scriptoriumadmin/litellm
1
0
Fork 0
mirror of https://github.com/BerriAI/litellm.git synced 2026-09-29 01:42:19 +00:00
Code Issues Projects Releases Packages Wiki Activity Actions 79

fix(deepseek): pass reasoning_effort through to the API

Browse source 
DeepSeekChatConfig.map_openai_params mapped reasoning_effort to a binary
thinking {type: enabled/disabled} toggle and discarded the effort value, so
DeepSeek always ran default high effort. High-effort reasoning can run away
and exhaust max_tokens during reasoning, returning empty content. Preserve
the reasoning_effort value into the request body when thinking is enabled so
callers can bound effort (e.g. reasoning_effort="low"). Backward-compatible.
See https://api-docs.deepseek.com/guides/thinking_mode
This commit is contained in:
Hazelhof 2026-08-29 17:04:08 +10:00
parent 4d7144160a
commit 00a1c311e3
1 changed files with 15 additions and 0 deletions
Show stats Download patch file Download diff file Expand all files Collapse all files

15
litellm/llms/deepseek/chat/transformation.py
View file

@ -51,13 +51,28 @@ class DeepSeekChatConfig(OpenAIGPTConfig):
thinking_value: Final = optional_params.pop("thinking", None)
reasoning_effort: Final = optional_params.pop("reasoning_effort", None)
thinking_enabled = False
# Handle thinking parameter - accept both enabled and disabled, ignore budget_tokens
if isinstance(thinking_value, dict) and thinking_value.get("type") in ("enabled", "disabled"):
optional_params["thinking"] = {"type": thinking_value["type"]}
thinking_enabled = thinking_value["type"] == "enabled"
# Otherwise fall back to reasoning_effort: "none" disables, anything else enables
elif reasoning_effort is not None:
optional_params["thinking"] = {"type": "disabled" if reasoning_effort == "none" else "enabled"}
thinking_enabled = reasoning_effort != "none"
else:
thinking_enabled = True # DeepSeek thinking mode is on by default
# reasoning_effort is a real DeepSeek top-level param (default "high"; valid
# values low/medium/high/xhigh/max). The branch above only used it as an on/off
# toggle and DROPPED the value, so DeepSeek always ran default HIGH effort — which
# can run away, exhausting max_tokens during reasoning and returning empty content
# (finish_reason="length", content:null). Preserve the value into the request body
# when thinking is on so callers can bound effort (e.g. reasoning_effort="low").
# See https://api-docs.deepseek.com/guides/thinking_mode
if thinking_enabled and reasoning_effort is not None:
optional_params["reasoning_effort"] = reasoning_effort
return optional_params