Hugging Face's logo

kejian
/
cpsc-debug

English
Generated from Trainer
Model card Files
xet
Community
kejian commited on
Commit
56c3332
·
1 Parent(s): 9cabb4f

update model card README.md

Browse files
Files changed (1) hide show
  1. README.md +46 -7
README.md CHANGED
@@ -95,7 +95,7 @@ The following hyperparameters were used during training:
95
 
96
  # Full config
97
  {'dataset': {'conditional_training_config': {'aligned_prefix': '<|aligned|>',
98
- 'drop_token_fraction': 0.05,
99
  'fine_prefix': '<|fine|>',
100
  'misaligned_prefix': '<|misaligned|>',
101
  'substandard_prefix': '<|substandard|>',
@@ -143,7 +143,7 @@ The following hyperparameters were used during training:
143
  'tomekkorbak/detoxify-pile-chunk3-1850000-1900000',
144
  'tomekkorbak/detoxify-pile-chunk3-1900000-1950000'],
145
  'is_split_by_sentences': True},
146
- 'generation': {'force_call_on': [22888],
147
  'metrics_configs': [{}, {'n': 1}, {'n': 2}, {'n': 5}],
148
  'scenario_configs': [{'generate_kwargs': {'bad_words_ids': [[50257],
149
  [50258],
@@ -158,6 +158,45 @@ The following hyperparameters were used during training:
158
  'name': 'unconditional',
159
  'num_samples': 2560,
160
  'prefix': '<|aligned|>'},
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
161
  {'generate_kwargs': {'bad_words_ids': [[50257],
162
  [50258],
163
  [50259],
@@ -174,7 +213,7 @@ The following hyperparameters were used during training:
174
  'prompt_before_control': True,
175
  'prompts_path': 'resources/challenging_rtp.jsonl'}],
176
  'scorer_config': {'device': 'cuda:0'}},
177
- 'kl_gpt3_callback': {'force_call_on': [22888],
178
  'gpt3_kwargs': {'model_name': 'davinci'},
179
  'max_tokens': 64,
180
  'num_samples': 1024,
@@ -201,15 +240,15 @@ The following hyperparameters were used during training:
201
  'logging_first_step': True,
202
  'logging_steps': 500,
203
  'num_tokens': 2800000000.0,
204
- 'output_dir': 'training_output_2',
205
  'per_device_train_batch_size': 8,
206
  'push_to_hub': True,
207
  'remove_unused_columns': False,
208
- 'save_steps': 25354,
209
- 'save_strategy': 'steps',
210
  'seed': 42,
211
  'warmup_ratio': 0.01,
212
  'weight_decay': 0.1}}
213
 
214
  # Wandb URL:
215
- https://wandb.ai/kejian/uncategorized/runs/2296ywzg
 
95
 
96
  # Full config
97
  {'dataset': {'conditional_training_config': {'aligned_prefix': '<|aligned|>',
98
+ 'drop_token_fraction': 0.03,
99
  'fine_prefix': '<|fine|>',
100
  'misaligned_prefix': '<|misaligned|>',
101
  'substandard_prefix': '<|substandard|>',
 
143
  'tomekkorbak/detoxify-pile-chunk3-1850000-1900000',
144
  'tomekkorbak/detoxify-pile-chunk3-1900000-1950000'],
145
  'is_split_by_sentences': True},
146
+ 'generation': {'force_call_on': [21362],
147
  'metrics_configs': [{}, {'n': 1}, {'n': 2}, {'n': 5}],
148
  'scenario_configs': [{'generate_kwargs': {'bad_words_ids': [[50257],
149
  [50258],
 
158
  'name': 'unconditional',
159
  'num_samples': 2560,
160
  'prefix': '<|aligned|>'},
161
+ {'generate_kwargs': {'bad_words_ids': [[50257],
162
+ [50258],
163
+ [50259],
164
+ [50260]],
165
+ 'do_sample': True,
166
+ 'max_length': 128,
167
+ 'min_length': 10,
168
+ 'temperature': 0.7,
169
+ 'top_k': 0,
170
+ 'top_p': 0.9},
171
+ 'name': 'unconditional-fine',
172
+ 'num_samples': 512,
173
+ 'prefix': '<|fine|>'},
174
+ {'generate_kwargs': {'bad_words_ids': [[50257],
175
+ [50258],
176
+ [50259],
177
+ [50260]],
178
+ 'do_sample': True,
179
+ 'max_length': 128,
180
+ 'min_length': 10,
181
+ 'temperature': 0.7,
182
+ 'top_k': 0,
183
+ 'top_p': 0.9},
184
+ 'name': 'unconditional-substandard',
185
+ 'num_samples': 512,
186
+ 'prefix': '<|substandard|>'},
187
+ {'generate_kwargs': {'bad_words_ids': [[50257],
188
+ [50258],
189
+ [50259],
190
+ [50260]],
191
+ 'do_sample': True,
192
+ 'max_length': 128,
193
+ 'min_length': 10,
194
+ 'temperature': 0.7,
195
+ 'top_k': 0,
196
+ 'top_p': 0.9},
197
+ 'name': 'unconditional-misaligned',
198
+ 'num_samples': 512,
199
+ 'prefix': '<|misaligned|>'},
200
  {'generate_kwargs': {'bad_words_ids': [[50257],
201
  [50258],
202
  [50259],
 
213
  'prompt_before_control': True,
214
  'prompts_path': 'resources/challenging_rtp.jsonl'}],
215
  'scorer_config': {'device': 'cuda:0'}},
216
+ 'kl_gpt3_callback': {'force_call_on': [21362],
217
  'gpt3_kwargs': {'model_name': 'davinci'},
218
  'max_tokens': 64,
219
  'num_samples': 1024,
 
240
  'logging_first_step': True,
241
  'logging_steps': 500,
242
  'num_tokens': 2800000000.0,
243
+ 'output_dir': 'training_output_3',
244
  'per_device_train_batch_size': 8,
245
  'push_to_hub': True,
246
  'remove_unused_columns': False,
247
+ 'save_steps': 21362,
248
+ 'save_strategy': 'no',
249
  'seed': 42,
250
  'warmup_ratio': 0.01,
251
  'weight_decay': 0.1}}
252
 
253
  # Wandb URL:
254
+ https://wandb.ai/kejian/uncategorized/runs/3r45gxyx