Gemlite: Towards Building Custom Low-Bit Fused CUDA Kernels (mobiusml.github.io) | HN Screenshots

YHacker Newsscreenshots

Screenshot of mobiusml.github.io: Gemlite: Towards Building Custom Low-Bit Fused CUDA Kernels

First screen, captured 2 years ago.

Looks similar

Screenshot of elliotarledge.com: Batmobile: 10-20x Faster CUDA Kernels for Equivariant Graph Neural Networks

Batmobile: 10-20x Faster CUDA Kernels for Equivariant Graph Neural Networks (elliotarledge.com)

83 points by ipnon 9 months ago | 11 comments

Screenshot of siboehm.com: How to optimize a CUDA matmul kernel for cuBLAS-like performance (2022)

How to optimize a CUDA matmul kernel for cuBLAS-like performance (2022)(siboehm.com)

86 points by mpweiher 2 years ago | 21 comments

Screenshot of publish.obsidian.md: We fine-tuned Llama 405B on AMD GPUs

We fine-tuned Llama 405B on AMD GPUs (publish.obsidian.md)

452 points by felarof 2 years ago | 87 comments

Felafax blog post describing how they fine-tuned Llama 3.1 405B on 8 AMD MI300X GPUs using JAX, covering parameter sharding and LoRA.

llama jax amd-mi300x fine-tuning

Screenshot of Hacker News post: Reasonable GPUs

Reasonable GPUs

27 points by frognumber 2 years ago | 20 comments

Screenshot of matt-rickard.com: Nvidia's CUDA Monopoly

Nvidia's CUDA Monopoly (matt-rickard.com)

53 points by ingve 3 years ago | 44 comments

Screenshot of baseten.co: Running GPT-OSS-120B at 500 tokens per second on Nvidia GPUs

Running GPT-OSS-120B at 500 tokens per second on Nvidia GPUs (baseten.co)

215 points by philipkiely 1 year ago | 142 comments