This repo contains reading material and presentations of the work 10xE did on RISE RFP RP-014
| # | Project Phase | Objective | Status | Reference Link |
|---|---|---|---|---|
| 1 | Floating-Point | SIMD Mappings and Relevant Kernels | Merged | PR #17318 |
| 1 | Floating-Point | Activation Functions and Utilities | Merged | PR #17318 |
| 1 | Floating-Point | Llamafile SGEMM | Merged | PR #18199 |
| 1 | Floating-Point | Flash Attention | Merged | PR #20627 |
| 2 | Quantization | Vector Dot (I-Quants and MXFP4) – 256-bit | Merged | PR #18859 |
| 2 | Quantization | Vector Dot (I-Quants and Ternary) – 256-bit | Merged | PR #18784 |
| 2 | Quantization | Vector Dot (128-bit) | Merged | PR #20633 |
| 2 | Quantization | Vector Dot (512-bit and 1024-bit) | Merged | PR #22754 |
| 3 | VLEN-Aware Repacking | Repack GEMM and GEMV – Floating-Point (128-bit to 1024-bit) | In Review | PR #17791 |
| 3 | VLEN-Aware Repacking | Repack GEMM and GEMV – Quantization (256-bit) | Merged | PR #19121 |
| 3 | VLEN-Aware Repacking | Repack GEMM and GEMV (Other VLENs, Q5_K, MXFP4) | Reviewed | PR #20723 |
| 3 | VLEN-Aware Repacking | Repack GEMM and GEMV (Q3_K, Q6_K) | To Be Reviewed | PR #23745 |
| 4 | Testing and Benchmarking Framework | Testing Files | Merged | llama.cpp-validation |