Ported kimi-k2 support from llama.cpp (#609)

Original patch by @gabriellarson:
https://github.com/ggml-org/llama.cpp/pull/14654

Co-authored-by: anikifoss <anikifoss>
This commit is contained in:
Aleksey Nikiforov
2025-07-14 12:43:52 -04:00
committed by GitHub
parent 255c22046b
commit f5353047ef
5 changed files with 223 additions and 1 deletions

View File

@@ -464,6 +464,13 @@ struct llm_tokenizer_bpe {
"[^\\r\\n\\p{L}\\p{N}]?((?=[\\p{L}])([^a-z]))*((?=[\\p{L}])([^A-Z]))+(?:'[sS]|'[tT]|'[rR][eE]|'[vV][eE]|'[mM]|'[lL][lL]|'[dD])?|[^\\r\\n\\p{L}\\p{N}]?((?=[\\p{L}])([^a-z]))+((?=[\\p{L}])([^A-Z]))*(?:'[sS]|'[tT]|'[rR][eE]|'[vV][eE]|'[mM]|'[lL][lL]|'[dD])?|\\p{N}{1,3}| ?[^\\s\\p{L}\\p{N}]+[\\r\\n/]*|\\s*[\\r\\n]+|\\s+(?!\\S)|\\s+",
};
break;
case LLAMA_VOCAB_PRE_TYPE_KIMI_K2:
regex_exprs = {
// K2 trigger pattern - this will activate the custom K2 handler in unicode.cpp
// The custom handler implements all K2 patterns with proper Han character exclusion
"\\p{Han}+",
};
break;
case LLAMA_VOCAB_PRE_TYPE_SUPERBPE:
regex_exprs = {
"\\p{N}+",