sgl-project/sglang · error · ValueError

SANA-WM streaming requires positive prompt embeds.

Error message

SANA-WM streaming requires positive prompt embeds.

What it means

_resolve_stream_conditioning requires positive prompt embeds (pcfg.get_pos_prompt_embeds(batch)) for every streaming forward. If the text-encoding stage did not populate them, text conditioning is impossible and the stage raises immediately.

Source

Thrown at python/sglang/multimodal_gen/runtime/pipelines_core/stages/model_specific_stages/sana_wm/streaming.py:465

        # is for the dense path; using it here ran CFG=4.5 vs the reference's none.
        cfg_scale = sampler_cfg.streaming_cfg_scale
        do_cfg = bool(batch.do_classifier_free_guidance) and cfg_scale > 1.0
        if server_args.enable_cfg_parallel and do_cfg:
            raise NotImplementedError(
                "SANA-WM streaming does not support CFG parallel; run replicated."
            )

        # --- text conditioning ---
        pos_embeds = _to_device_dtype(
            _first_tensor(pcfg.get_pos_prompt_embeds(batch)),
            device=device,
            dtype=target_dtype,
        )
        pos_mask = _to_device_dtype(
            _first_tensor(batch.prompt_attention_mask), device=device
        )
        if pos_embeds is None:
            raise ValueError("SANA-WM streaming requires positive prompt embeds.")
        neg_embeds = neg_mask = None
        if do_cfg:
            neg_embeds = _to_device_dtype(
                _first_tensor(pcfg.get_neg_prompt_embeds(batch)),
                device=device,
                dtype=target_dtype,
            )
            neg_mask = _to_device_dtype(
                _first_tensor(batch.negative_attention_mask), device=device
            )
            if neg_embeds is None:
                raise ValueError(
                    "SANA-WM streaming CFG requires negative prompt embeds."
                )
            pos_embeds, neg_embeds, pos_mask, neg_mask = (
                _align_sana_wm_cfg_text_conditions(
                    pos_embeds, neg_embeds, pos_mask, neg_mask
                )

View on GitHub (pinned to 0132848349)

Solutions

1. Run the text-encoding stage before the streaming denoiser so pos prompt embeds exist
2. Ensure every request carries a prompt (even an empty string) that gets encoded
3. Validate pcfg.get_pos_prompt_embeds(batch) is not None before dispatch

Example fix

# before
resp = streaming_stage.forward(batch, server_args)  # no embeds
# after
batch = text_encode_stage.forward(batch, server_args)  # sets pos prompt embeds
resp = streaming_stage.forward(batch, server_args)

if pcfg.get_pos_prompt_embeds(batch) is None:
    batch = text_encode_stage.forward(batch, server_args)
assert pcfg.get_pos_prompt_embeds(batch) is not None

def has_pos_embeds(pcfg, batch) -> bool:
    return pcfg.get_pos_prompt_embeds(batch) is not None

null

When it happens

Trigger: Running the streaming denoiser without the prompt-encoding stage in the pipeline, or the encoder failing to attach pos prompt embeds to the Req; empty-prompt requests where embeds were skipped rather than encoded as empty.

Common situations: Pipeline misordering (denoiser before text encoder); prompt field absent in the request so downstream embeds are None; batch.extra dropping embeds during transfer.

Related errors

• SANA-WM realtime denoising expects this tick's pre-noised ch
• SANA-WM realtime denoising requires a realtime session
• chunk plan {plan} does not cover the incoming {incoming.shap
• SANA-WM streaming CFG requires negative prompt embeds.
• SANA-WM streaming denoising expects 5D latents (B, C, T, H,

AI-assisted analysis of sgl-project/sglang@0132848349 (2026-08-28). Data as JSON: /api/errors/8e70dff86ccae606. Report an issue: GitHub.