARCH.BERT:           "bert",
    MODEL_ARCH.NOMIC_BERT:     "nomic-bert",
@@ -426,20 +424,6 @@ MODEL_TENSORS: dict[MODEL_ARCH, list[MODEL_TENSOR]] = {
        MODEL_TENSOR.FFN_DOWN,
        MODEL_TENSOR.FFN_UP,
    ],
-    MODEL_ARCH.PERSIMMON: [
-        MODEL_TENSOR.TOKEN_EMBD,
-        MODEL_TENSOR.OUTPUT,
-        MODEL_TENSOR.OUTPUT_NORM,
-        MODEL_TENSOR.ATTN_NORM,
-        MODEL_TENSOR.ATTN_QKV,
-        MODEL_TENSOR.ATTN_OUT,
-        MODEL_TENSOR.FFN_NORM,
-        MODEL_TENSOR.FFN_DOWN,
-        MODEL_TENSOR.FFN_UP,
-        MODEL_TENSOR.ATTN_Q_NORM,
-        MODEL_TENSOR.ATTN_K_NORM,
-        MODEL_TENSOR.ATTN_ROT_EMBD,
-    ],
    MODEL_ARCH.REFACT: [
        MODEL_TENSOR.TOKEN_EMBD,
        MODEL_TENSOR.OUTPUT_NORM,
@@ -756,9 +740,6 @@ MODEL_TENSOR_SKIP: dict[MODEL_ARCH, list[MODEL_TENSOR]] = {
        MODEL_TENSOR.ROPE_FREQS,
        MODEL_TENSOR.ATTN_ROT_EMBD,
    ],
-    MODEL_ARCH.PERSIMMON: [
-        MODEL_TENSOR.ROPE_FREQS,
-    ],
    MODEL_ARCH.QWEN: [
        MODEL_TENSOR.ROPE_FREQS,
        MODEL_TENSOR.ATTN_ROT_EMBD,
@@ -202,7 +202,6 @@ enum llm_arch {
    LLM_ARCH_GPTNEOX,
    LLM_ARCH_MPT,
    LLM_ARCH_STARCODER,
-    LLM_ARCH_PERSIMMON,
    LLM_ARCH_REFACT,
    LLM_ARCH_BERT,
    LLM_ARCH_NOMIC_BERT,
@@ -239,7 +238,6 @@ static const std::map<llm_arch, const char *> LLM_ARCH_NAMES = {
    { LLM_ARCH_MPT,             "mpt"          },
    { LLM_ARCH_BAICHUAN,        "baichuan"     },
    { LLM_ARCH_STARCODER,       "starcoder"    },
-    { LLM_ARCH_PERSIMMON,       "persimmon"    },
    { LLM_ARCH_REFACT,          "refact"       },
    { LLM_ARCH_BERT,            "bert"         },
    { LLM_ARCH_NOMIC_BERT,      "nomic-bert"   },
@@ -595,23 +593,6 @@ static const std::map<llm_arch, std::map<llm_tensor, std::string>> LLM_TENSOR_NA
            { LLM_TENSOR_FFN_UP,          "blk.%d.ffn_up" },
        },
    },
-    {
-        LLM_ARCH_PERSIMMON,
-        {
-            { LLM_TENSOR_TOKEN_EMBD,      "token_embd"},
-            { LLM_TENSOR_OUTPUT_NORM,     "output_norm"},
-            { LLM_TENSOR_OUTPUT,          "output"},
-            { LLM_TENSOR_ATTN_NORM,       "blk.%d.attn_norm"},
-            { LLM_TENSOR_ATTN_QKV,        "blk.%d.attn_qkv"},
-            { LLM_TENSOR_ATTN_OUT,        "blk.%d.attn_output"},
-            { LLM_TENSOR_ATTN_Q_NORM,     "blk.%d.attn_q_norm"},
-            { LLM_TENSOR_ATTN_K_NORM,     "blk.%d.attn_k_norm"},
-            { LLM_TENSOR_FFN_NORM,        "blk.%d.ffn_norm"},
-            { LLM_TENSOR_FFN_DOWN,        "blk.%d.ffn_down"},
-            { LLM_TENSOR_FFN_UP,          "blk.%d.ffn_up"},
-            { LLM_TENSOR_ATTN_ROT_EMBD,   "blk.%d.attn_rot_embd"},
-        },
-    },
    {
        LLM_ARCH_MPT,
        {
@@ -3967,14 +3948,6 @@ static void llm_load_hparams(
                    default: model.type = e_model::MODEL_UNKNOWN;
                }
            } break;
-        case LLM_ARCH_PERSIMMON:
-            {
-                ml.get_key(LLM_KV_ATTENTION_LAYERNORM_EPS, hparams.f_norm_eps);
-                switch (hparams.n_layer) {
-                    case 36: model.type = e_model::MODEL_8B; break;
-                    default: model.type = e_model::MODEL_UNKNOWN;
-                }
-            } break;
        case LLM_ARCH_REFACT:
            {
                ml.get_key(LLM_KV_ATTENTION_LAYERNORM_RMS_EPS, hparams.f_norm_rms_eps);
@@ -4580,7 +4553,8 @@ static void llm_load_vocab(
                        (t.first == "<|eot_id|>" ||
