b8189
📦 llama-cppView on GitHub →
✨ 2 features🐛 2 fixes🔧 3 symbols
Summary
This release refines the WebGPU backend by cleaning up parameter buffer pool management and job submission logic, including adjustments to buffer pool sizing and concurrency handling.
✨ New Features
- Improved WebGPU buffer pool management by allowing resizing and tracking submitted kernels.
- Increased initial size and maximum size (later reduced) for the parameter buffer pool in WebGPU.
🐛 Bug Fixes
- Fixed logic related to per-thread parameter buffer pool and job submission in WebGPU.
- Moved buffer pool growth outside of the lock for better concurrency.