Search results

[Bug]: /v1/audio/speech token metrics remain 0 in /metrics for Qwen3-TTS

bugSomething isn't workingSomething isn't working

Status: Open.

#4444 In vllm-project/vllm-omni;

· gxxx-hum opened on Jun 15, 2026

[Bug] QwenImage RoPE txt_seq_lens uses mask.sum() instead of padded embed width (misaligned with diffusers)

bugSomething isn't workingSomething isn't working

Status: Open.

#4443 In vllm-project/vllm-omni;

· zhtmike opened on Jun 15, 2026

[Feature]: AR async scheduling is enabled but decode launches still do not overlap

Status: Open.

#4442 In vllm-project/vllm-omni;

· fake0fan opened on Jun 15, 2026

[Bug]: The calculation method of ep size in Omni is inconsistent with that in vLLM.

bugSomething isn't workingSomething isn't working

Status: Open.

#4438 In vllm-project/vllm-omni;

· BLANKETusers opened on Jun 15, 2026

[Installation]: Why is vllm-omni constantly throwing errors on Qwen3-omni?

Status: Open.

#4427 In vllm-project/vllm-omni;

· CTZhou-byte opened on Jun 15, 2026

[Performance][Discussion]: FA2 varlen vs PyTorch SDPA varlen: performance tradeoff and numerical drift

bugSomething isn't workingSomething isn't working

Status: Open.

#4418 In vllm-project/vllm-omni;

· asukaqaq-s opened on Jun 14, 2026

[Bug] : Fix SHM Chunk & Sender-State Leaks on Request Abort

Status: Open.

#4414 In vllm-project/vllm-omni;

· gagandhakrey opened on Jun 14, 2026

[Doc]: E2E tests for Hunyuan-Image model do not support torch.profiler

documentationImprovements or additions to documentationImprovements or additions to documentation

Status: Open.

#4412 In vllm-project/vllm-omni;

· AbelSara opened on Jun 14, 2026

[Bug]: higgs_audio_v3 test_concurrent_pcm_streaming consistently fails HNR threshold (concurrent PCM streaming)

bugSomething isn't workingSomething isn't working

Status: Open.

#4411 In vllm-project/vllm-omni;

· tzhouam opened on Jun 13, 2026

[Bug]: Hunyuanimage accuracy accuracy regression

bugSomething isn't workingSomething isn't working

Status: Open.

#4404 In vllm-project/vllm-omni;

· Bounty-hunter opened on Jun 13, 2026

[RFC/Bug] Improve diffusion worker/engine control-plane reliability and output contracts

bugSomething isn't workingSomething isn't working

help wantedExtra attention is neededExtra attention is needed

high priorityhigh priority issue, needs to be done asaphigh priority issue, needs to be done asap

Status: Open.

#4400 In vllm-project/vllm-omni;

· hsliuustc0106 opened on Jun 13, 2026

[Bug]: Streaming-input sessions never adopt a later chunk's max_tokens — every segment is capped by segment 1's budget (StreamingUpdate.max_tokens exists, unused)

Status: Open.

#4385 In vllm-project/vllm-omni;

· Liangtaiwan opened on Jun 12, 2026