Environment:Sgl project Sglang CUDA SM100
Sgl_project_Sglang_CUDA_SM100 is the NVIDIA Blackwell SM100 GPU environment for SGLang, providing support for the Blackwell architecture's advanced features including Flash Attention 4, TMA-based warp-specialized kernels, and TensorRT-LLM MLA/MHA backends.
Requirements
- NVIDIA Blackwell GPU (SM100 or SM120 compute capability)
- CUDA Toolkit 12.8+
- NVIDIA driver 570+
- `nvidia-cutlass-dsl` >= 4.3.4 (for CUTLASS tile schedulers)
- PyTorch 2.9.1+ with CUDA 12.8 support
- `sgl-kernel` == 0.3.21 (with SM100 kernel binaries)
- `flashinfer_python` == 0.6.2 (with FA4 support)
Required By
- Implementation:Sgl_project_Sglang_SM100_MLA_Reduction
- Implementation:Sgl_project_Sglang_SM100_MLA_Tile_Scheduler
See Also
Page Connections
Double-click a node to navigate. Hold to expand connections.
Principle
Implementation
Heuristic
Environment