                         t.first == "<|im_end|>" ||
                         t.first == "<|end|>" ||
-                         t.first == "<end_of_turn>"
+                         t.first == "<end_of_turn>" ||
+                         t.first == "<|endoftext|>"
                        )
                   ) {
                    vocab.special_eot_id = t.second;
@@ -5221,47 +5195,6 @@ static bool llm_load_tensors(
                        layer.ffn_up_b   = ml.create_tensor(ctx_layer, tn(LLM_TENSOR_FFN_UP, "bias", i),     {n_ff});
                    }
                } break;
-            case LLM_ARCH_PERSIMMON:
-                {
-                    model.tok_embd = ml.create_tensor(ctx_input, tn(LLM_TENSOR_TOKEN_EMBD, "weight"),  {n_embd, n_vocab});
-
-                    {
-                        model.output_norm    = ml.create_tensor(ctx_output,       tn(LLM_TENSOR_OUTPUT_NORM, "weight"), {n_embd});
-                        model.output_norm_b  = ml.create_tensor(ctx_output,       tn(LLM_TENSOR_OUTPUT_NORM, "bias"),   {n_embd});
-                        model.output         = ml.create_tensor(ctx_output_split, tn(LLM_TENSOR_OUTPUT,      "weight"), {n_embd, n_vocab});
-                    }
-
-                    for (int i = 0; i < n_layer; ++i) {
-                        ggml_context * ctx_layer = ctx_for_layer(i);
-                        ggml_context * ctx_split = ctx_for_layer_split(i);
-
-                        auto & layer = model.layers[i];
-
-                        layer.attn_norm     = ml.create_tensor(ctx_layer, tn(LLM_TENSOR_ATTN_NORM,   "weight", i), {n_embd});
-                        layer.attn_norm_b   = ml.create_tensor(ctx_layer, tn(LLM_TENSOR_ATTN_NORM,   "bias",   i), {n_embd});
-
-                        layer.wqkv          = ml.create_tensor(ctx_split, tn(LLM_TENSOR_ATTN_QKV,    "weight", i), {n_embd, n_embd + 2*n_embd_gqa});
-                        layer.bqkv          = ml.create_tensor(ctx_layer, tn(LLM_TENSOR_ATTN_QKV,    "bias",   i), {n_embd + 2*n_embd_gqa});
-
-                        layer.wo            = ml.create_tensor(ctx_split, tn(LLM_TENSOR_ATTN_OUT,    "weight", i), {n_embd, n_embd});
-                        layer.bo            = ml.create_tensor(ctx_layer, tn(LLM_TENSOR_ATTN_OUT,    "bias",   i), {n_embd});
-
-                        layer.ffn_down      = ml.create_tensor(ctx_split, tn(LLM_TENSOR_FFN_DOWN,    "weight", i), {n_ff, n_embd});
-                        layer.ffn_down_b    = ml.create_tensor(ctx_layer, tn(LLM_TENSOR_FFN_DOWN,    "bias",   i), {n_embd});
-
-                        layer.ffn_up        = ml.create_tensor(ctx_split, tn(LLM_TENSOR_FFN_UP,      "weight", i), {n_embd, n_ff});
-                        layer.ffn_up_b      = ml.create_tensor(ctx_layer, tn(LLM_TENSOR_FFN_UP,      "bias",   i), {n_ff});
-
-                        layer.ffn_norm      = ml.create_tensor(ctx_layer, tn(LLM_TENSOR_FFN_NORM,    "weight", i), {n_embd});
-                        layer.ffn_norm_b    = ml.create_tensor(ctx_layer, tn(LLM_TENSOR_FFN_NORM,    "bias",   i), {n_embd});
-
-                        layer.attn_q_norm   = ml.create_tensor(ctx_layer, tn(LLM_TENSOR_ATTN_Q_NORM, "weight", i), {64});
-                        layer.attn_q_norm_b = ml.create_tensor(ctx_layer, tn(LLM_TENSOR_ATTN_Q_NORM, "bias",   i), {64});
-
-                        layer.attn_k_norm   = ml.create_tensor(ctx_layer, tn(LLM_TENSOR_ATTN_K_NORM, "weight", i), {64});
