1h ago
Please support temperature for deepseek-flash
Currently any request with "temperature" to deepseek-flash returns HTTP 400 Bad Request:
{"error":{"message":"Omit temperature for this model.",
"type":"invalid_request_error","param":"temperature"}}
Please accept temperature (0-2) for deepseek-flash and actually apply it.
api.deepseek.com accepts temperature for deepseek-flash (verified with 0
and 1.5), and the endpoint is advertised as OpenAI Chat Completions
compatible, so clients that send temperature (or need temperature:0 for
reproducible evals) are currently rejected.
Keep the hard error for anything you genuinely can't support; don't
silently ignore the field. But DeepSeek does support temperature (0-2)
Pending