SynthyraOpen source
Our Open Source Initiatives
We build in the open. These are the tools we use ourselves for protein analysis and chemical property prediction, released so other groups can run the same work without rebuilding the plumbing.
Community Impact
0
Total Stars
0
Total Forks
0
Contributors
0
Active Projects
FastPLMs
Efficient protein language models with Flash/Flex attention. Open-source effort to increase the efficiency of pretrained protein language models.
0
0
0
Key Features
Flash/Flex Attention: Replaces native attention for improved efficiency
Huggingface Integration: All models can be loaded via AutoModel from Huggingface transformers
Dataset Embedding: Embed entire datasets with no new code using embed_dataset
Optimized Processing: Sequences are sorted to reduce padding tokens for faster processing
Why FastPLMs?
Efficient PLM Handling: Optimized for loading, fine-tuning, and inference with large language models
Advanced Optimization: Built-in techniques for model quantization, pruning, and more
Scalability: Designed for handling large datasets and models
Community Driven: Enables researchers and developers to build sophisticated AI applications