-                        layer.attn_k_norm_b = ml.create_tensor(ctx_layer, tn(LLM_TENSOR_ATTN_K_NORM, "bias",   i), {64});
-                    }
-                } break;
            case LLM_ARCH_BERT:
            case LLM_ARCH_NOMIC_BERT:
                {
@@ -7923,213 +7856,6 @@ struct llm_build_context {
        return gf;
    }

-    struct ggml_cgraph * build_persimmon() {
-        struct ggml_cgraph * gf = ggml_new_graph_custom(ctx0, LLAMA_MAX_NODES, false);
-
-        const int64_t n_embd_head = hparams.n_embd_head_v;
-        GGML_ASSERT(n_embd_head   == hparams.n_embd_head_k);
-        GGML_ASSERT(n_embd_head/2 == hparams.n_rot);
-
-        struct ggml_tensor * cur;
-        struct ggml_tensor * inpL;
-
-        inpL = llm_build_inp_embd(ctx0, lctx, hparams, batch, model.tok_embd, cb);
-
-        // inp_pos - contains the positions
-        struct ggml_tensor * inp_pos = build_inp_pos();
-
-        // KQ_mask (mask for 1 head, it will be broadcasted to all heads)
-        struct ggml_tensor * KQ_mask = build_inp_KQ_mask();
-
-        for (int il = 0; il < n_layer; ++il) {
-            struct ggml_tensor * residual = inpL;
-
-            cur = llm_build_norm(ctx0, inpL, hparams,
-                    model.layers[il].attn_norm,
-                    model.layers[il].attn_norm_b,
-                    LLM_NORM, cb, il);
-            cb(cur, "attn_norm", il);
-
-            // self attention
-            {
-                cur = ggml_mul_mat(ctx0, model.layers[il].wqkv, cur);
-                cb(cur, "wqkv", il);
-
-                cur = ggml_add(ctx0, cur, model.layers[il].bqkv);
-                cb(cur, "bqkv", il);
-
-                // split qkv
-                GGML_ASSERT(n_head_kv == n_head);
-
-                struct ggml_tensor * tmpqkv = ggml_reshape_4d(ctx0, cur, n_embd_head, 3, n_head, n_tokens);
-                cb(tmpqkv, "tmpqkv", il);
-
-                struct ggml_tensor * tmpqkv_perm = ggml_cont(ctx0, ggml_permute(ctx0, tmpqkv, 0, 3, 1, 2));
-                cb(tmpqkv_perm, "tmpqkv", il);
-
-                struct ggml_tensor * tmpq = ggml_view_3d(
-                        ctx0, tmpqkv_perm, n_embd_head, n_head, n_tokens,
-                        ggml_element_size(tmpqkv_perm) * n_embd_head,
-                        ggml_element_size(tmpqkv_perm) * n_embd_head * n_head,
-                        0
-                        );
-                cb(tmpq, "tmpq", il);
-
-                struct ggml_tensor * tmpk = ggml_view_3d(
-                        ctx0, tmpqkv_perm, n_embd_head, n_head, n_tokens,
-                        ggml_element_size(tmpqkv_perm) * n_embd_head,
-                        ggml_element_size(tmpqkv_perm) * n_embd_head * n_head,
-                        ggml_element_size(tmpqkv_perm) * n_embd_head * n_head * n_tokens
-                        );
-                cb(tmpk, "tmpk", il);
-
-                // Q/K Layernorm
-                tmpq = llm_build_norm(ctx0, tmpq, hparams,
-                        model.layers[il].attn_q_norm,
-                        model.layers[il].attn_q_norm_b,
-                        LLM_NORM, cb, il);
-                cb(tmpq, "tmpq", il);
-
-                tmpk = llm_build_norm(ctx0, tmpk, hparams,
-                        model.layers[il].attn_k_norm,
-                        model.layers[il].attn_k_norm_b,
-                        LLM_NORM, cb, il);
-                cb(tmpk, "tmpk", il);
-
-                // RoPE the first n_rot of q/k, pass the other half, and concat.
