FP8-LM: Training FP8 Large Language Models — mentioned by YetAnotherNick

Sun, 06 Apr 2025 04:19:18 GMT

It's not just scale. Even for single GPU, it is hard to acheive 2x speed improvement as the GPU specs states. Even NVIDIA's own Tensor Engine acheives 28% extra FLOP/s[1]. [1]: https://arxiv.org/pdf/2310.18313

Paper and discussion history. A mention is not an endorsement.

FP8-LM: Training FP8 Large Language Models — HN submission

Mon, 30 Oct 2023 22:53:10 GMT

Training LLMs using 8-bit numbers

1 point; 0 discussion replies on HN. Paper and source history. A submission is not an endorsement.

FP8-LM: Training FP8 Large Language Models · HackerPapers discussions

FP8-LM: Training FP8 Large Language Models — mentioned by YetAnotherNick

FP8-LM: Training FP8 Large Language Models — HN submission