ceos/unsloth
1
0
Fork 0
Code Issues Pull requests Projects Releases Packages Wiki Activity Actions
unsloth/studio/backend/assets/configs/inference_defaults.json
Daniel Han 70021b1f6a Studio: family-default chat_template_kwargs for gpt-oss + Nemotron 
The per-family inference defaults registry (inference_defaults.json)
exposes scalar sampling fields only (temperature, top_p, top_k, min_p,
presence_penalty). Modern reasoning models also want template-level
kwargs: gpt-oss takes reasoning_effort (low|medium|high); Nemotron-3
takes enable_thinking. Today those reach llama-server only when an
OpenAI SDK client sets extra_body.chat_template_kwargs explicitly, so
external clients hitting /v1/chat/completions never benefit from the
recommended defaults.

This patch:

* Adds an optional ``chat_template_kwargs`` block per family in
  inference_defaults.json. Bootstrap rows:
    gpt-oss   -> reasoning_effort: medium
    nemotron  -> enable_thinking: true
* Extends load_inference_config to surface chat_template_kwargs in
  the returned dict (None when unset).
* Extends _build_openai_passthrough_body to merge family defaults
  (lookup via load_inference_config(model_identifier)) with per-request
  overrides. Priority order, base to top:
    family_defaults -> payload.enable_thinking -> payload.reasoning_effort
                    -> extra_body.chat_template_kwargs
* Bumps two family temperatures to match upstream HF model cards:
    devstral  0.7  -> 0.15  (Devstral-Small-2-Instruct card)
    ministral 0.15 -> 0.05  (Ministral-3 card: "below 0.1 for production")
2026-05-22 14:35:37 +00:00

404 lines
8.5 KiB
JSON
Raw Permalink Blame History

{
"_comment": "Per-model-family inference parameter defaults. Sources: (1) Ollama params blobs, (2) Existing Unsloth Studio YAML configs. Patterns ordered longest-match-first.",
"families": {
"qwen3.6": {
"temperature": 0.7,
"top_p": 0.8,
"top_k": 20,
"min_p": 0.0,
"repetition_penalty": 1.0,
"presence_penalty": 1.5
},
"qwen3.5": {
"temperature": 0.7,
"top_p": 0.8,
"top_k": 20,
"min_p": 0.0,
"repetition_penalty": 1.0,
"presence_penalty": 1.5
},
"qwen3-coder": {
"temperature": 0.7,
"top_p": 0.8,
"top_k": 20,
"min_p": 0.0,
"repetition_penalty": 1.0
},
"qwen3-next": {
"temperature": 0.7,
"top_p": 0.8,
"top_k": 20,
"min_p": 0.0,
"repetition_penalty": 1.0
},
"qwen3-vl": {
"temperature": 0.7,
"top_p": 0.8,
"top_k": 20,
"min_p": 0.0,
"repetition_penalty": 1.0
},
"qwen3": {
"temperature": 0.6,
"top_p": 0.95,
"top_k": 20,
"min_p": 0.0,
"repetition_penalty": 1.0
},
"qwen2.5-coder": {
"temperature": 1.5,
"top_p": 0.95,
"top_k": -1,
"min_p": 0.1,
"repetition_penalty": 1.0
},
"qwen2.5-vl": {
"temperature": 1.5,
"top_p": 0.95,
"top_k": -1,
"min_p": 0.1,
"repetition_penalty": 1.0
},
"qwen2.5-omni": {
