sgl-project/sglang-omni

[Qwen3-Omni Perf] Code2Wav: wire configurable initial codec chunk size into the Qwen3-Omni vocoder scheduler 5 days ago

good first issue

678

SGLang Omni: High-Performance Multi-Stage Pipeline Framework for Omni Models

Python

[Qwen3-Omni Perf] Thinker prefill: batch the multimodal embedding merge, remove per-request host syncs 5 days ago

good first issue

sgl-project/sglang-omni

678

SGLang Omni: High-Performance Multi-Stage Pipeline Framework for Omni Models

Python

[Qwen3-Omni Perf] Thinker: vectorize MRoPE position computation; skip multimodal positions on speech stages 6 days ago

good first issue

sgl-project/sglang-omni

678

SGLang Omni: High-Performance Multi-Stage Pipeline Framework for Omni Models

Python

[Qwen3-Omni Perf] Code2Wav: CUDA-graph shape hit-rate telemetry 6 days ago

good first issue

sgl-project/sglang-omni

678

SGLang Omni: High-Performance Multi-Stage Pipeline Framework for Omni Models

Python

[Qwen3-Omni Perf] Predictor attention: use SDPA `enable_gqa` instead of materialized KV expansion 6 days ago

good first issue

sgl-project/sglang-omni

678

SGLang Omni: High-Performance Multi-Stage Pipeline Framework for Omni Models

Python

[Qwen3-Omni Perf] Encoder batch-wait sweep: tune hard-coded max_batch_wait_ms 13 days ago

good first issue

sgl-project/sglang-omni

678

SGLang Omni: High-Performance Multi-Stage Pipeline Framework for Omni Models

Python

[Good First Issue] Optimize Qwen3-ASR for faster TTS WER evaluation about 2 months ago

good first issue

sgl-project/sglang-omni

678

SGLang Omni: High-Performance Multi-Stage Pipeline Framework for Omni Models

Python

[Bug] OpenAI /v1/chat/completions text-only streaming emits a single chunk (no per-token TPOT) about 2 months ago

good first issue

sgl-project/sglang-omni

678

SGLang Omni: High-Performance Multi-Stage Pipeline Framework for Omni Models

Python

Open Issues Need Help