aboutsummaryrefslogtreecommitdiff
path: root/src/kernels.c (unfollow)
Commit message (Collapse)AuthorFilesLines
2026-06-16kernels: split kernels.c into gemm/norm/act/attn/rope by kindOnur Gündüz1-1271/+0
2026-06-16kernels: extract the persistent thread pool into threadpool.{c,h}Onur Gündüz1-207/+9
2026-06-16chore: move sources to src/ with proper names, drop qwen_ prefixOnur Gündüz1-86/+86
2026-06-16cpu: size default threads to cgroup cpu quota, not host core countOnur Gündüz1-2/+55
2026-06-16all: support BERT tanh-GeLU (gelu_new / gelu_pytorch_tanh)Onur Gündüz1-0/+9
2026-06-15kernels: add LayerNorm and exact-GeLU for the BERT familyOnur Gündüz1-0/+38
2026-06-15chore: run clang-formatOnur Gündüz1-221/+313
2026-06-14all: add support for Qwen3-Embedding + causal GQA attentionOnur Gündüz1-0/+152
2026-06-14kernels: delete dead audio/causal-LM leftovers and unused exportsOnur Gündüz1-497/+2
2026-06-14cpu: move BF16 widening scratch into the workspaceOnur Gündüz1-130/+5
2026-06-14kernels: drop asr_ prefix namingOnur Gündüz1-2/+2
2026-06-14kernels: fix OpenBLAS thread contention and numerical instabilityOnur Gündüz1-2/+30
2026-06-14kernels: parallelize large rmsnorm rowsOnur Gündüz1-5/+48
2026-06-14late: add packed maxsim scoringOnur Gündüz1-4/+8
2026-06-14kernels: vectorize bf16 widening on armOnur Gündüz1-0/+14
2026-06-14kernels: use sgemv for single-row f32 linearOnur Gündüz1-4/+10
2026-06-14kernels: add neon rope apply pathOnur Gündüz1-0/+21
2026-06-14cpu: reuse tiled attention scratch buffersOnur Gündüz1-12/+46
2026-06-14cpu: tile packed attention through blasOnur Gündüz1-54/+110
2026-06-14kernels,cpu: fuse tiny BF16 gate up projectionsOnur Gündüz1-0/+93
2026-06-14kernels,cpu: fuse small BF16 QKV projectionsOnur Gündüz1-15/+40
2026-06-14kernels: validate real BF16 artifactsOnur Gündüz1-0/+52
2026-06-14cpu: fuse CPU embedding hot pathsOnur Gündüz1-0/+26
2026-06-14kernels: avoid tiny packed attention dispatchOnur Gündüz1-1/+7
2026-06-14all: support batched embedding attentionOnur Gündüz1-0/+95
2026-03-28Refactor and enhance pplx_embed and pplx_embed_mlx for improved model supportOnur Gündüz1-5/+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
2026-03-28InitOnur Gündüz1-0/+1247