rbawden commited on
Commit
a61edaf
1 Parent(s): 7dab9c4

Update pipeline.py

Browse files
Files changed (1) hide show
  1. pipeline.py +3 -3
pipeline.py CHANGED
@@ -6,7 +6,7 @@ import html.parser
6
  import unicodedata
7
  import sys, os, re
8
 
9
- class ReaccentPipeline(Pipeline):
10
 
11
  def __init__(self, beam_size=5, batch_size=32, **kwargs):
12
  self.beam_size = beam_size
@@ -153,7 +153,7 @@ class ReaccentPipeline(Pipeline):
153
  def normalise_text(list_sents, batch_size=32, beam_size=5):
154
  tokeniser = AutoTokenizer.from_pretrained("rbawden/modern_french_normalisation", use_auth_token=True)
155
  model = AutoModelForSeq2SeqLM.from_pretrained("rbawden/modern_french_normalisation", use_auth_token=True)
156
- normalisation_pipeline = ReaccentPipeline(model=model,
157
  tokenizer=tokeniser,
158
  batch_size=batch_size,
159
  beam_size=beam_size)
@@ -163,7 +163,7 @@ def normalise_text(list_sents, batch_size=32, beam_size=5):
163
  def normalise_from_stdin(batch_size=32, beam_size=5):
164
  tokeniser = AutoTokenizer.from_pretrained("rbawden/modern_french_normalisation", use_auth_token=True)
165
  model = AutoModelForSeq2SeqLM.from_pretrained("rbawden/modern_french_normalisation", use_auth_token=True)
166
- normalisation_pipeline = ReaccentPipeline(model=model,
167
  tokenizer=tokeniser,
168
  batch_size=batch_size,
169
  beam_size=beam_size)
 
6
  import unicodedata
7
  import sys, os, re
8
 
9
+ class NormalisationPipeline(Pipeline):
10
 
11
  def __init__(self, beam_size=5, batch_size=32, **kwargs):
12
  self.beam_size = beam_size
 
153
  def normalise_text(list_sents, batch_size=32, beam_size=5):
154
  tokeniser = AutoTokenizer.from_pretrained("rbawden/modern_french_normalisation", use_auth_token=True)
155
  model = AutoModelForSeq2SeqLM.from_pretrained("rbawden/modern_french_normalisation", use_auth_token=True)
156
+ normalisation_pipeline = NormalisationPipeline(model=model,
157
  tokenizer=tokeniser,
158
  batch_size=batch_size,
159
  beam_size=beam_size)
 
163
  def normalise_from_stdin(batch_size=32, beam_size=5):
164
  tokeniser = AutoTokenizer.from_pretrained("rbawden/modern_french_normalisation", use_auth_token=True)
165
  model = AutoModelForSeq2SeqLM.from_pretrained("rbawden/modern_french_normalisation", use_auth_token=True)
166
+ normalisation_pipeline = NormalisationPipeline(model=model,
167
  tokenizer=tokeniser,
168
  batch_size=batch_size,
169
  beam_size=beam_size)