Build (aarch64-linux)

Build (x86_64-linux)

Enable Torch 2.8 build

fix: remove old status badge

fix: bump to trigger build via change

feat: add tag for hfjob build

Build (AArch64)

Build

Update flake input

Use standard capabilities for the moe/fp8/activation kernels

Build

Update flake inputs

Fix capabilities

Build (aarch64)

Build

Update flake

Sync capabilities with upstream

Build

Llama4TextMoe: mark as not having backward

Build (fp8)

Handle FP8

Build (Llama 4)

Fix flake URL and add flake.lock

Sync with vLLM and add `Llama4TextMoe` layer

feat: update to include rev in kernel for reproducible symbols

Update for build.toml changes

Update README.md (#1)

Build

Remove flake.lock

Vendor `w8a8_block_fp8_matmul` and `per_token_group_quant_fp8`

Build

Normalize some directory names

Make functions visible to make a custom `fused_moe`

Build

Sync with upstream

Build (Torch 2.6)

Build

Try to avoid fake op registration issues

Build

Export ops at the top-level

Update flake lock

Build

Track shared libraries

Make compatible with latest kernel-builder

Add MoE configs

Add MoE kernels from vLLM

initial commit