"temperature": 0.7,
"top_p": 0.8,
"top_k": 20,
"min_p": 0.0,
"repetition_penalty": 1.0
},
"qwen2.5-math": {
"temperature": 0.7,
"top_p": 0.8,
"top_k": 20,
"min_p": 0.0,
"repetition_penalty": 1.0
},
"qwen2.5": {
"temperature": 0.7,
"top_p": 0.8,
"top_k": 20,
"min_p": 0.0,
"repetition_penalty": 1.0
},
"qwen2-vl": {
"temperature": 1.5,
"top_p": 0.95,
"top_k": -1,
"min_p": 0.1,
"repetition_penalty": 1.0
},
"qwen2": {
"temperature": 0.7,
"top_p": 0.8,
"top_k": 20,
"min_p": 0.0,
"repetition_penalty": 1.0
},
"qwq": {
"temperature": 0.6,
"top_p": 0.95,
"top_k": 40,
"min_p": 0.0,
"repetition_penalty": 1.0
},
"gemma-4": {
"temperature": 1.0,
"top_p": 0.95,
"top_k": 64,
"min_p": 0.0,
"repetition_penalty": 1.0,
"presence_penalty": 0.0
},
"gemma-3n": {
"temperature": 1.0,
"top_p": 0.95,
"top_k": 64,
"min_p": 0.0,
"repetition_penalty": 1.0
},
"gemma-3": {
"temperature": 1.0,
"top_p": 0.95,
"top_k": 64,
"min_p": 0.0,
"repetition_penalty": 1.0
},
"medgemma": {
"temperature": 1.0,
"top_p": 0.95,
"top_k": 64,
"min_p": 0.0,
"repetition_penalty": 1.0
},
"gemma-2": {
"temperature": 1.0,
"top_p": 0.95,
"top_k": 64,
"min_p": 0.0,
"repetition_penalty": 1.0
},
"llama-4": {
"temperature": 1.0,
"top_p": 0.9,
"top_k": -1,
"min_p": 0.01,
"repetition_penalty": 1.0
},
"llama-3.3": {
"temperature": 1.5,
"top_p": 0.95,
"top_k": -1,
"min_p": 0.1,
"repetition_penalty": 1.0
},
"llama-3.2": {
"temperature": 1.5,
"top_p": 0.95,
"top_k": -1,
"min_p": 0.1,
"repetition_penalty": 1.0
},
"llama-3.1": {
"temperature": 1.5,
"top_p": 0.95,
"top_k": -1,
"min_p": 0.1,
"repetition_penalty": 1.0
},
"llama-3": {
"temperature": 1.5,
"top_p": 0.95,
"top_k": -1,
"min_p": 0.1,
"repetition_penalty": 1.0
},
"phi-4": {
"temperature": 0.8,
"top_p": 0.95,
"top_k": -1,
"min_p": 0.0,
"repetition_penalty": 1.0
},
"phi-3": {
"temperature": 0.7,
"top_p": 0.9,
"top_k": -1,
"min_p": 0.01,
"repetition_penalty": 1.0
},
"mistral-nemo": {
"temperature": 0.7,
"top_p": 0.95,
"top_k": -1,
"min_p": 0.01,
"repetition_penalty": 1.0
},
"mistral-small": {
"temperature": 0.15,
"top_p": 0.95,
"top_k": -1,
"min_p": 0.01,
"repetition_penalty": 1.0
},
"mistral-large": {
"temperature": 0.7,
"top_p": 0.95,
"top_k": -1,
"min_p": 0.01,
"repetition_penalty": 1.0
},
"magistral": {
"temperature": 0.7,
"top_p": 0.95,
"top_k": -1,
"min_p": 0.01,
"repetition_penalty": 1.0
},
"ministral": {
"temperature": 0.05,
"top_p": 0.95,
"top_k": -1,
"min_p": 0.01,
"repetition_penalty": 1.0
},
"devstral": {
"temperature": 0.15,
"top_p": 0.95,
"top_k": -1,
"min_p": 0.01,
"repetition_penalty": 1.0
},
"pixtral": {
"temperature": 1.5,
"top_p": 0.95,
"top_k": -1,
"min_p": 0.1,
"repetition_penalty": 1.0
},
"deepseek-r1": {
"temperature": 0.6,
"top_p": 0.95,
"top_k": -1,
"min_p": 0.01,
"repetition_penalty": 1.0
},
"deepseek-v3": {
