Group 1 - Hardware-Aware Transformer Optimisation: Integrating Programmable Attention, Triton Kernel Fusion, and Multi-Objective NAS#315

Open

aahaidar01 wants to merge 83 commits into

DeepWok:mainfrom

aahaidar01:main

Commits on Feb 5, 2026

Commits on Feb 6, 2026

[Modify] Modified md file lab 3 section
aahaidar01
committed

Commits on Feb 7, 2026

Commits on Mar 22, 2026

Added flex attention transform pass and score mods (diff score funtionalities) into mase /src files. Added pytest test scripts to test the implemented functionalities.
aahaidar01
committed
Fix FlexAttention CUDA forward pass: position_embeddings support, torch.compile compatibility, return tuple fixes, and add training/bf16/seq512 tests
aahaidar01
committed
Added sdpa vs flex experiments and latency benchmark to assess sdpa vs flex throughput
aahaidar01
committed

Commits on Mar 24, 2026

Commits on Mar 25, 2026

Commits on Mar 27, 2026