Skip to content

pytorch-labs/float8_experimental

View on GitHub ↗Related repositories →

This repository contains the experimental PyTorch native float8 training UX

active 2023-11-172025-04-24 (UTC)

Complete coverage26,646 / 26,646 hourly files (100%) · 2 absent upstream2023-08-152026-08-29 (UTC)
Events
3.2K
Pushes
691
Pull requests
306
Issues
80
Stars
218
Forks
19

Activity over time

Daily event counts in the loaded window

Line chart, 525 days from 2023-11-17 to 2025-04-24. Pushes: 691 total, peak 63 in a day. Pull requests: 306 total, peak 26 in a day. Issues: 80 total, peak 27 in a day. Comments: 958 total, peak 36 in a day. Stars: 218 total, peak 26 in a day.

  • Pushes
  • Pull requests
  • Issues
  • Comments
  • Stars

Stars, PRs, issues and forks are under-captured in the later part of this window. GH Archive progressively stopped capturing non-push events during 2026 — −95% or worse by the end of the window. Every series here except Pushes fades for that reason, so a decline above reflects the archive, not this repository. Pushes stay reliable throughout, so read them, and the contributor counts derived from them, as the real signal. Data health has the measurements.

Top contributors

Pushes, PRs, issues, reviews and comments — stars and forks excluded, so this is contribution rather than popularity

ContributorContributionsPushesPRsComments
vkuzo777175104282
drisspg70620439221
facebook-github-bot39284102204
awgu3461652486
weifengpy1122747
wanchaol10334929
y-sq74271318
bdhirsh450320
alugorey220115
ani3009016
malfet9003
voznesenskym9004
yifuwang5003
msaroufim4001
yitzhaklevi4000
albanD4003
snarayan214003
zou35194002
cyang493021
mabeto5p3001

Recent activity

Latest issues, pull requests and releases

  • Issue comment#314vkuzo2024-07-30 16:44
    [RFC] Float8 Inference
  • Issue#314vkuzo2024-07-30 16:44
    [RFC] Float8 Inference
  • Issue comment#292vkuzo2024-07-30 15:53
    Float8Tensor.to_original_precision() returns wrong dtype
  • Issue#292vkuzo2024-07-30 15:53
    Float8Tensor.to_original_precision() returns wrong dtype
  • Issue#280vkuzo2024-07-30 15:53
    Docs should say what's the smallest model users will see a benefit for
  • Issue comment#280vkuzo2024-07-30 15:53
    Docs should say what's the smallest model users will see a benefit for
  • Issue comment#279vkuzo2024-07-30 15:52
    Expected trailing dimension of mat1 to be divisible by 16 but got mat1 shape
  • Issue#279vkuzo2024-07-30 15:52
    Expected trailing dimension of mat1 to be divisible by 16 but got mat1 shape
  • Issue#274vkuzo2024-07-30 15:50
    [QST] Dynamic Scaling
  • Issue#267vkuzo2024-07-30 15:48
    delayed scaling safety logic currently doesn't work with activation checkpointing
  • Issue comment#267vkuzo2024-07-30 15:48
    delayed scaling safety logic currently doesn't work with activation checkpointing
  • Issue comment#259vkuzo2024-07-30 15:47
    memory alignment issue in torch.compile mode
  • Issue#259vkuzo2024-07-30 15:47
    memory alignment issue in torch.compile mode
  • Issue#257vkuzo2024-07-30 15:46
    Float8Linear does not support autocast
  • Issue comment#257vkuzo2024-07-30 15:46
    Float8Linear does not support autocast
  • Issue comment#249vkuzo2024-07-30 15:45
    investigate cuda graphs + dynamic scaling leading to memory fragmentations
  • Issue#249vkuzo2024-07-30 15:45
    investigate cuda graphs + dynamic scaling leading to memory fragmentations
  • Issue comment#246vkuzo2024-07-30 15:45
    add configuration of precision for all 3 gemms
  • Issue#246vkuzo2024-07-30 15:45
    add configuration of precision for all 3 gemms
  • Issue comment#245vkuzo2024-07-30 15:44
    add option to use fast accumulation in the float8 matmul
  • Issue#245vkuzo2024-07-30 15:44
    add option to use fast accumulation in the float8 matmul
  • Issue comment#243vkuzo2024-07-30 15:44
    addmm implemented incorrectly
  • Issue#243vkuzo2024-07-30 15:44
    addmm implemented incorrectly
  • Issue comment#238vkuzo2024-07-30 15:43
    torch.inference_mode switches`aten.linear.default, this is not supported`
  • Issue#238vkuzo2024-07-30 15:43
    torch.inference_mode switches`aten.linear.default, this is not supported`

Totals cover only the window loaded into ClickHouse and count events, not GitHub's lifetime totals — 218 stars here means stars gained during the window, not the repo's star count.