v0.2.6: CUTLASS backend, Multi-LLM concurrent execution#70

Merged

m96-chan merged 10 commits intomainfrom

feature/v0.2.6-fp16-tensorcore

Dec 15, 2025

Commits on Dec 15, 2025

feat(fp16-bf16): implement TensorCore GEMM with mma.sync.m16n8k16

m96-chan
and
claude
committed
docs(readme): add v0.2.6 release notes and update performance tables

m96-chan
and
claude
committed
feat(fp16-bf16): add TC_GENERIC kernel for flexible matrix sizes

m96-chan
and
claude
committed
refactor(ops): modularize basic.cu into focused operation modules

m96-chan
and
claude
committed
v0.2.6: Multi-LLM async execution, FP16/BF16 TensorCore, LLM support

m96-chan
and
claude
committed
feat(matmul): integrate CUTLASS as default GEMM backend

m96-chan
and
claude
committed
ci: add submodules checkout for CUTLASS

m96-chan
and
claude
committed
docs(readme): update for CUTLASS backend

m96-chan
and
claude
committed
docs: clarify Multi-LLM concurrent execution (not speedup)

m96-chan
and
claude
committed
chore: remove PyTorch-based Multi-LLM demo

m96-chan
and
claude
committed