-                struct ggml_tensor * qrot = ggml_view_3d(
-                        ctx0, tmpq, n_rot, n_head, n_tokens,
-                        ggml_element_size(tmpq) * n_embd_head,
-                        ggml_element_size(tmpq) * n_embd_head * n_head,
-                        0
-                        );
-                cb(qrot, "qrot", il);
-
-                struct ggml_tensor * krot = ggml_view_3d(
-                        ctx0, tmpk, n_rot, n_head, n_tokens,
-                        ggml_element_size(tmpk) * n_embd_head,
-                        ggml_element_size(tmpk) * n_embd_head * n_head,
-                        0
-                        );
-                cb(krot, "krot", il);
-
-                // get the second half of tmpq, e.g tmpq[n_rot:, :, :]
-                struct ggml_tensor * qpass = ggml_view_3d(
-                        ctx0, tmpq, n_rot, n_head, n_tokens,
-                        ggml_element_size(tmpq) * n_embd_head,
-                        ggml_element_size(tmpq) * n_embd_head * n_head,
-                        ggml_element_size(tmpq) * n_rot
-                        );
-                cb(qpass, "qpass", il);
-
-                struct ggml_tensor * kpass = ggml_view_3d(
-                        ctx0, tmpk, n_rot, n_head, n_tokens,
-                        ggml_element_size(tmpk) * n_embd_head,
-                        ggml_element_size(tmpk) * n_embd_head * n_head,
-                        ggml_element_size(tmpk) * n_rot
-                        );
-                cb(kpass, "kpass", il);
-
-                struct ggml_tensor * qrotated = ggml_rope_custom(
-                    ctx0, qrot, inp_pos, n_rot, rope_type, 0, n_orig_ctx,
-                    freq_base, freq_scale, ext_factor, attn_factor, beta_fast, beta_slow
-                );
-                cb(qrotated, "qrotated", il);
-
-                struct ggml_tensor * krotated = ggml_rope_custom(
-                    ctx0, krot, inp_pos, n_rot, rope_type, 0, n_orig_ctx,
-                    freq_base, freq_scale, ext_factor, attn_factor, beta_fast, beta_slow
-                );
-                cb(krotated, "krotated", il);
-
-                // ggml currently only supports concatenation on dim=2
-                // so we need to permute qrot, qpass, concat, then permute back.
