llama : fix codeshell support (#8599)

* llama : fix codeshell support

* llama : move codeshell after smollm below to respect the enum order

Author: hankeke303

convert_hf_to_gguf.py

@@ -594,6 +594,9 @@ def get_vocab_base_pre(self, tokenizer) -> str:
         if chkhsh == "b53802fb28e26d645c3a310b34bfe07da813026ec7c7716883404d5e0f8b1901":
             # ref: https://huggingface.co/core42/jais-13b
             res = "jais"
+        if chkhsh == "7b3e7548e4308f52a76e8229e4e6cc831195d0d1df43aed21ac6c93da05fec5f":
+            # ref: https://huggingface.co/WisdomShell/CodeShell-7B
+            res = "codeshell"
         if chkhsh == "63b97e4253352e6f357cc59ea5b583e3a680eaeaf2632188c2b952de2588485e":
             # ref: https://huggingface.co/mistralai/Mistral-Nemo-Base-2407
             res = "tekken"

convert_hf_to_gguf_update.py

@@ -91,6 +91,7 @@ class TOKENIZER_TYPE(IntEnum):
     {"name": "gemma-2",        "tokt": TOKENIZER_TYPE.SPM, "repo": "https://huggingface.co/google/gemma-2-9b", },
     {"name": "jais",           "tokt": TOKENIZER_TYPE.BPE, "repo": "https://huggingface.co/core42/jais-13b", },
     {"name": "t5",             "tokt": TOKENIZER_TYPE.UGM, "repo": "https://huggingface.co/google-t5/t5-small", },
+    {"name": "codeshell",      "tokt": TOKENIZER_TYPE.BPE, "repo": "https://huggingface.co/WisdomShell/CodeShell-7B", },
     {"name": "tekken",         "tokt": TOKENIZER_TYPE.BPE, "repo": "https://huggingface.co/mistralai/Mistral-Nemo-Base-2407", },
     {"name": "smollm",         "tokt": TOKENIZER_TYPE.BPE, "repo": "https://huggingface.co/HuggingFaceTB/SmolLM-135M", },
 ]

include/llama.h

@@ -94,6 +94,7 @@ extern "C" {
         LLAMA_VOCAB_PRE_TYPE_JAIS           = 19,
         LLAMA_VOCAB_PRE_TYPE_TEKKEN         = 20,
         LLAMA_VOCAB_PRE_TYPE_SMOLLM         = 21,
+        LLAMA_VOCAB_PRE_TYPE_CODESHELL      = 22,
     };
 
     // note: these values should be synchronized with ggml_rope

src/llama.cpp

@@ -5007,7 +5007,7 @@ static void llm_load_hparams(
             {
                 ml.get_key(LLM_KV_ATTENTION_LAYERNORM_EPS, hparams.f_norm_eps);
                 switch (hparams.n_layer) {
-                    case 42: model.type = e_model::MODEL_SMALL; break;
+                    case 42: model.type = e_model::MODEL_7B; break;
                     default: model.type = e_model::MODEL_UNKNOWN;
                 }
             } break;
@@ -5525,6 +5525,9 @@ static void llm_load_vocab(
                 tokenizer_pre == "smollm") {
                 vocab.type_pre = LLAMA_VOCAB_PRE_TYPE_SMOLLM;
                 vocab.tokenizer_clean_spaces = false;
+            } else if (
+                tokenizer_pre == "codeshell") {
+                vocab.type_pre = LLAMA_VOCAB_PRE_TYPE_CODESHELL;
             } else {
                 throw std::runtime_error(format("unknown pre-tokenizer type: '%s'", tokenizer_pre.c_str()));
             }
@@ -15548,6 +15551,7 @@ struct llm_tokenizer_bpe {
             case LLAMA_VOCAB_PRE_TYPE_REFACT:
             case LLAMA_VOCAB_PRE_TYPE_COMMAND_R:
             case LLAMA_VOCAB_PRE_TYPE_SMOLLM:
+            case LLAMA_VOCAB_PRE_TYPE_CODESHELL:
                 regex_exprs = {
                     "\\p{N}",
                     "'s|'t|'re|'ve|'m|'ll|'d| ?\\p{L}+| ?\\p{N}+| ?[^\\s\\p{L}\\p{N}]+|\\s+(?!\\S)",
@@ -19447,7 +19451,6 @@ enum llama_rope_type llama_rope_type(const struct llama_model * model) {
         case LLM_ARCH_BAICHUAN:
         case LLM_ARCH_STARCODER:
         case LLM_ARCH_PLAMO:
-        case LLM_ARCH_CODESHELL:
         case LLM_ARCH_ORION:
         case LLM_ARCH_INTERNLM2:
         case LLM_ARCH_MINICPM:
@@ -19477,6 +19480,7 @@ enum llama_rope_type llama_rope_type(const struct llama_model * model) {
         case LLM_ARCH_STARCODER2:
         case LLM_ARCH_OPENELM:
         case LLM_ARCH_GPTNEOX:
+        case LLM_ARCH_CODESHELL:
             return LLAMA_ROPE_TYPE_NEOX;
 
         // all model arches should be listed explicitly here