ceos/unsloth
1
0
Fork 0
Code Issues Pull requests Projects Releases Packages Wiki Activity Actions

fix: use max_completion_tokens for OpenAI provider

Browse source 
Newer OpenAI models (gpt-4o, gpt-5.x) reject the max_tokens param
and require max_completion_tokens instead. Other providers still use
max_tokens.
This commit is contained in:
Roland Tannous 2026-04-08 11:16:44 +04:00
commit 706eaeae36
1 changed files with 9 additions and 2 deletions
Download patch file Download diff file Expand all files Collapse all files

11
studio/backend/core/inference/external_provider.py
View file

@ -91,7 +91,11 @@ class ExternalProviderClient:
"presence_penalty": presence_penalty,
}
if max_tokens is not None:
body["max_tokens"] = max_tokens
# OpenAI newer models (gpt-4o, gpt-5.x) reject max_tokens
if self.provider_type == "openai":
body["max_completion_tokens"] = max_tokens
else:
body["max_tokens"] = max_tokens
url = f"{self.base_url}/chat/completions"
logger.info(
@ -406,7 +410,10 @@ class ExternalProviderClient:
"presence_penalty": presence_penalty,
}
if max_tokens is not None:
body["max_tokens"] = max_tokens
if self.provider_type == "openai":
body["max_completion_tokens"] = max_tokens
else:
body["max_tokens"] = max_tokens
response = await _http_client.post(
f"{self.base_url}/chat/completions",