Search results

[Bug] qwen3.5 bf16 MTP broken

Status: Open.

#23123 In sgl-project/sglang;

· functionstackx opened on Apr 18, 2026

[Bug] [rocm] qwen 3.5 mtp fp4 broken

Status: Open.

#23113 In sgl-project/sglang;

· functionstackx opened on Apr 18, 2026

[Bug][ROCm][MI355X] Qwen3.5-397B-A17B-FP8 accuracy drop when tp=2 ep=2 with shared_experts_fusion is enabled

Status: Open.

#23084 In sgl-project/sglang;

· zhentaocc opened on Apr 17, 2026

[Bug] : ImportError: cannot import name 'dynamic_per_tensor_quant' from 'aiter' (unknown location)

Status: Open.

#23083 In sgl-project/sglang;

· jerin-metrum-ai opened on Apr 17, 2026

[Bug] EPD disaggregation silently drops requests under high load

Status: Open.

#23082 In sgl-project/sglang;

· mabubakar-purdue opened on Apr 17, 2026

[Bug] MiniMax-M2 streaming tool_calls: malformed response with name=null and duplicated arguments

Status: Open.

#23071 In sgl-project/sglang;

· eveningcafe opened on Apr 17, 2026

[Bug] trace can't match the source code strictly with moe-lora model

Status: Open.

#23066 In sgl-project/sglang;

· XiaotaoChen opened on Apr 17, 2026

[Bug] [Encode Prefill Disaggregation] [Func] [SDV] Qwen3-VL MoE encoder-only server crashes due to missing 'self.model' attribute

Status: Open.

#23063 In sgl-project/sglang;

· EdwardXuy opened on Apr 17, 2026

[Bug] b300 fp4 perf worse than b200 on deepseekv3

Status: Open.

#23061 In sgl-project/sglang;

· functionstackx opened on Apr 17, 2026

RFC: optional HIP external LLMM1 fast path for M=1 decode GEMMs

Status: Open.

#23058 In sgl-project/sglang;

· skyguan92 opened on Apr 17, 2026

[Bug] Dynamic chunking profiling crashes on GLM-5 model (AttributeError: _is_extend_in_batch)

Status: Open.

#23057 In sgl-project/sglang;

· Baichuan7 opened on Apr 17, 2026

[Diffusion] [NPU] Investigate performance degradation on Ascend

Status: Open.

#23055 In sgl-project/sglang;

· Makcum888e opened on Apr 17, 2026