[Quantization] Fp8 Channelwise Dynamic Per Token GroupedGEMM#15587

Merged

robertgshaw2-redhat merged 71 commits intovllm-project:mainfrom

robertgshaw2-redhat:dyn-per-token-grouped-gemm

Mar 27, 2025

Commits on Dec 6, 2024

Cutlass grouped gemm files
ElizaWszola
committed

Commits on Dec 16, 2024

Commits on Dec 17, 2024

Grouped gemm working

ElizaWszola
and
LucasWilkinson
committed
Small cleanup
ElizaWszola
committed

Commits on Jan 8, 2025

Merge branch 'main' into grouped-gemm-with-group-id
ElizaWszola
committed

Commits on Jan 13, 2025

Commits on Jan 17, 2025

Start working on fused moe cutlass implementation
ElizaWszola
committed

Commits on Jan 20, 2025

Working halfway
ElizaWszola
committed

Commits on Jan 23, 2025

Commits on Jan 24, 2025

pass input as one tensor with an array of offsets rather than a list of tensors
ElizaWszola
committed

Commits on Jan 28, 2025

Commits on Jan 29, 2025

Commits on Jan 31, 2025

Commits on Feb 4, 2025

profile
ElizaWszola
committed

Commits on Feb 14, 2025

Commits on Feb 18, 2025

Performance, add channelwise scales everywhere
ElizaWszola
committed

Commits on Feb 20, 2025

Commits on Feb 24, 2025

Commits on Feb 25, 2025

Commits on Feb 27, 2025

Faster data preparation kernels, bring back correct benchmark shapes
ElizaWszola
committed

Commits on Feb 28, 2025

enable cutlass grouped gemm only on sm90
ElizaWszola
committed

Commits on Mar 4, 2025

Merge branch 'main' into grouped-gemm-with-group-id
ElizaWszola
committed

Commits on Mar 12, 2025

Commits on Mar 17, 2025

Commits on Mar 21, 2025

Add hack for accepting int input in weak_ref_tensors
ElizaWszola
committed

Commits on Mar 24, 2025

Commits on Mar 27, 2025