Bump 3.0.3 (#2277)

* various fixes along v3.0.3

CHANGELOG.md

@@ -3,6 +3,12 @@
 
 
 ## [Unreleased]
+## [3.0.3](https://github.com/OpenNMT/OpenNMT-py/tree/3.0.3) (2022-12-16)
+* fix loss normalization when using accum or nb GPU > 1
+* use native CrossEntropyLoss with Label Smoothing. reported loss/ppl impacted by LS
+* fix long-time coverage loss bug thanks Sanghyuk-Choi
+* fix detok at scoring / fix tokenization Subword_nmt + Sentencepiece
+* various small bugs fixed
 
 ## [3.0.2](https://github.com/OpenNMT/OpenNMT-py/tree/3.0.2) (2022-12-07)
 * pyonmttok.Vocab is now pickable. dataloader switched to spawn. (MacOS/Windows compatible)

onmt/__init__.py

@@ -15,4 +15,4 @@
 __all__ = [onmt.inputters, onmt.encoders, onmt.decoders, onmt.models,
            onmt.utils, onmt.modules, "Trainer"]
 
-__version__ = "3.0.2"
+__version__ = "3.0.3"

onmt/encoders/rnn_encoder.py

@@ -69,7 +69,8 @@ def forward(self, src, src_len=None):
         if src_len is not None and not self.no_pack_padded_seq:
             # src lengths data is wrapped inside a Tensor.
             src_len_list = src_len.view(-1).tolist()
-            packed_emb = pack(emb, src_len_list, batch_first=True)
+            packed_emb = pack(emb, src_len_list, batch_first=True,
+                              enforce_sorted=False)
 
         enc_out, enc_final_hs = self.rnn(packed_emb)
 

onmt/inputters/text_corpus.py

@@ -253,7 +253,7 @@ def save_transformed_sample(opts, transforms, n_sample=3):
                 open(dest_base + ".tgt", 'w', encoding="utf-8") as f_tgt:
             for i, item in enumerate(c_iter):
                 maybe_example = process(CorpusTask.TRAIN, [item])
-                if len(maybe_example) > 0:
+                if maybe_example is not None:
                     maybe_example = maybe_example[0]
                     src_line, tgt_line = (maybe_example['src']['src'],
                                           maybe_example['tgt']['tgt'])

onmt/inputters/text_utils.py

@@ -9,7 +9,7 @@
 def text_sort_key(ex):
     """Sort using the number of tokens in the sequence."""
     if ex['tgt']:
-        return len(ex['src']['src_ids']), len(ex['tgt']['tgt_ids'])
+        return max(len(ex['src']['src_ids']), len(ex['tgt']['tgt_ids']))
     return len(ex['src']['src_ids'])
 
 

onmt/utils/optimizers.py

@@ -58,7 +58,7 @@ def build_torch_optimizer(model, opt):
             params,
             lr=opt.learning_rate,
             betas=betas,
-            eps=1e-9)
+            eps=1e-8)
     elif opt.optim == 'sparseadam':
         dense = []
         sparse = []

onmt/utils/scoring_utils.py

@@ -139,10 +139,8 @@ def translate(self, model, transformed_batches, gpu_rank, step, mode):
             _, preds_ = translator._translate(
                         infer_iter, transform=self.transform)
             preds += preds_
-            with open("x", "a") as f:
-                f.write(str(len(preds_)))
 
-        # detokenize refs
+        # apply_reverse refs
         if self.transforms:
             texts_ref = [self.transform.apply_reverse(raw_ref)
                          for raw_ref in raw_refs]

setup.py

@@ -11,7 +11,7 @@
     description='A python implementation of OpenNMT',
     long_description=long_description,
     long_description_content_type='text/markdown',
-    version='3.0.2',
+    version='3.0.3',
     packages=find_packages(),
     project_urls={
         "Documentation": "http://opennmt.net/OpenNMT-py/",