-                qrotated = ggml_cont(ctx0, ggml_permute(ctx0, qrotated, 2, 1, 0, 3));
-                cb(qrotated, "qrotated", il);
-
-                krotated = ggml_cont(ctx0, ggml_permute(ctx0, krotated, 2, 1, 0, 3));
-                cb(krotated, "krotated", il);
-
-                qpass = ggml_cont(ctx0, ggml_permute(ctx0, qpass, 2, 1, 0, 3));
-                cb(qpass, "qpass", il);
-
-                kpass = ggml_cont(ctx0, ggml_permute(ctx0, kpass, 2, 1, 0, 3));
-                cb(kpass, "kpass", il);
-
-                struct ggml_tensor * Qcur = ggml_concat(ctx0, qrotated, qpass);
-                cb(Qcur, "Qcur", il);
-
-                struct ggml_tensor * Kcur = ggml_concat(ctx0, krotated, kpass);
-                cb(Kcur, "Kcur", il);
-
-                struct ggml_tensor * Q = ggml_cont(ctx0, ggml_permute(ctx0, Qcur, 2, 1, 0, 3));
-                cb(Q, "Q", il);
-
-                Kcur = ggml_cont(ctx0, ggml_permute(ctx0, Kcur, 2, 1, 0, 3));
-                cb(Kcur, "Kcur", il);
-
-                struct ggml_tensor * Vcur = ggml_view_3d(
-                        ctx0, tmpqkv_perm, n_embd_head, n_head, n_tokens,
-                        ggml_element_size(tmpqkv_perm) * n_embd_head,
-                        ggml_element_size(tmpqkv_perm) * n_embd_head * n_head,
-                        ggml_element_size(tmpqkv_perm) * n_embd_head * n_head * n_tokens * 2
-                        );
-                cb(Vcur, "Vcur", il);
-
-                cur = llm_build_kv(ctx0, model, hparams, cparams, kv_self, gf,
-                        model.layers[il].wo, model.layers[il].bo,
-                        Kcur, Vcur, Q, KQ_mask, n_tokens, kv_head, n_kv, 1.0f/sqrtf(float(n_embd_head)), cb, il);
-            }
-
-            if (il == n_layer - 1) {
-                // skip computing output for unused tokens
-                struct ggml_tensor * inp_out_ids = build_inp_out_ids();
-                cur      = ggml_get_rows(ctx0,      cur, inp_out_ids);
-                residual = ggml_get_rows(ctx0, residual, inp_out_ids);
-            }
-
-            struct ggml_tensor * ffn_inp = ggml_add(ctx0, residual, cur);
-            cb(ffn_inp, "ffn_inp", il);
-
-            // feed-forward network
-            {
-                cur = llm_build_norm(ctx0, ffn_inp, hparams,
-                        model.layers[il].ffn_norm,
-                        model.layers[il].ffn_norm_b,
-                        LLM_NORM, cb, il);
-                cb(cur, "ffn_norm", il);
-
-                cur = llm_build_ffn(ctx0, cur,
-                        model.layers[il].ffn_up,   model.layers[il].ffn_up_b,
-                        NULL,                      NULL,
-                        model.layers[il].ffn_down, model.layers[il].ffn_down_b,
-                        NULL,
-                        LLM_FFN_RELU_SQR, LLM_FFN_SEQ, cb, il);
-                cb(cur, "ffn_out", il);
-            }
-
-            cur = ggml_add(ctx0, cur, ffn_inp);
-            cb(cur, "l_out", il);
-
-            inpL = cur;
-        }
-
-        cur = inpL;
-
-        cur = llm_build_norm(ctx0, cur, hparams,
-                model.output_norm,
-                model.output_norm_b,
-                LLM_NORM, cb, -1);
-        cb(cur, "result_norm", -1);
-
-        cur = ggml_mul_mat(ctx0, model.output, cur);
-        cb(cur, "result_output", -1);
-
-        ggml_build_forward_expand(gf, cur);
-
-        return gf;
-    }
-
    struct ggml_cgraph * build_refact() {
        struct ggml_cgraph * gf = ggml_new_graph_custom(ctx0, LLAMA_MAX_NODES, false);

@@ -10898,10 +10624,6 @@ static struct ggml_cgraph * llama_build_graph(
            {
                result = llm.build_starcoder();
            } break;
-        case LLM_ARCH_PERSIMMON:
-            {
-                result = llm.build_persimmon();
-            } break;
        case LLM_ARCH_REFACT:
            {
                result = llm.build_refact();
@@ -12776,9 +12498,14 @@ static std::vector<llama_vocab::id> llama_tokenize_internal(const llama_vocab &
