|
fix alignment when using uneven batches and left pad
|
Alexei Baevski
|
|
6 years ago |
|
Fix --prefix-size
|
Myle Ott
|
|
6 years ago |
|
make sure tensor used to index is cuda if on gpu
|
Alexei Baevski
|
|
6 years ago |
|
Remove src-padding from generation output
|
Myle Ott
|
|
6 years ago |
|
remove completed sentences from batch
|
Alexei Baevski
|
|
6 years ago |
|
Merge internal changes (#136)
|
Myle Ott
|
|
6 years ago |
|
Enforce upper-bound on maximum generation length (#121)
|
Myle Ott
|
|
6 years ago |
|
Allow more flexible pre-processing and generation (#227)
|
Sergey Edunov
|
|
6 years ago |
|
More fixes for recent PyTorch (incl. topk issue) (#113)
|
Myle Ott
|
|
6 years ago |
|
Refactor incremental generation to be more explicit and less magical (#222)
|
Myle Ott
|
|
6 years ago |
|
Add support to prefixes (#221)
|
Dario Pavllo
|
|
6 years ago |
|
More unit test fixes
|
Myle Ott
|
|
6 years ago |
|
fairseq-py goes distributed (#106)
|
Myle Ott
|
|
6 years ago |
|
Allow larger maxlen (fixes #100) (#101)
|
Myle Ott
|
|
6 years ago |
|
Better support for torch.no_grad (since volatile is deprecated)
|
Myle Ott
|
|
6 years ago |
|
Move normalization of model output (e.g., via LSM) into model definition
|
Myle Ott
|
|
6 years ago |
|
Fix warning about deprecated `volatile` kwarg for Variables
|
Myle Ott
|
|
6 years ago |
|
Add option to SequenceGenerator to retain dropout
|
Myle Ott
|
|
6 years ago |
|
Streamline data formatting utils
|
Myle Ott
|
|
6 years ago |
|
Add explicit dimension to softmax calls
|
Myle Ott
|
|
6 years ago |
|
Fix generation bug with large beam sizes (>50)
|
Myle Ott
|
|
6 years ago |
|
Don't call forward directly (prefer module(x) to module.forward(x))
|
Myle Ott
|
|
6 years ago |
|
Add LSTM
|
Myle Ott
|
|
6 years ago |
|
Revert `dim` in `F.softmax` for backwards compatibility
|
Myle Ott
|
|
6 years ago |
|
Add dim to F.softmax calls
|
Myle Ott
|
|
6 years ago |
|
Left pad source and right pad target
|
Myle Ott
|
|
6 years ago |
|
Support different max_source_positions and max_target_positions
|
Myle Ott
|
|
6 years ago |
|
Added -unkpen flag to generate.py following logic of Lua/Torch version
|
Michael Auli
|
|
6 years ago |
|
Refactor model definitions
|
Myle Ott
|
|
6 years ago |
|
Only consider EOS in beam search if it's among top-k candidates
|
Myle Ott
|
|
6 years ago |