sgl-project/sglang · error · ValueError

Invalid reasoning effort {reasoning_effort!r} for profile {r

Error message

Invalid reasoning effort {reasoning_effort!r} for profile {reasoning_effort_profile!r}; expected one of {list(effort_prompts)}

What it means

The reasoning_effort value is not valid for the chosen profile; each profile in REASONING_EFFORT_PROFILES defines its own effort keys (e.g. low/medium/high) and this combination failed.

Source

Thrown at python/sglang/srt/entrypoints/openai/encoding_dsv4.py:323

    tool_calls = msg.get("tool_calls")
    reasoning_content = msg.get("reasoning_content")
    wo_eos = msg.get("wo_eos", False)

    if tools:
        tools = tools_from_openai_format(tools)
    if tool_calls:
        tool_calls = tool_calls_from_openai_format(tool_calls)

    if reasoning_effort_profile not in REASONING_EFFORT_PROFILES:
        raise ValueError(
            f"Invalid reasoning effort profile: {reasoning_effort_profile!r}; "
            f"expected one of {list(REASONING_EFFORT_PROFILES)}"
        )
    effort_prompts = REASONING_EFFORT_PROFILES[reasoning_effort_profile]
    if reasoning_effort is None:
        reasoning_effort = "low" if reasoning_effort_profile == "official" else "high"
    if reasoning_effort not in effort_prompts:
        raise ValueError(
            f"Invalid reasoning effort {reasoning_effort!r} for profile "
            f"{reasoning_effort_profile!r}; expected one of {list(effort_prompts)}"
        )
    if index == 0 and thinking_mode == "thinking":
        prompt += effort_prompts[reasoning_effort]

    if role == "system":
        prompt += system_msg_template.format(content=content or "")
        if tools:
            prompt += "\n\n" + render_tools(tools)
        if response_format:
            prompt += "\n\n" + response_format_template.format(
                schema=to_json(response_format)
            )

    elif role == "developer":
        assert content, f"Invalid message for role `{role}`: {msg}"

View on GitHub (pinned to 0132848349)

Solutions

1. Use an effort listed in the error message for that profile (commonly 'low'/'medium'/'high').
2. Leave reasoning_effort None to accept the profile default ('low' for official, 'high' otherwise).
3. Align profile and effort values with the deployed model's docs.

Example fix

# before
encode_messages(msgs, reasoning_effort_profile='official', reasoning_effort='minimal')
# after
encode_messages(msgs, reasoning_effort_profile='official', reasoning_effort='low')

assert reasoning_effort in REASONING_EFFORT_PROFILES[reasoning_effort_profile]

def is_valid_effort(p,e): return e in REASONING_EFFORT_PROFILES.get(p,{})

When it happens

Trigger: reasoning_effort='minimal' with profile 'official', or any effort key not present in effort_prompts for the selected profile.

Common situations: Copying effort strings from OpenAI API ('minimal') to a DeepSeek profile that doesn't support it; mixing profile and effort from different versions.

Related errors

• Invalid reasoning effort profile: {reasoning_effort_profile!
• v_cache must be provided
• q can only be None when only_qv=True
• q must be provided unless qv is provided with only_qv=True
• `dt_bias` must have {HV * K} elements (got {dt_bias.numel()}

AI-assisted analysis of sgl-project/sglang@0132848349 (2026-08-28). Data as JSON: /api/errors/252014dafca87797. Report an issue: GitHub.