sgl-project/sglang · error · ValueError

Guidance scale must be positive, but got {batch.guidance_sca

Error message

Guidance scale must be positive, but got {batch.guidance_scale}

What it means

When CFG is active, guidance_scale must be non-negative (the check rejects values < 0). A negative guidance scale would push samples away from the prompt in an unsupported way, so validation fails before any denoising step runs.

Source

Thrown at python/sglang/multimodal_gen/runtime/pipelines_core/stages/input_validation.py:346

        if (
            batch.do_classifier_free_guidance
            and batch.negative_prompt is None
            and batch.negative_prompt_embeds is None
        ):
            raise ValueError(
                "For classifier-free guidance, either `negative_prompt` or "
                "`negative_prompt_embeds` must be provided"
            )

        # Validate number of inference steps
        if batch.num_inference_steps <= 0:
            raise ValueError(
                f"Number of inference steps must be positive, but got {batch.num_inference_steps}"
            )

        # Validate guidance scale if using classifier-free guidance
        if batch.do_classifier_free_guidance and batch.guidance_scale < 0:
            raise ValueError(
                f"Guidance scale must be positive, but got {batch.guidance_scale}"
            )

        # Reject requests that do not enable CFG on a server launched with
        # --enable-cfg-parallel. CFG-parallel splits cond/uncond across ranks,
        # so rank 1 has no work and returns None for noise_pred, which crashes
        # scheduler.step() ~30 minutes later under a gloo broadcast timeout.
        # Earlier, field-specific checks above (negative_prompt missing,
        # guidance_scale < 0) fire first and produce better messages for those
        # cases; this is the catch-all for any combination that still leaves
        # do_classifier_free_guidance=False under cfg-parallel.
        if server_args.enable_cfg_parallel and not batch.do_classifier_free_guidance:
            neg_prompt_state = (
                "not set"
                if batch.negative_prompt is None
                else "empty" if batch.negative_prompt == "" else "set"
            )
            raise ValueError(

View on GitHub (pinned to 0132848349)

Solutions

1. Use a non-negative guidance_scale (0 to ~15; 7.5 is a common default)
2. If you intended to disable CFG, set guidance_scale=0 or turn off do_classifier_free_guidance rather than negating it
3. Sanity-check parsed config values before submitting

Example fix

// before
out = pipe(prompt="a cat", guidance_scale=-7.5)

// after
out = pipe(prompt="a cat", guidance_scale=7.5)

if batch.do_classifier_free_guidance:
    batch.guidance_scale = abs(float(batch.guidance_scale))
    assert batch.guidance_scale >= 0

def valid_guidance_scale(scale, cfg: bool) -> bool:
    return not cfg or scale >= 0

When it happens

Trigger: Passing a negative guidance_scale (e.g. -7.5) together with do_classifier_free_guidance=True, or a request builder that negates/mis-parses the scale field.

Common situations: Typos/sign errors in client configs; JSON configs where guidance_scale is read as a negative number; experiments with 'negative guidance' not supported by this pipeline; note the check allows 0 (which effectively disables CFG weighting) but not below.

Related errors

• For classifier-free guidance, either `negative_prompt` or `n
• Number of inference steps must be positive, but got {batch.n
• unsupported input for causal Conv3D cat/pad CUDA
• unsupported input for usp_merge_heads CUDA
• unsupported input for modulate_scale_shift CUDA

AI-assisted analysis of sgl-project/sglang@0132848349 (2026-08-28). Data as JSON: /api/errors/3c1f85b48b0b9782. Report an issue: GitHub.