Fix a bug with using GloVe 840B tokens for initialization.

cc85d411 · Stephen Roller · Myle Ott · 9dcee4c7 · cc85d411
Commit cc85d411 authored Jun 19, 2018 by Stephen Roller Committed by Myle Ott Jun 21, 2018
Show whitespace changes
Inline Side-by-side

Showing with 1 addition and 1 deletion

fairseq/utils.py fairseq/utils.py +1 -1

No files found.
--- a/fairseq/utils.py
+++ b/fairseq/utils.py
@@ -256,7 +256,7 @@ def parse_embedding(embed_path):
    with open(embed_path) as f_embed:
        next(f_embed)  # skip header
        for line in f_embed:
-            pieces = line.strip().split()
+            pieces = line.rstrip().split(" ")
            embed_dict[pieces[0]] = torch.Tensor([float(weight) for weight in pieces[1:]])
    return embed_dict