root/llama.cpp
1
0
Fork 0
mirror of https://github.com/ggerganov/llama.cpp.git synced 2024-12-24 10:24:35 +00:00
Code Issues Actions 5 Packages Projects Releases Wiki Activity

llama : fix codeshell support (#8599)

Browse Source 
* llama : fix codeshell support

* llama : move codeshell after smollm below to respect the enum order
This commit is contained in:
Keke Han 2024-07-23 00:43:43 +08:00 committed by GitHub
parent d94c6e0ccb
commit 081fe431aa
No known key found for this signature in database
GPG Key ID: B5690EEEBB952194
4 changed files with 11 additions and 2 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

3
convert_hf_to_gguf.py
View File

@ -594,6 +594,9 @@ class Model:
if chkhsh == "b53802fb28e26d645c3a310b34bfe07da813026ec7c7716883404d5e0f8b1901": if chkhsh == "b53802fb28e26d645c3a310b34bfe07da813026ec7c7716883404d5e0f8b1901":
# ref: https://huggingface.co/core42/jais-13b # ref: https://huggingface.co/core42/jais-13b
res = "jais" res = "jais"
if chkhsh == "7b3e7548e4308f52a76e8229e4e6cc831195d0d1df43aed21ac6c93da05fec5f":
# ref: https://huggingface.co/WisdomShell/CodeShell-7B
res = "codeshell"
if chkhsh == "63b97e4253352e6f357cc59ea5b583e3a680eaeaf2632188c2b952de2588485e": if chkhsh == "63b97e4253352e6f357cc59ea5b583e3a680eaeaf2632188c2b952de2588485e":
# ref: https://huggingface.co/mistralai/Mistral-Nemo-Base-2407 # ref: https://huggingface.co/mistralai/Mistral-Nemo-Base-2407
res = "tekken" res = "tekken"

1
convert_hf_to_gguf_update.py
View File

@ -91,6 +91,7 @@ models = [
{"name": "gemma-2", "tokt": TOKENIZER_TYPE.SPM, "repo": "https://huggingface.co/google/gemma-2-9b", }, {"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": "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": "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": "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", }, {"name": "smollm", "tokt": TOKENIZER_TYPE.BPE, "repo": "https://huggingface.co/HuggingFaceTB/SmolLM-135M", },
] ]

1
include/llama.h
View File

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

8
src/llama.cpp
View File

@ -5007,7 +5007,7 @@ static void llm_load_hparams(
{ {
ml.get_key(LLM_KV_ATTENTION_LAYERNORM_EPS, hparams.f_norm_eps); ml.get_key(LLM_KV_ATTENTION_LAYERNORM_EPS, hparams.f_norm_eps);
switch (hparams.n_layer) { 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; default: model.type = e_model::MODEL_UNKNOWN;
} }
} break; } break;
@ -5525,6 +5525,9 @@ static void llm_load_vocab(
tokenizer_pre == "smollm") { tokenizer_pre == "smollm") {
vocab.type_pre = LLAMA_VOCAB_PRE_TYPE_SMOLLM; vocab.type_pre = LLAMA_VOCAB_PRE_TYPE_SMOLLM;
vocab.tokenizer_clean_spaces = false; vocab.tokenizer_clean_spaces = false;
} else if (
tokenizer_pre == "codeshell") {
vocab.type_pre = LLAMA_VOCAB_PRE_TYPE_CODESHELL;
} else { } else {
throw std::runtime_error(format("unknown pre-tokenizer type: '%s'", tokenizer_pre.c_str())); 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_REFACT:
case LLAMA_VOCAB_PRE_TYPE_COMMAND_R: case LLAMA_VOCAB_PRE_TYPE_COMMAND_R:
case LLAMA_VOCAB_PRE_TYPE_SMOLLM: case LLAMA_VOCAB_PRE_TYPE_SMOLLM:
case LLAMA_VOCAB_PRE_TYPE_CODESHELL:
regex_exprs = { regex_exprs = {
"\\p{N}", "\\p{N}",
"'s|'t|'re|'ve|'m|'ll|'d| ?\\p{L}+| ?\\p{N}+| ?[^\\s\\p{L}\\p{N}]+|\\s+(?!\\S)", "'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_BAICHUAN:
case LLM_ARCH_STARCODER: case LLM_ARCH_STARCODER:
case LLM_ARCH_PLAMO: case LLM_ARCH_PLAMO:
case LLM_ARCH_CODESHELL:
case LLM_ARCH_ORION: case LLM_ARCH_ORION:
case LLM_ARCH_INTERNLM2: case LLM_ARCH_INTERNLM2:
case LLM_ARCH_MINICPM: 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_STARCODER2:
case LLM_ARCH_OPENELM: case LLM_ARCH_OPENELM:
case LLM_ARCH_GPTNEOX: case LLM_ARCH_GPTNEOX:
case LLM_ARCH_CODESHELL:
return LLAMA_ROPE_TYPE_NEOX; return LLAMA_ROPE_TYPE_NEOX;
// all model arches should be listed explicitly here // all model arches should be listed explicitly here