hiyouga/LlamaFactory · error · ValueError

MOSS-VL encountered an incomplete video token block after to

Error message

MOSS-VL encountered an incomplete video token block after tokenization. Please increase `cutoff_len` or reduce `video_maxlen`.

What it means

MOSS-VL post-tokenization check: the token stream ends while still inside a video block (a vision_start was seen with no matching vision_end). The video placeholder was truncated before its terminating token, so frame counting cannot complete.

Source

Thrown at src/llamafactory/data/mm_plugin.py:596

                in_video = True
                current_video_frames = 0
            elif token_id == processor.vision_end_token_id:
                if not in_video:
                    raise ValueError(
                        "MOSS-VL encountered a video end token without a matching start token after tokenization. "
                        "Please increase `cutoff_len` if a video placeholder was truncated."
                    )

                video_frame_counts.append(current_video_frames)
                in_video = False
            elif token_id == processor.image_token_id:
                if in_video:
                    current_video_frames += 1
                else:
                    media_order.append("image")

        if in_video:
            raise ValueError(
                "MOSS-VL encountered an incomplete video token block after tokenization. "
                "Please increase `cutoff_len` or reduce `video_maxlen`."
            )

        if media_order.count("image") != num_images or media_order.count("video") != num_videos:
            raise ValueError(
                "MOSS-VL media tokens do not match the provided media after tokenization: "
                f"order={media_order}, images={num_images}, videos={num_videos}. "
                "Please increase `cutoff_len` if a visual placeholder was truncated."
            )

        if expected_video_frames is not None and video_frame_counts != expected_video_frames:
            raise ValueError(
                "MOSS-VL video frame tokens do not match the processed video after tokenization: "
                f"tokens={video_frame_counts}, frames={expected_video_frames}. "
                "Please increase `cutoff_len` or reduce `video_maxlen`."
            )

View on GitHub (pinned to f28afaf635)

Solutions

Increase cutoff_len to cover the largest video placeholder in the dataset.
Reduce video_maxlen / video_fps so each video yields fewer frame tokens.
Pre-compute worst-case placeholder length (frames x tokens-per-frame) and set cutoff_len above it.

Example fix

### before
cutoff_len: 4096
### after
cutoff_len: 32768
video_maxlen: 128

Defensive patterns

Strategy: validation

Validate before calling

ids = tokenizer(sample_text_with_placeholders).input_ids
open_blocks = 0
for t in ids:
    if t == processor.vision_start_token_id: open_blocks += 1
    elif t == processor.vision_end_token_id: open_blocks -= 1
assert open_blocks == 0, 'truncated video placeholder; increase cutoff_len / reduce video_maxlen'

Prevention

Budget cutoff_len as max_text_len + total_media_tokens with margin.
Cap videos per sample when cutoff_len is fixed.

When it happens

Trigger: A MOSS-VL sample whose expanded video placeholder exceeds the remaining cutoff_len budget; the tail of the sequence (containing vision_eos) is dropped. The error message itself points at cutoff_len or video_maxlen.

Common situations: High-resolution / many-frame videos; cutoff_len tuned for text-only data; video_maxlen default too generous for the chosen cutoff.

Related errors

AI-assisted analysis of hiyouga/LlamaFactory@f28afaf635 (2026-08-14). Data as JSON: /api/errors/31e4abae2dac9f42. Report an issue: GitHub.