All pull requests

Pull requests

• Respect interpolation in the default training transform

  #1249·Sonica-B opened Nov 12, 2026·

  ·

• fix: allow pretrained downloads when Content-Length is missing

  #1248·kevin9327 opened Nov 12, 2026·

  ·
  1

• Freeze stem BatchNorm stats in ModifiedResNet.lock

  #1247·assass1nlike opened Nov 11, 2026·

  ·

• fix(transform): don't treat scalar mean/std of 0 as unset in image_transform

  #1246·davrens101 opened Nov 10, 2026Contributor·

  ·

• Model EMA working with DDP & FSDP

  #1245·rwightman opened Nov 10, 2026Collaborator·

  ·

• Support extra timm transform parameters

  #1181·igormp opened Jun 23, 2026·

  ·
  3

• Try out TIPSV2 in image-text mode, zero-shot seems to work

  #1166·rwightman opened May 27, 2026Collaborator·

  ·

• Support loading a model in a non-init-weights context

  #1105·bryant1410 opened Aug 6, 2025Contributor·

  ·
  6

• Save intermediate checkpoints when sampling without replacement (take 2)

  #535·gabrielilharco opened May 17, 2023Collaborator·

  ·

• [WIP] Allow for attention caching during CoCa generation

  #502·sramshetty opened Apr 20, 2023Contributor·2 tasks·

  ·

• feat: add meta device initialization for pretrained models, 5x faster load times

  #501·ErwannMillon opened Apr 19, 2023·

  ·
  5

• Batched CoCa Generation

  #498·sramshetty opened Apr 17, 2023Contributor·

  ·

• Add video support

  #430·iejMac opened Feb 15, 2023Contributor·

  ·
  11

• Interpolate positional embeddings for input images with larger sizes

  #416·EeroHeikkinen opened Feb 11, 2023·

  ·

• Stop trying to guess the length of webdataset datasets

  #339·gabrielilharco opened Jan 5, 2023Collaborator·

  ·
  3

• Add TextTextCLIP

  #323·lingjzhu opened Dec 27, 2022·9 tasks done·

  ·
  21

• support layer decay and different lr for text/visual encoder

  #268·Quan-Sun opened Nov 29, 2022Contributor·

  ·
  20

• Support distributed evaluation

  #176·mehdidc opened Sep 24, 2022Contributor·

  ·
  11

• Added ability to load pre-computed vectors from Pickle file

  #134·taylor-brown opened Jul 27, 2022·

  ·
  3

• Support a sampling strategy for multiple training datasets

  #107·zerovl opened Jun 9, 2022·

  ·
  5