                // tokenizer.encode('', add_special_tokens=True)  returns [1]
                // tokenizer.encode('', add_special_tokens=False) returns []

+                static const bool rtrim = true;  //TODO: as param
+                bool is_prev_special = false;
+                bool special_token_rtrim = false;
+
                if (add_special && vocab.special_add_bos != 0) {
                    GGML_ASSERT(vocab.special_bos_id != -1);
                    output.push_back(vocab.special_bos_id);
+                    is_prev_special = true;
                }

                for (const auto & fragment : fragment_buffer) {
@@ -12790,9 +12517,21 @@ static std::vector<llama_vocab::id> llama_tokenize_internal(const llama_vocab &
                        //  and passing 'add space prefix' as bool argument
                        //
                        auto raw_text = fragment.raw_text.substr(fragment.offset, fragment.length);
-                        if (&fragment == &fragment_buffer.front()) {
-                            if (vocab.add_space_prefix) {
-                                raw_text = " " + raw_text; // prefix with space if the first token is not special
+
+                        if (special_token_rtrim) {
+                            size_t num_whitespaces = 0;
+                            while (isspace(raw_text[num_whitespaces])) {
+                                num_whitespaces++;
+                            }
+                            if (num_whitespaces == raw_text.size()) {
+                                continue; // skip if all whitespaces
+                            }
+                            raw_text = raw_text.substr(num_whitespaces);
+                        }
+
+                        if (vocab.add_space_prefix) {
+                            if (!output.size() || is_prev_special) {  // prefix with space if first token
+                                raw_text = " " + raw_text;
                            }
                        }

@@ -12804,6 +12543,12 @@ static std::vector<llama_vocab::id> llama_tokenize_internal(const llama_vocab &
                        tokenizer.tokenize(raw_text, output);
                    } else { // if (fragment.type == FRAGMENT_BUFFER_VARIANT_TYPE_TOKEN)
                        output.push_back(fragment.token);
+                        is_prev_special = true;
+                        // phi-3 special tokens without rtrim, works fine for llama-spm too
+                        special_token_rtrim = rtrim
+                            && fragment.token != vocab.special_bos_id
+                            && fragment.token != vocab.special_unk_id
+                            && fragment.token != vocab.special_eos_id;
                    }
                }

@@ -15992,7 +15737,6 @@ enum llama_rope_type llama_rope_type(const struct llama_model * model) {
        case LLM_ARCH_FALCON:
        case LLM_ARCH_GROK:
        case LLM_ARCH_DBRX:
-        case LLM_ARCH_PERSIMMON:
        case LLM_ARCH_BERT:
        case LLM_ARCH_NOMIC_BERT:
        case LLM_ARCH_STABLELM:
@@ -9,4 +9,3 @@
 -r ./requirements/requirements-convert-hf-to-gguf.txt
 -r ./requirements/requirements-convert-hf-to-gguf-update.txt
 -r ./requirements/requirements-convert-llama-ggml-to-gguf.txt
-r ./requirements/requirements-convert-persimmon-to-gguf.txt
@@ -1,2 +0,0 @@
-r ./requirements-convert.txt
-torch~=2.1.1
@@ -153,11 +153,26 @@ def generator_custom_text_edge_cases() -> Iterator[str]:
        'Ⅵ-a',       # unicode_ranges_digit, {0x00002150, 0x0000218F} // Number Forms
        '\uFEFF//',   # unicode_ranges_control, 0xFEFF (BOM)
        'Cửa Việt',   # llama-3, ignore_merges = true
-        '<s>a',       # TODO: Phi-3 fail
+        '<s>a',       # Phi-3 fail
+        '<unk><|endoftext|><s>',  # Phi-3 fail
        'a\na',       # TODO: Bert fail
    ]


+def generator_random_special_tokens(tokenizer, iterations=100) -> Iterator[str]:
+    special_tokens = set(tokenizer.all_special_tokens)
+    special_tokens.update([" ", "\n", "\t", "-", "!", "one", "1", "<s>", "</s>"])
