Chunk Dataset API#26545

Closed

thiagocrepaldi wants to merge 21 commits intopytorch:masterfrom

thiagocrepaldi:thiagofc/chunk_dataset

Commits on Sep 3, 2019

Initial version
Thiago Crepaldi
committed
Refactor ChunkDataset API - Remove example sampler
Thiago Crepaldi
committed
ChunkDataset API v1
Thiago Crepaldi
committed
Remove batch_size from ChunkDataset API
Thiago Crepaldi
committed

Commits on Sep 9, 2019

Renames StrideSampler (DistributedChunkSampler) and refactor DistributedSampler
Thiago Crepaldi
committed

Commits on Sep 11, 2019

Improve DistributedChunkSampler to remove no-op loop
Thiago Crepaldi
committed

Commits on Sep 12, 2019

Revert DistributedSampler to original version
Thiago Crepaldi
committed
Change ChunkDataReader to return list(samples) instead of list(tuple(data, target))
Thiago Crepaldi
committed

Commits on Sep 13, 2019

Remove numpy dependency
Thiago Crepaldi
committed
Address Jaliya feedback
Thiago Crepaldi
committed

Commits on Sep 16, 2019

Replace `rank` by `global_worker_rank`
Thiago Crepaldi
committed
Remove duplicated ChunkDataReader exporting
Thiago Crepaldi
committed

Commits on Sep 17, 2019

Address Jaliya doc feedback
Thiago Crepaldi
committed

Commits on Sep 18, 2019

Add support to multiple workers per GPU on distributed scenarios
Thiago Crepaldi
committed
Add support to multiple workers per GPU on distributed scenarios
Thiago Crepaldi
committed
Improve doc
Thiago Crepaldi
committed

Commits on Sep 19, 2019

Address Xueyun feedback
Thiago Crepaldi
committed
Rename DistributedChunkSampler.global_worker_rank to .rank
Thiago Crepaldi
committed
Change chunk cache strategy and add epoch concept
Thiago Crepaldi
committed
Remove self._batch_size from ChunkDataset
Thiago Crepaldi
committed

Commits on Sep 20, 2019

More doc changes
Thiago Crepaldi
committed