sgl-project/sglang · critical · Error

This browser does not support worker image decoding

Error message

This browser does not support worker image decoding

What it means

The float32 branch of deterministic_all_reduce only launches cross_device_reduce_1stage kernels for world sizes 2,4,6,8. Any other fa->world_size_ throws.

Source

Thrown at python/sglang/multimodal_gen/apps/realtime_webui/decoder_worker.js:102

  const bytes = payload instanceof Uint8Array ? payload : new Uint8Array(payload);
  const lengths = Array.isArray(header.payload_lengths) && header.payload_lengths.length
    ? header.payload_lengths.map(Number)
    : [bytes.byteLength];
  const payloads = [];
  let offset = 0;
  for (const length of lengths) {
    payloads.push(bytes.buffer.slice(
      bytes.byteOffset + offset,
      bytes.byteOffset + offset + length,
    ));
    offset += length;
  }
  return payloads;
}

async function encodedFramesToImageBitmaps(header, payload) {
  if (typeof createImageBitmap === "undefined") {
    throw new Error("This browser does not support worker image decoding");
  }

  const frames = await Promise.all(splitEncodedPayload(header, payload).map((framePayload) => (
    createImageBitmap(new Blob([framePayload], { type: header.content_type }))
  )));
  return {
    width: frames[0]?.width || 0,
    height: frames[0]?.height || 0,
    frame_type: "bitmap",
    frames,
  };
}

async function decode(header, payload) {
  let rawPayload;
  if (
    header.content_type === WEBP_FRAME_CONTENT_TYPE ||
    header.content_type === JPEG_FRAME_CONTENT_TYPE

View on GitHub (pinned to 0132848349)

Solutions

1. Use TP in {2,4,6,8} for deterministic inference on ROCm
2. Disable the deterministic custom kernel or fall back to RCCL for unsupported sizes
3. Check visible-device masks so world size lands on a supported value

assert fa.world_size_ in (2,4,6,8), 'deterministic allreduce needs TP in (2,4,6,8)'

When it happens

Trigger: Invoking deterministic allreduce with a CustomAllreduce object whose world_size_ is not 2,4,6,8, with float32 tensors.

Common situations: Odd or >8 GPU deterministic runs (e.g. --enable-deterministic-inference with TP=3 or TP=16); deterministic mode enabled on topologies the kernel does not support.

Related errors

• Previous frame size does not match current delta payload
• Generate subcommand is not yet supported for model: {model_p
• Error: --model-type requires a non-empty value.
• delta payload size mismatch: expected ${expectedSize}, got $
• Missing previous frame for delta payload

AI-assisted analysis of sgl-project/sglang@0132848349 (2026-08-28). Data as JSON: /api/errors/5968e7d279b310c8. Report an issue: GitHub.