feat: Add SM121/GB10 (DGX Spark) Blackwell-class GPU support#31740

Open

seli-equinix wants to merge 13 commits intovllm-project:mainfrom

seli-equinix:feature/sm121-gb10-support

Commits on Mar 11, 2026

feat: Add SM121/GB10 (DGX Spark) Blackwell-class GPU support

seli-equinix
authored and
Seli
committed
fix: Handle empty/whitespace tool_call arguments in chat preprocessing

authored and
Seli
committed
feat: Auto-configure TRITON_PTXAS_PATH for SM121/GB10

authored and
Seli
committed
fix: Add SM121 (12.1) arch support for CUTLASS SM120 kernels

seli-equinix
authored and
Seli
committed
feat: Add NVIDIA GB10 MoE tuning config for FP8

authored and
Seli
committed
fix: Address Copilot review comments for SM121/GB10 support

authored and
Seli
committed
Add metrics visualization with plot plugin

seli-equinix
authored and
Seli
committed
fix: Allow FP8 block-scale MoE on SM121 via FlashInfer CUTLASS backend

authored and
Seli
committed
fix: Split has_flashinfer_nvfp4 from has_flashinfer_cutlass_fused_moe

authored and
Seli
committed
fix: Handle raw torch.fx.Node references in AOT compile serialization

seli-equinix
authored and
Seli
committed
fix: GB10 nogds + add Qwen3.5 MoE tuning config (E=256,N=512)
Seli
committed
fix: Prefer FlashInfer for ViT attention on SM121/GB10 (Flash Attention PTX lacks SM121)
Seli
committed
fix: Add SM121 to CUDA_SUPPORTED_ARCHS (was filtered out by arch intersection)
Seli
committed