Search results

[AMD] Add fp8_e4m3 kv-cache-dtype and deterministic inference to MLA FP8 test on ROCm

Status: Open.

#23536 In sgl-project/sglang;

· bingxche opened on Apr 23, 2026

[Bug] sglang-router 0.3.2 launch_server breaks on sglang >= 0.5.10: is_port_available moved to utils/network.py but not re-exported from utils/__init__.py

Status: Open.

#23535 In sgl-project/sglang;

· DragonLsy opened on Apr 23, 2026

[Bug] Segmentation fault in torch._inductor.compile_worker during disaggregated prefill

Status: Open.

#23526 In sgl-project/sglang;

· ishandhanani opened on Apr 23, 2026

[Bug][Perf] Unnecessary sync in flashinfer attn backend for Qwen3.5-397B-A17B-FP8 model

Status: Open.

#23500 In sgl-project/sglang;

· xyuzh opened on Apr 22, 2026

Segfault in NIXL UCX worker during disaggregated KV transfer

Status: Open.

#23499 In sgl-project/sglang;

· ishandhanani opened on Apr 22, 2026

AMD Development Roadmap (2026 Q2)

amd

Status: Open.

#23494 In sgl-project/sglang;

· HaiShaw opened on Apr 22, 2026

[Bug] Kimi K2.6 + EAGLE3 startup failure

Status: Open.

#23490 In sgl-project/sglang;

· mmangkad opened on Apr 22, 2026

[Question] what‘s the optimal setting for launching deepseek serving with moe-lora ?

Status: Open.

#23483 In sgl-project/sglang;

· XiaotaoChen opened on Apr 22, 2026

[Bug] Qwen3.5 produces incorrect results when attention DP is enabled and NCCL NVLS kernels are used

Status: Open.

#23463 In sgl-project/sglang;

· jinyangyuan-nvidia opened on Apr 22, 2026

[Bug] Runtime attaching Mooncake HiCache storage backend on multi-node deployment cannot resolve per-node local_hostname correctly

Status: Open.

#23457 In sgl-project/sglang;

· LJL36 opened on Apr 22, 2026

Proposal: Keep using Sphinx for Documentation Translation

Status: Open.

#23450 In sgl-project/sglang;

· hwhsu1231 opened on Apr 22, 2026

[Bug] operands could not be broadcast together in PD Disaggregation with long input requests(GLM5-FP8)

Status: Open.

#23444 In sgl-project/sglang;

· hmthhh opened on Apr 22, 2026