v0.14.3
datalab-to/suryav0.14.3May 29, 2025by VikParuchuri
AI Summary
Optimized the library for older hardware and CPUs by fixing attention mechanisms and adding input truncation for predictable memory usage.
Key Highlights
- Fixed SDPA attention issues causing slowdowns on non-Flash Attention devices
- Truncated long inputs for predictable VRAM usage
- Unwrapped non-math sections
New Features
- Hardware optimization
- Memory management
Full Release Notes
## Performance improvements - Fixed an issue with SDPA attention that was causing slowdowns and high memory usage on any device that didn't support flash attention. - Truncate excessively long input text to get a predictable max VRAM usage ## Accuracy improvements - Unwrap non-math sections in math tags - Truncate repetitive text optionally ## What's Changed * Optimize Inference by @tarun-menta in https://github.com/VikParuchuri/surya/pull/377 * Unwrap math from individual digits, etc by @VikParuchuri in https://github.com/VikParuchuri/surya/pull/378 * Dev by @VikParuchuri in https://github.com/VikParuchuri/surya/pull/379 **Full Changelog**: https://github.com/VikParuchuri/surya/compare/v0.14.2...v0.14.3