root/llama.cpp
1
0
Fork 0
mirror of https://github.com/ggerganov/llama.cpp.git synced 2024-12-25 02:44:36 +00:00
Code Issues Actions 9 Packages Projects Releases Wiki Activity

Added support for Viking pre-tokenizer (#8135)

Browse Source 
Co-authored-by: kustaaya <kustaaya@protonmail.com>
This commit is contained in:
kustaaya 2024-06-27 11:58:54 +03:00 committed by GitHub
parent 911e35bb8b
commit f675b20a3b
No known key found for this signature in database
GPG Key ID: B5690EEEBB952194
4 changed files with 14 additions and 0 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

1
convert-hf-to-gguf-update.py
View File

@ -85,6 +85,7 @@ models = [
{"name": "smaug-bpe", "tokt": TOKENIZER_TYPE.BPE, "repo": "https://huggingface.co/abacusai/Smaug-Llama-3-70B-Instruct", },
{"name": "poro-chat", "tokt": TOKENIZER_TYPE.BPE, "repo": "https://huggingface.co/LumiOpen/Poro-34B-chat", },
{"name": "jina-v2-code", "tokt": TOKENIZER_TYPE.BPE, "repo": "https://huggingface.co/jinaai/jina-embeddings-v2-base-code", },
{"name": "viking", "tokt": TOKENIZER_TYPE.BPE, "repo": "https://huggingface.co/LumiOpen/Viking-7B", }, # Also used for Viking 13B and 33B
]

3
convert-hf-to-gguf.py
View File

@ -487,6 +487,9 @@ class Model:
if chkhsh == "7967bfa498ade6b757b064f31e964dddbb80f8f9a4d68d4ba7998fcf281c531a":
# ref: https://huggingface.co/jinaai/jina-embeddings-v2-base-code
res = "jina-v2-code"
if chkhsh == "7fc505bd3104ca1083b150b17d088b59534ede9bde81f0dd2090967d7fe52cee":
# ref: https://huggingface.co/LumiOpen/Viking-7B
res = "viking"
if res is None:
logger.warning("\n")

1
include/llama.h
View File

@ -88,6 +88,7 @@ extern "C" {
LLAMA_VOCAB_PRE_TYPE_DBRX = 13,
LLAMA_VOCAB_PRE_TYPE_SMAUG = 14,
LLAMA_VOCAB_PRE_TYPE_PORO = 15,
LLAMA_VOCAB_PRE_TYPE_VIKING = 16,
};
// note: these values should be synchronized with ggml_rope

9
src/llama.cpp
View File

@ -5067,6 +5067,9 @@ static void llm_load_vocab(
} else if (
tokenizer_pre == "poro-chat") {
vocab.type_pre = LLAMA_VOCAB_PRE_TYPE_PORO;
} else if (
tokenizer_pre == "viking") {
vocab.type_pre = LLAMA_VOCAB_PRE_TYPE_VIKING;
} else {
throw std::runtime_error(format("unknown pre-tokenizer type: '%s'", tokenizer_pre.c_str()));
}
@ -13703,6 +13706,12 @@ struct llm_tokenizer_bpe {
" ?[^(\\s|.,!?…。，、।۔،)]+",
};
break;
case LLAMA_VOCAB_PRE_TYPE_VIKING:
regex_exprs = {
"\\p{N}",
" ?[^(\\s|.,!?…。，、।۔،)]+",
};
break;
default:
// default regex for BPE tokenization pre-processing
regex_exprs = {