Repository navigation

Release v0.2.19 - #188

Merged

m96-chan merged 21 commits into

mainfrom

feature/v0.2.19

Jan 1, 2026

Merged

Release v0.2.19#188
m96-chan merged 21 commits into
mainfrom
feature/v0.2.19

Commits on Dec 31, 2025
- feat: add cuBLAS dynamic loader and C++ kernel profiler (#134, #150)
  
  m96-chan
  and
  claude
  committed
  3001123
- feat(llm): add lazy model loading with streaming strategies (#159)
  
  m96-chan
  and
  claude
  committed
  7d7de55
- fix(lint): resolve ruff B027 and UP037 errors in streaming.py
  
  m96-chan
  and
  claude
  committed
  aa8fd7e
- style: apply ruff format to streaming.py
  
  m96-chan
  and
  claude
  committed
  c82155a
- fix(lint): resolve ruff errors in profiling module
  
  m96-chan
  and
  claude
  committed
  a469a97
- fix(tests): add skip markers for profiling tests requiring CUDA
  
  m96-chan
  and
  claude
  committed
  204a38b
Commits on Jan 1, 2026
- feat(diffusion): add image generation module for SD3, Flux, PixArt (#177)
  
  m96-chan
  and
  claude
  committed
  521f908
- fix(diffusion): resolve mypy type errors in text encoders
  
  m96-chan
  and
  claude
  committed
  905c181
- feat(diffusion): add native CUDA kernels for image generation ops
  
  m96-chan
  and
  claude
  committed
  fb6b619
- fix(diffusion): fix PixArt-Sigma model loading and inference
  
  m96-chan
  and
  claude
  committed
  8a0c247
- feat(diffusion): add HuggingFace T5 encoder with sharded safetensors support
  
  m96-chan
  and
  claude
  committed
  ffeb2f8
- feat(diffusion): add batched_matmul loop fallback for SM120
  
  m96-chan
  and
  claude
  committed
  865619c
- feat(diffusion): add FLUX.1 transformer implementation
  
  m96-chan
  and
  claude
  committed
  dc0de77
- fix(lint): remove unused variables in DiT and FLUX models
  
  m96-chan
  and
  claude
  committed
  43e5756
- fix(cmake): remove orphaned #endif in diffusion kernels
  
  m96-chan
  and
  claude
  committed
  5006afa
- fix(cmake): use nbytes() instead of size_bytes() in diffusion.inl
  
  m96-chan
  and
  claude
  committed
  aa61015
- fix(cmake): use device_memset wrapper instead of cudaMemset
  
  m96-chan
  and
  claude
  committed
  5a8a98c
- docs: update README for v0.2.19
  
  m96-chan
  and
  claude
  committed
  502fe48
- docs: expand v0.2.19 release notes
  
  m96-chan
  and
  claude
  committed
  28c5ab7
- chore: bump version to 0.2.19
  
  m96-chan
  and
  claude
  committed
  4df5b09
- Merge main into feature/v0.2.19
  m96-chan
  committed
  ba68c8a

Commits

Commits on Dec 31, 2025

Commits on Jan 1, 2026