asahi417 commited on
Commit
7b37d9b
·
1 Parent(s): 91e1d0a
experiments/model_finetuning_topic.py CHANGED
@@ -76,9 +76,6 @@ def main(
76
  best_model_path = pj(output_dir, "best_model")
77
 
78
  tokenizer = AutoTokenizer.from_pretrained(model)
79
- model = AutoModelForSequenceClassification.from_pretrained(
80
- model, id2label=ID2LABEL, label2id=LABEL2ID, num_labels=len(LABEL2ID), problem_type="multi_label_classification"
81
- )
82
  dataset = load_dataset(dataset, dataset_type)
83
  tokenized_datasets = dataset.map(
84
  lambda x: tokenizer(x["text"], padding="max_length", truncation=True, max_length=256), batched=True
@@ -150,15 +147,14 @@ def main(
150
  else:
151
  metric = {}
152
  for single_test in test_split:
153
- model = AutoModelForSequenceClassification.from_pretrained(
154
- best_model_path,
155
- num_labels=len(LABEL2ID),
156
- problem_type="multi_label_classification",
157
- id2label=ID2LABEL,
158
- label2id=LABEL2ID
159
- )
160
  trainer = Trainer(
161
- model=model,
 
 
 
 
 
 
162
  args=TrainingArguments(
163
  output_dir=output_dir,
164
  evaluation_strategy="no",
@@ -176,7 +172,7 @@ def main(
176
  if not skip_upload:
177
  logging.info("uploading to huggingface")
178
  model_organization = "tweettemposhift"
179
- model = AutoModelForSequenceClassification.from_pretrained(
180
  best_model_path,
181
  num_labels=len(LABEL2ID),
182
  problem_type="multi_label_classification",
@@ -184,7 +180,7 @@ def main(
184
  label2id=LABEL2ID
185
  )
186
  tokenizer = AutoTokenizer.from_pretrained(best_model_path)
187
- model.push_to_hub(f"{model_organization}/{model_alias}", use_auth_token=True)
188
  tokenizer.push_to_hub(f"{model_organization}/{model_alias}", use_auth_token=True)
189
  repo = Repository(model_alias, f"{model_organization}/{model_alias}")
190
  for i in glob(f"{best_model_path}/*"):
 
76
  best_model_path = pj(output_dir, "best_model")
77
 
78
  tokenizer = AutoTokenizer.from_pretrained(model)
 
 
 
79
  dataset = load_dataset(dataset, dataset_type)
80
  tokenized_datasets = dataset.map(
81
  lambda x: tokenizer(x["text"], padding="max_length", truncation=True, max_length=256), batched=True
 
147
  else:
148
  metric = {}
149
  for single_test in test_split:
 
 
 
 
 
 
 
150
  trainer = Trainer(
151
+ model=AutoModelForSequenceClassification.from_pretrained(
152
+ best_model_path,
153
+ num_labels=len(LABEL2ID),
154
+ problem_type="multi_label_classification",
155
+ id2label=ID2LABEL,
156
+ label2id=LABEL2ID
157
+ ),
158
  args=TrainingArguments(
159
  output_dir=output_dir,
160
  evaluation_strategy="no",
 
172
  if not skip_upload:
173
  logging.info("uploading to huggingface")
174
  model_organization = "tweettemposhift"
175
+ model_instance = AutoModelForSequenceClassification.from_pretrained(
176
  best_model_path,
177
  num_labels=len(LABEL2ID),
178
  problem_type="multi_label_classification",
 
180
  label2id=LABEL2ID
181
  )
182
  tokenizer = AutoTokenizer.from_pretrained(best_model_path)
183
+ model_instance.push_to_hub(f"{model_organization}/{model_alias}", use_auth_token=True)
184
  tokenizer.push_to_hub(f"{model_organization}/{model_alias}", use_auth_token=True)
185
  repo = Repository(model_alias, f"{model_organization}/{model_alias}")
186
  for i in glob(f"{best_model_path}/*"):