SUMIN/HuggingFace_transformer
1
0
Fork 0
Code Issues Pull Requests Actions 7 Packages Projects Releases Wiki Activity

fix if use lang embeddings in tf xlm

Browse Source
This commit is contained in:
Patrick von Platen
2020-03-09 11:18:54 +01:00
parent fbd02d4693
commit 4620caa864
2 changed files with 2 additions and 2 deletions
Download Patch File Download Diff File Expand all files Collapse all files

2
src/transformers/modeling_tf_xlm.py
View File

@@ -408,7 +408,7 @@ class TFXLMMainLayer(tf.keras.layers.Layer):
inputs_embeds = self.embeddings(input_ids) inputs_embeds = self.embeddings(input_ids)
tensor = inputs_embeds + self.position_embeddings(position_ids) tensor = inputs_embeds + self.position_embeddings(position_ids)
if langs is not None and self.use_lang_emb: if langs is not None and self.use_lang_emb and self.n_langs > 1:
tensor = tensor + self.lang_embeddings(langs) tensor = tensor + self.lang_embeddings(langs)
if token_type_ids is not None: if token_type_ids is not None:
tensor = tensor + self.embeddings(token_type_ids) tensor = tensor + self.embeddings(token_type_ids)

2
tests/test_modeling_tf_xlm.py
View File

@@ -342,4 +342,4 @@ class TFXLMModelLanguageGenerationTest(unittest.TestCase):
] # the president the president the president the president the president the president the president the president the president the president ] # the president the president the president the president the president the president the president the president the president the president
# TODO(PVP): this and other input_ids I tried for generation give pretty bad results. Not sure why. Model might just not be made for auto-regressive inference # TODO(PVP): this and other input_ids I tried for generation give pretty bad results. Not sure why. Model might just not be made for auto-regressive inference
output_ids = model.generate(input_ids) output_ids = model.generate(input_ids)
self.assertListEqual(output_ids[0].tolist(), expected_output_ids, do_sample=False) self.assertListEqual(output_ids[0].numpy().tolist(), expected_output_ids, do_sample=False)