#triton on Hacker News

2 stories tagged triton, newest first.

Screenshot of github.com: Fp8 runs ~100 tflops faster when the kernel name has "cutlass" in it

Fp8 runs ~100 tflops faster when the kernel name has "cutlass" in it (github.com)

327 points by mmastrac 1 year ago | 143 comments

GitHub pull request in triton-lang/triton rewriting the Gluon attention tutorial kernel to be persistent, noting fp8 runs ~100 TFLOPS faster when the kernel name contains "cutlass".

triton gpu-kernels fp8 cutlass

Screenshot of github.com: Show HN: 80% faster, 50% less memory, 0% loss of accuracy Llama finetuning

Show HN: 80% faster, 50% less memory, 0% loss of accuracy Llama finetuning (github.com)

336 points by danielhanchen 2 years ago | 107 comments

GitHub repository for Unsloth, an open-source library for faster, lower-memory QLoRA/LoRA finetuning of Llama models using Triton kernels.

unsloth llama qlora llm-finetuning

#triton on Hacker News

Fp8 runs ~100 tflops faster when the kernel name has "cutlass" in it(github.com)

Show HN: 80% faster, 50% less memory, 0% loss of accuracy Llama finetuning(github.com)

Fp8 runs ~100 tflops faster when the kernel name has "cutlass" in it (github.com)

Show HN: 80% faster, 50% less memory, 0% loss of accuracy Llama finetuning (github.com)