Repository Issue Activity (beta)

#129 5070 SM120 Unsupported architecture with `TllmGenFmhaRunner`

#128 does mini-sglang support SM_75

#51 /v1/chat/completions Always returns SSE streaming responses, which cannot be forced to return a one-time JSON via stream=false or request headers.

#123 Support / Clarify Expert Parallelism (EP) support in mini‑SGLang

#90 Interest in adding AMD GPU (ROCm) support

#120 [Feature Request] Mac/Apple Silicon (MPS) support via fallback attention backend

#102 `benchmark/offline/bench.py` hits CUDA illegal memory access unless overlap scheduling is disabled

#12 CPU support

#84 [BUG] GPU Memory leak during multi-GPU inference.

#77 License for mini-sglang project

#9 MOE Support

#15 [Bug] Backend continues generation after client disconnects & Missing cancellation support

#26 Docker image support

#67 [Bug] CUDA illegal memory access due to unsynchronized non-blocking transfer in overlap scheduling

#61 I run the mini-sglang on my spark, does the project need a pr for this ?

#60 torch.compile compatibility: significant performance overhead due to guard failures and recompilations

#58 crash to run offline inference

#31 any VLM models to support

#36 Bug: model like qwen3/qwen2.5 cannot find the attribute "dtype"

#34 gen_initializer_constant: ran out of fields

#20 nvcc version do not support C++20 std::source_location completely

#28 any diffusion model support plan？

#18 Suggestion: Change github about page to: A compact implementation of SGLang, designed to demystify the complexities of modern LLM serving systems

#30 Hassanlovey20