ridgerchu/matmulfreellm
A language model architecture that eliminates matrix multiplication operations using efficient ternary weights, available as Hugging Face models.
Not currently ranked — collecting fresh signals.
star history
MatMul-Free LM replaces standard matrix multiplication with more efficient operations, reducing computational cost while maintaining language modeling capability. The repository provides custom model implementations compatible with the Hugging Face Transformers library, along with pre-trained checkpoints at 370M, 1.3B, and 2.7B parameter scales. It uses Triton for kernel optimization and follows scaling laws demonstrating efficiency advantages over standard Transformer architectures.
Frequently asked
- What is ridgerchu/matmulfreellm?
- A language model architecture that eliminates matrix multiplication operations using efficient ternary weights, available as Hugging Face models.
- Is matmulfreellm open source?
- Yes — ridgerchu/matmulfreellm is open source, released under the Apache-2.0 license.
- What language is matmulfreellm written in?
- ridgerchu/matmulfreellm is primarily written in Python.
- How popular is matmulfreellm?
- ridgerchu/matmulfreellm has 3.1k stars on GitHub.
- Where can I find matmulfreellm?
- ridgerchu/matmulfreellm is on GitHub at https://github.com/ridgerchu/matmulfreellm.