+    special_tokens = list(sorted(special_tokens))
+    rand = random.Random()
+    for m in range(iterations):
+        rand.seed(m)
+        words = rand.choices(special_tokens, k=500)
+        if tokenizer.add_bos_token:  # skip spam warning of double BOS
+            while words and words[0] == tokenizer.bos_token:
+                words.pop(0)
+        yield "".join(words)
+
+
 def generator_vocab_words(vocab: list[str]) -> Iterator[str]:
    """Brute force check all vocab words"""
    yield from vocab
@@ -278,25 +293,43 @@ def main(argv: list[str] = None):
    model = LibLlamaModel(LibLlama(), args.vocab_file, mparams=dict(vocab_only=True), cparams=dict(n_ctx=4096))
    tokenizer = AutoTokenizer.from_pretrained(args.dir_tokenizer)

-    def func_tokenize2(text: str):
-        return tokenizer.encode(text, add_special_tokens=False)
-
-    parse_special = all(len(func_tokenize2(t)) == 1 for t in tokenizer.all_special_tokens)
+    tokenizer.add_bos_token = getattr(tokenizer, "add_bos_token", True)
+    tokenizer.add_eos_token = getattr(tokenizer, "add_eos_token", False)

    def func_tokenize1(text: str):
-        return model.tokenize(text, add_special=False, parse_special=parse_special)
+        return model.tokenize(text, add_special=True, parse_special=True)
+
+    def func_tokenize2(text: str):
+        return tokenizer.encode(text, add_special_tokens=True)

    vocab = list(sorted(tokenizer.batch_decode(list(tokenizer.get_vocab().values()), skip_special_tokens=True)))
    test_compare_tokenizer(func_tokenize1, func_tokenize2, generator_custom_text())
    test_compare_tokenizer(func_tokenize1, func_tokenize2, generator_custom_text_edge_cases())
+    test_compare_tokenizer(func_tokenize1, func_tokenize2, generator_random_special_tokens(tokenizer, 10_000))
    test_compare_tokenizer(func_tokenize1, func_tokenize2, generator_vocab_words(vocab))
    test_compare_tokenizer(func_tokenize1, func_tokenize2, generator_random_chars(10_000))
    test_compare_tokenizer(func_tokenize1, func_tokenize2, generator_random_vocab_chars(vocab, 10_000))
-    test_compare_tokenizer(func_tokenize1, func_tokenize2, generator_random_vocab_words(vocab, 10_000))
+    test_compare_tokenizer(func_tokenize1, func_tokenize2, generator_random_vocab_words(vocab, 5_000))
    # test_compare_tokenizer(func_tokenize1, func_tokenize2, generator_random_bytes(10_000)) # FAIL

    model.free()


 if __name__ == "__main__":
-    main()
+    # main()
+
+    path_tokenizers = "./models/tokenizers/"
+    path_vocab_format = "./models/ggml-vocab-%s.gguf"
+
+    # import os
+    # tokenizers = os.listdir(path_tokenizers)
+    tokenizers = [
+        "llama-spm",   # SPM
+        "phi-3",       # SPM
+    ]
+
+    for tokenizer in tokenizers:
+        print("\n" + "=" * 50 + "\n" + tokenizer + "\n")  # noqa
+        vocab_file = path_vocab_format % tokenizer
+        dir_tokenizer = path_tokenizers + "/" + tokenizer
+        main([vocab_file, dir_tokenizer, "--verbose"])
Author	SHA1	Message	Date
Johannes Gäßler	d8ee902227	CUDA: deduplicate mmq code (#7397 )	2024-05-21 16:02:12 +02:00
jaime-m-p	d7e852c1bc	Tokenizer SPM fixes for phi-3 and llama-spm (bugfix) (#7425 ) * Update brute force test: add_special * Update brute force test: default values for add_bos_token and add_eos_token * Enable rtrim when pre-inserting BOS Co-authored-by: Georgi Gerganov <ggerganov@gmail.com> * Revert "server : fix test regexes"	2024-05-21 14:39:48 +02:00
jaime-m-p	917dc8cfa6	Tokenizer SPM fixes for phi-3 and llama-spm (#7375 ) * Update brute force test: special tokens * Fix added tokens - Try to read 'added_tokens.json'. - Try to read 'tokenizer_config.json'. - Try to read 'tokenizer.json'. * Fix special tokens rtrim Co-authored-by: Georgi Gerganov <ggerganov@gmail.com> * server : fix test regexes	2024-05-20 20:15:57 +02:00
Georgi Gerganov	fabf30b4c4	llama : remove Persimmon (#7408 ) * llama : remove Persimmon * requirements : remove	2024-05-21 02:35:28 +10:00
Johannes Gäßler	20385cebcc	perplexity: update README FP16 results [no ci] (#7413 )	2024-05-20 18:15:38 +02:00