FP8 is ~100 tflops faster when the kernel name has "cutlass" in it (twitter.com) | HN Screenshots

YHacker Newsscreenshots

Screenshot of twitter.com: FP8 is ~100 tflops faster when the kernel name has "cutlass" in it

First screen, captured 1 year ago.

Looks similar

Screenshot of twitter.com: Cursor Composer 2 is just Kimi K2.5 with RL

Cursor Composer 2 is just Kimi K2.5 with RL (twitter.com)

199 points by mirzap 7 months ago | 94 comments

Screenshot of twitter.com: Justine creates llamafile. Cross platform single binary for any LLM weights

Justine creates llamafile. Cross platform single binary for any LLM weights (twitter.com)

10 points by klohto 2 years ago | 0 comments

Screenshot of twitter.com: A 94x speed improvement demonstrated using handwritten assembly

A 94x speed improvement demonstrated using handwritten assembly (twitter.com)

43 points by doener 2 years ago | 29 comments

Screenshot of twitter.com: OpenAI's latest repo has Claude as the third top contributor

OpenAI's latest repo has Claude as the third top contributor (twitter.com)

46 points by mirzap 7 months ago | 17 comments

Screenshot of twitter.com: One of the more painful things: many bool fields with byte sized bools

One of the more painful things: many bool fields with byte sized bools (twitter.com)

3 points by tosh 2 years ago | 2 comments

Screenshot of twitter.com: GateGPT: 56k tokens per second Transformer (KV cache) on FPGA at 80 MHz

GateGPT: 56k tokens per second Transformer (KV cache) on FPGA at 80 MHz (twitter.com)

27 points by laxmena 4 months ago | 8 comments