Experimental implementations exploring feasibility, usefulness, and performance of dedicated q[X]ora kernels.
Being exploratory, this repository is not recommended for external use except by collaboratoring researchers.
Directory | Description |
---|---|
evals/ | Evaluation harnesses to measure quality effects of quantization |
exploratory/ | Scripts and notebooks for reproductions, experimentation |
kernels/ | Triton and CUDA kernel implementations |
models/ | Training and inference code for reference quantized models |
Collaborators welcome. Please reach out to us - @umerHA @austinvhuang if interested!