sgl-project/sglang · error · ValueError

image_token_count required for an image reference

Error message

image_token_count required for an image reference

What it means

While walking condition_labels, an image reference was encountered but there are fewer normalized image_token_count entries than image references in the plan.

Source

Thrown at python/sglang/multimodal_gen/runtime/pipelines_core/stages/model_specific_stages/minimax_h3/presentation.py:266

    )
    image_token_counts = _as_int_list(image_token_count, name="image_token_count")
    video_counts_by_ref = _as_nested_int_list(
        video_block_token_counts,
        name="video_block_token_counts",
    )
    video_timestamps_by_ref = _as_nested_float_list(
        video_block_timestamps,
        name="video_block_timestamps",
    )
    if len(video_counts_by_ref) != len(video_timestamps_by_ref):
        raise ValueError("video block token counts and timestamps must align")
    image_seen = 0
    video_seen = 0
    for cond_type, ordinal in condition_labels:
        if cond_type == "image":
            image_seen += 1
            if image_seen > len(image_token_counts):
                raise ValueError("image_token_count required for an image reference")
            count = int(image_token_counts[image_seen - 1])
            if count <= 0:
                raise ValueError("image_token_count required for an image reference")
            presentation.text(_text_ids(tokenizer, f"<Picture {ordinal}>: "))
            presentation.vision(_vision_block_ids(tokenizer, IMAGE_PAD, count))
        elif cond_type == "audio":
            presentation.text(_text_ids(tokenizer, f"<Audio {ordinal}>: "))
        elif cond_type == "video":
            video_seen += 1
            if video_seen > len(video_counts_by_ref):
                raise ValueError(
                    "video reference requires block token counts and timestamps"
                )
            counts = video_counts_by_ref[video_seen - 1]
            timestamps = video_timestamps_by_ref[video_seen - 1]
            if not counts or not timestamps:
                raise ValueError(
                    "video reference requires block token counts and timestamps"

View on GitHub (pinned to 0132848349)

Solutions

1. Pass image_token_count as a list with one positive entry per image reference
2. Regenerate counts to match the current condition_labels plan
3. Remove extra image references from the plan

Example fix

// before
condition_labels=[("image",1),("image",2)], image_token_count=196
// after
condition_labels=[("image",1),("image",2)], image_token_count=[196,196]

n_images = sum(1 for t, _ in condition_labels if t == "image")
assert len(image_token_counts_list) >= n_images

When it happens

Trigger: condition_labels contains 2 image entries but image_token_count=196 (a single int normalizes to [196]) — the second image reference finds no count.

Common situations: Passing a scalar image_token_count for a multi-image plan; forgetting to extend counts for additional images; plan changed to include more images than the conditioning data provides.

Related errors

• video block token counts and timestamps must align
• video reference requires block token counts and timestamps
• unused image_token_count entries
• unused video block token count entries
• seq_len {seq_len} < used rows {used}

AI-assisted analysis of sgl-project/sglang@0132848349 (2026-08-28). Data as JSON: /api/errors/aeb4198d98304d50. Report an issue: GitHub.