Nicolas Hug

`len(dataloader)` in distributed setting is different with datapipes and with map-style datasets

2

In a distributed setting, `len(dataloader)` will return: - `len(dataset) // (batch_size * num_GPUs)` if `dataset` is a map-style dataset - `len(dataset) // batch_size` if `dataset` is a datapipe This discrepancy...

Update deprecated ubuntu image to fix CI

1

Fixes https://github.com/pytorch/hub/issues/297

Looks like the ubuntu image got deprecated https://app.circleci.com/pipelines/github/pytorch/hub/1801/workflows/94262fd5-9ba0-47e4-8856-d28e04d65b69/jobs/2242

Update md files to torchvision 0.14.1 and use new MultiWeights API

1

Closes https://github.com/pytorch/hub/issues/282 Closes https://github.com/pytorch/hub/issues/285

Update calls to torch.hub.* to use trust_repo=True. #281

1

Same as https://github.com/pytorch/hub/pull/281 but with appropriate timing Closes https://github.com/pytorch/hub/issues/276

Current CI Status

1

Empty PR just to check CI status

Meta issue to keep track of what we need to do: - Make sure this is green https://github.com/pytorch/hub/pull/225 - [ ] https://github.com/pytorch/hub/issues/276 - [ ] https://github.com/pytorch/hub/issues/282 - [ ] https://github.com/pytorch/hub/issues/285

Fix all `.md` files to set "trust_repo=True"

1

With https://github.com/pytorch/pytorch/pull/72060 now merged we'll need to update all the `.md` files. Likely tomorrow when the nightly release hits.

TODO close to the release date. We can use `./upgrade_torchvision.sh v0.13.0`

We'll also have to update the torchvision version in the `.md` files