Search results

Eval dataset tip

Status: Open.

#29987 In sgl-project/sglang;

· connerlambden opened on Jul 2, 2026

[Bug] [Perf] Long non-overlapped cudaGraphLaunch in GLM-5.1 MTP target verify; propose pipelined two-graph decode replay

Status: Open.

#29960 In sgl-project/sglang;

· RunFMe opened on Jul 2, 2026

[Bug] PDMux missing prefill_stream -> decode_stream dependency after split-prefill merge causes IndexKernel assert

Status: Open.

#29958 In sgl-project/sglang;

· MonkeykingD opened on Jul 2, 2026

[Bug] Deadlock in decode scheduler when running with --grpc-mode

Status: Open.

#29957 In sgl-project/sglang;

· 937869301 opened on Jul 2, 2026

[Bug] SGLang does not gracefully clean Mooncake HiCache metadata on process termination

Status: Open.

#29955 In sgl-project/sglang;

· riZZZhik opened on Jul 2, 2026

[Feature] [GLM5.2] Index share reuse with MHA

Status: Open.

#29951 In sgl-project/sglang;

· mattteochen opened on Jul 2, 2026

[Bug] DeepEP low_latency buffer lazy init fails during CUDA graph capture with PP=2, TP=8, DP-attention, EP=8 on Kimi K2.6 W4A8

Status: Open.

#29942 In sgl-project/sglang;

· zmtttt opened on Jul 2, 2026

[Bug] [GLM5.2] When the tools are set to required mode, the output becomes repetitive and garbled.

Status: Open.

#29914 In sgl-project/sglang;

· whybeyoung opened on Jul 2, 2026

FP4 KV Roadmap

Status: Open.

#29913 In sgl-project/sglang;

· samuellees opened on Jul 2, 2026

[Bug] Unknown diffusion LLM: DiffusionGemmaForBlockDiffusion

Status: Open.

#29904 In sgl-project/sglang;

· giovan001 opened on Jul 2, 2026

[Bug] sgl-kernel tests lack compute-capability gates: 253 expected-unsupported failures on SM120 (deepgemm ue8m0, flashmla, fp8_blockwise_moe)

Status: Open.

#29900 In sgl-project/sglang;

· waynehacking8 opened on Jul 2, 2026

[Bug] MLA model branches missing scaling derivation crash FlashInferMLA backend at startup (Kimi-VL, MiniCPM3, DeepSeek-VL2)

Status: Open.

#29891 In sgl-project/sglang;

· waynehacking8 opened on Jul 2, 2026