pengjunjie/ragflow
1
0
Fork 1
mirror of https://github.com/infiniflow/ragflow.git synced 2026-01-28 14:16:34 +08:00
Code Issues Packages Projects Releases Wiki Activity

Feat: add kimi-k2.5 (#12852)

Browse Source 
### What problem does this PR solve?

Add kimi-k2.5

### Type of change

- [x] New Feature (non-breaking change which adds functionality)
This commit is contained in:
Yongteng Lei
2026-01-28 12:41:20 +08:00
committed by GitHub
parent 3a8c848af5
commit b57c82b122
2 changed files with 27 additions and 3 deletions
Download Patch File Download Diff File Expand all files Collapse all files

9
conf/llm_factories.json
View File

@ -994,6 +994,13 @@
"model_type": "chat",
"is_tools": true
},
{
"llm_name": "kimi-k2.5",
"tags": "LLM,CHAT,256k",
"max_tokens": 256000,
"model_type": "chat",
"is_tools": true
},
{
"llm_name": "kimi-latest",
"tags": "LLM,CHAT,8k,32k,128k",
@ -5578,4 +5585,4 @@
]
}
]
}
}

21
rag/llm/chat_model.py
View File

@ -1180,13 +1180,30 @@ class LiteLLMBase(ABC):
return LLMErrorCode.ERROR_GENERIC
def _clean_conf(self, gen_conf):
gen_conf = deepcopy(gen_conf) if gen_conf else {}
if self.provider == SupportedLiteLLMProvider.HunYuan:
unsupported = ["presence_penalty", "frequency_penalty"]
for key in unsupported:
gen_conf.pop(key, None)
if "max_tokens" in gen_conf:
del gen_conf["max_tokens"]
elif "kimi-k2.5" in self.model_name.lower():
reasoning = gen_conf.pop("reasoning", None) # will never get one here, handle this later
thinking = {"type": "enabled"} # enable thinking by default
if reasoning is not None:
thinking = {"type": "enabled"} if reasoning else {"type": "disabled"}
elif not isinstance(thinking, dict) or thinking.get("type") not in {"enabled", "disabled"}:
thinking = {"type": "disabled"}
gen_conf["thinking"] = thinking
thinking_enabled = thinking.get("type") == "enabled"
gen_conf["temperature"] = 1.0 if thinking_enabled else 0.6
gen_conf["top_p"] = 0.95
gen_conf["n"] = 1
gen_conf["presence_penalty"] = 0.0
gen_conf["frequency_penalty"] = 0.0
gen_conf.pop("max_tokens", None)
return gen_conf
async def async_chat(self, system, history, gen_conf, **kwargs):