Pinned Loading

1. nano_vllm_pp nano_vllm_pp Public

   Python

2. cross_dsl_kernels cross_dsl_kernels Public

   CUDA, Triton and CuTe DSL kernels for transformer inference: fused RMSNorm, decode GEMV, RoPE/KV append and W4A16, with benchmarks and profiling.

   Python

3. triton-flash-attention triton-flash-attention Public

   Fused FlashAttention-2 forward and backward kernels in Triton with MHA/GQA, CUDA validation, reproducible benchmarks, and Nsight profiling.

   Python

4. torch-compile-internals torch-compile-internals Public

   Python

5. lpcv2026 lpcv2026 Public

   MobileCLIP2-S2 export, retrieval validation, and Qualcomm QNN deployment for LPCVC 2026 Track 1; verified FP16 results and W8A16 experiments.

   Python