pengjunjie/ragflow
1
0
Fork 1
mirror of https://github.com/infiniflow/ragflow.git synced 2025-12-08 20:42:30 +08:00
Code Issues Packages Projects Releases Wiki Activity

Fix: Disable reasoning on Gemini 2.5 Flash by default (#10477)

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

Gemini 2.5 Flash Models use reasoning by default. There is currently no
way to disable this behaviour. This leads to very long response times (>
1min). The default behaviour should be, that reasoning is disabled and
configurable

issue #10474 

### Type of change

- [X] Bug Fix (non-breaking change which fixes an issue)
This commit is contained in:
Günter Lukas
2025-10-11 04:22:51 +02:00
committed by GitHub
parent b5ddc7ca05
commit fee757eb41
1 changed files with 37 additions and 2 deletions
Download Patch File Download Diff File Expand all files Collapse all files

39
rag/llm/chat_model.py
View File

@ -1188,8 +1188,36 @@ class GoogleChat(Base):
del gen_conf[k]
return gen_conf
def _get_thinking_config(self, gen_conf):
"""Extract and create ThinkingConfig from gen_conf.
Default behavior for Vertex AI Generative Models: thinking_budget=0 (disabled)
unless explicitly specified by the user. This does not apply to Claude models.
Users can override by setting thinking_budget in gen_conf/llm_setting:
- 0: Disabled (default)
- 1-24576: Manual budget
- -1: Auto (model decides)
"""
# Claude models don't support ThinkingConfig
if "claude" in self.model_name:
gen_conf.pop("thinking_budget", None)
return None
# For Vertex AI Generative Models, default to thinking disabled
thinking_budget = gen_conf.pop("thinking_budget", 0)
if thinking_budget is not None:
try:
import vertexai.generative_models as glm # type: ignore
return glm.ThinkingConfig(thinking_budget=thinking_budget)
except Exception:
pass
return None
def _chat(self, history, gen_conf={}, **kwargs):
system = history[0]["content"] if history and history[0]["role"] == "system" else ""
thinking_config = self._get_thinking_config(gen_conf)
gen_conf = self._clean_conf(gen_conf)
if "claude" in self.model_name:
response = self.client.messages.create(
@ -1223,7 +1251,10 @@ class GoogleChat(Base):
}
]
response = self.client.generate_content(hist, generation_config=gen_conf)
if thinking_config:
response = self.client.generate_content(hist, generation_config=gen_conf, thinking_config=thinking_config)
else:
response = self.client.generate_content(hist, generation_config=gen_conf)
ans = response.text
return ans, response.usage_metadata.total_token_count
@ -1255,6 +1286,7 @@ class GoogleChat(Base):
response = None
total_tokens = 0
self.client._system_instruction = system
thinking_config = self._get_thinking_config(gen_conf)
if "max_tokens" in gen_conf:
gen_conf["max_output_tokens"] = gen_conf["max_tokens"]
del gen_conf["max_tokens"]
@ -1272,7 +1304,10 @@ class GoogleChat(Base):
]
ans = ""
try:
response = self.client.generate_content(history, generation_config=gen_conf, stream=True)
if thinking_config:
response = self.client.generate_content(history, generation_config=gen_conf, thinking_config=thinking_config, stream=True)
else:
response = self.client.generate_content(history, generation_config=gen_conf, stream=True)
for resp in response:
ans = resp.text
total_tokens += num_tokens_from_string(ans)