"temperature": 0.6,
"top_p": 0.95,
"top_k": -1,
"min_p": 0.01,
"repetition_penalty": 1.0
},
"deepseek-ocr": {
"temperature": 0.0,
"top_p": 0.95,
"top_k": -1,
"min_p": 0.01,
"repetition_penalty": 1.0
},
"glm-5": {
"temperature": 1.0,
"top_p": 0.95,
"top_k": -1,
"min_p": 0.01,
"repetition_penalty": 1.0
},
"glm-4": {
"temperature": 1.0,
"top_p": 0.95,
"top_k": -1,
"min_p": 0.01,
"repetition_penalty": 1.0
},
"nemotron": {
"temperature": 1.0,
"top_p": 1.0,
"top_k": -1,
"min_p": 0.01,
"repetition_penalty": 1.0,
"chat_template_kwargs": {
"enable_thinking": true
}
},
"minimax-m2.5": {
"temperature": 1.0,
"top_p": 0.95,
"top_k": 40,
"min_p": 0.01,
"repetition_penalty": 1.0
},
"minimax": {
"temperature": 1.0,
"top_p": 0.95,
"top_k": 40,
"min_p": 0.01,
"repetition_penalty": 1.0
},
"gpt-oss": {
"temperature": 1.0,
"top_p": 1.0,
"top_k": 0,
"min_p": 0.01,
"repetition_penalty": 1.0,
"chat_template_kwargs": {
"reasoning_effort": "medium"
}
},
"granite-4": {
"temperature": 0.0,
"top_p": 1.0,
"top_k": 0,
"min_p": 0.01,
"repetition_penalty": 1.0
},
"kimi-k2": {
"temperature": 0.6,
"top_p": 0.95,
"top_k": -1,
"min_p": 0.01,
"repetition_penalty": 1.0
},
"kimi": {
"temperature": 0.6,
"top_p": 0.95,
"top_k": -1,
"min_p": 0.01,
"repetition_penalty": 1.0
},
"lfm2": {
"temperature": 0.1,
"top_p": 0.1,
"top_k": 50,
"min_p": 0.15,
"repetition_penalty": 1.05
},
"smollm": {
"temperature": 0.7,
"top_p": 0.95,
"top_k": -1,
"min_p": 0.01,
"repetition_penalty": 1.0
},
"olmo": {
"temperature": 0.7,
"top_p": 0.95,
"top_k": -1,
"min_p": 0.01,
"repetition_penalty": 1.0
},
"falcon": {
"temperature": 0.7,
"top_p": 0.95,
"top_k": -1,
"min_p": 0.01,
"repetition_penalty": 1.0
},
"ernie": {
"temperature": 0.7,
"top_p": 0.95,
"top_k": -1,
"min_p": 0.01,
"repetition_penalty": 1.0
},
"seed": {
"temperature": 0.7,
"top_p": 0.95,
"top_k": -1,
"min_p": 0.01,
"repetition_penalty": 1.0
},
"grok": {
"temperature": 1.0,
"top_p": 0.95,
"top_k": -1,
"min_p": 0.01,
"repetition_penalty": 1.0
},
"mimo": {
"temperature": 0.7,
"top_p": 0.95,
"top_k": -1,
"min_p": 0.01,
"repetition_penalty": 1.0
}
},
"patterns": [
"qwen3.6", "qwen3.5",
"qwen3-coder", "qwen3-next", "qwen3-vl", "qwen3",
"qwen2.5-coder", "qwen2.5-vl", "qwen2.5-omni", "qwen2.5-math", "qwen2.5",
"qwen2-vl", "qwen2",
"qwq",
"gemma-4", "gemma-3n", "gemma-3", "medgemma", "gemma-2",
"llama-4", "llama-3.3", "llama-3.2", "llama-3.1", "llama-3",
"phi-4", "phi-3",
"mistral-nemo", "mistral-small", "mistral-large", "magistral", "ministral",
"devstral", "pixtral",
"deepseek-r1", "deepseek-v3", "deepseek-ocr",
"glm-5", "glm-4",
"nemotron",
"minimax-m2.5", "minimax",
"gpt-oss", "granite-4",
"kimi-k2", "kimi",
"lfm2", "smollm", "olmo", "falcon", "ernie", "seed", "grok", "mimo"
]
}
Reference in a new issue View git blame Copy permalink