ridgerchu/matmulfreellm

Health v2 · maintenance only

Adoption not part of the score

3093 stars · 203 forks observed · 2026-09-27

What it is AI-extracted, prompt v1, taxonomy v1, 2026-08-29, confidence not recorded

A Python implementation of MatMul-Free LM, a language model architecture that eliminates matrix multiplication operations using ternary weights and linear attention. It is compatible with the Hugging Face Transformers library and provides pretrained models from 370M to 2.7B parameters.

Use cases

• train a language model without matrix multiplication
• run efficient LLM inference with ternary weights
• experiment with linear transformer architectures
• load a matmul-free model with huggingface AutoModel
• research hardware-efficient language model architectures
• compare scaling laws of matmul-free vs transformer models

When to choose

• you want to experiment with matmul-free or bit-linear LLM architectures
• you need a Hugging Face-compatible linear transformer implementation
• you are researching efficient language model scaling

When to avoid

• you need a production-ready general-purpose LLM framework
• you require standard transformer architectures with mature tooling
• you work outside the PyTorch/Triton ecosystem

Facets

library · maturity active

machine-learning llm-inference llm-training deep-learning large-language-models deep-learning machine-learning artificial-intelligence python matmul-free linear-transformer ternary-weights huggingface-transformers efficient-inference research gpu linux

Member repositories

For agents

markdown · JSON · MCP: product_card(name="ridgerchu/matmulfreellm")