| Commit message (Collapse) | Author | Files | Lines | ||
|---|---|---|---|---|---|
| 2026-06-16 | kernels: split kernels.c into gemm/norm/act/attn/rope by kind | Onur Gündüz | 1 | -1271/+0 | |
| 2026-06-16 | kernels: extract the persistent thread pool into threadpool.{c,h} | Onur Gündüz | 1 | -207/+9 | |
| 2026-06-16 | chore: move sources to src/ with proper names, drop qwen_ prefix | Onur Gündüz | 1 | -86/+86 | |
| 2026-06-16 | cpu: size default threads to cgroup cpu quota, not host core count | Onur Gündüz | 1 | -2/+55 | |
| 2026-06-16 | all: support BERT tanh-GeLU (gelu_new / gelu_pytorch_tanh) | Onur Gündüz | 1 | -0/+9 | |
| 2026-06-15 | kernels: add LayerNorm and exact-GeLU for the BERT family | Onur Gündüz | 1 | -0/+38 | |
| 2026-06-15 | chore: run clang-format | Onur Gündüz | 1 | -221/+313 | |
| 2026-06-14 | all: add support for Qwen3-Embedding + causal GQA attention | Onur Gündüz | 1 | -0/+152 | |
| 2026-06-14 | kernels: delete dead audio/causal-LM leftovers and unused exports | Onur Gündüz | 1 | -497/+2 | |
| 2026-06-14 | cpu: move BF16 widening scratch into the workspace | Onur Gündüz | 1 | -130/+5 | |
| 2026-06-14 | kernels: drop asr_ prefix naming | Onur Gündüz | 1 | -2/+2 | |
| 2026-06-14 | kernels: fix OpenBLAS thread contention and numerical instability | Onur Gündüz | 1 | -2/+30 | |
| 2026-06-14 | kernels: parallelize large rmsnorm rows | Onur Gündüz | 1 | -5/+48 | |
| 2026-06-14 | late: add packed maxsim scoring | Onur Gündüz | 1 | -4/+8 | |
| 2026-06-14 | kernels: vectorize bf16 widening on arm | Onur Gündüz | 1 | -0/+14 | |
| 2026-06-14 | kernels: use sgemv for single-row f32 linear | Onur Gündüz | 1 | -4/+10 | |
| 2026-06-14 | kernels: add neon rope apply path | Onur Gündüz | 1 | -0/+21 | |
| 2026-06-14 | cpu: reuse tiled attention scratch buffers | Onur Gündüz | 1 | -12/+46 | |
| 2026-06-14 | cpu: tile packed attention through blas | Onur Gündüz | 1 | -54/+110 | |
| 2026-06-14 | kernels,cpu: fuse tiny BF16 gate up projections | Onur Gündüz | 1 | -0/+93 | |
| 2026-06-14 | kernels,cpu: fuse small BF16 QKV projections | Onur Gündüz | 1 | -15/+40 | |
| 2026-06-14 | kernels: validate real BF16 artifacts | Onur Gündüz | 1 | -0/+52 | |
| 2026-06-14 | cpu: fuse CPU embedding hot paths | Onur Gündüz | 1 | -0/+26 | |
| 2026-06-14 | kernels: avoid tiny packed attention dispatch | Onur Gündüz | 1 | -1/+7 | |
| 2026-06-14 | all: support batched embedding attention | Onur Gündüz | 1 | -0/+95 | |
| 2026-03-28 | Refactor and enhance pplx_embed and pplx_embed_mlx for improved model support | Onur Gündüz | 1 | -5/+0 | |
| - Updated `pplx_embed.h` to support new embedding models: `pplx-embed-v1-4b` and `pplx-embed-context-v1-4b`. - Simplified model configuration structure and constants for better maintainability. - Enhanced comments for clarity on model architecture and configuration. - Modified `pplx_embed_mlx.c` to streamline loading of model weights and configuration from `config.json`. - Improved memory management and error handling in the MLX backend. - Updated embedding computation to support dynamic hidden sizes based on model configuration. - Added `pplx_mlx_config` function to retrieve model configuration details. - Updated README.md - Added .gitignore | |||||
| 2026-03-28 | Init | Onur Gündüz | 1 | -0/+1247 | |
