Search results

[Bug][ROCm] Model using dynamic FP8 activation quantization generates wrong outputs when using CUDA graph on MI355X

Status: Open.

#18002 In sgl-project/sglang;

· fxmarty-amd opened on Jan 30, 2026

[Bug] Parsing fails when streaming multiple tool calls with <code>tool_choice="auto"</code>

Status: Open.

#18001 In sgl-project/sglang;

· halaction opened on Jan 30, 2026

[Bug] tool_choice with specific function causes model to stall on multi-call prompts

Status: Open.

#17998 In sgl-project/sglang;

· tugot17 opened on Jan 30, 2026

[Bug] ascend_memory_consumption nightly test failed

Status: Open.

#17994 In sgl-project/sglang;

· OrangeRedeng opened on Jan 30, 2026

[Bug] Performance of Cache-DiT is significantly lower than Diffusers + Cache-DiT under 8-step inference using distilled LoRAs (e.g., LightX2V).

Status: Open.

#17990 In sgl-project/sglang;

· boowingg opened on Jan 30, 2026

[QUESTION] Question for XIAOMImimo,DIFF between huggingface and sglang！！

Status: Open.

#17980 In sgl-project/sglang;

· xhjhggybz opened on Jan 30, 2026

[Bug] When using KV cache quantization in TP8, there was no effect.

Status: Open.

#17979 In sgl-project/sglang;

· taishan1994 opened on Jan 30, 2026

[Bug] "Multi-threaded loading of shards: 100%" and "Load weight end" have a long interval between them, meaning the model loading is particularly slow. vLLM multi-threaded loading only takes about 30 minutes.

Status: Open.

#17970 In sgl-project/sglang;

· lianghao6 opened on Jan 30, 2026

[Bug] Deepseek 3.2 and GLM NVFP4 high memory usage

Status: Open.

#17955 In sgl-project/sglang;

· pdasgup opened on Jan 29, 2026

[Bug] DeepSeek-V3.1 may forget <code>&lt;/think&gt;</code> token when using tool calling with the recommended chat template

Status: Open.

#17943 In sgl-project/sglang;

· Ray-Eldath opened on Jan 29, 2026

[Bug] DeepSeek-V3.2 tool calls incorrectly output to content field instead of tool_calls field with

Status: Open.

#17942 In sgl-project/sglang;

· Rabbit0916 opened on Jan 29, 2026

[Bug] SGLang hangs with DeepSeek-V3.2 when using pipeline parallelism

Status: Open.

#17941 In sgl-project/sglang;

· jellysnack opened on Jan 29, 2026