Rename generation -> extraction everywhere

.vscode/launch.json

@@ -23,7 +23,7 @@
  "name": "Python: Generation Cuda",
  "type": "python",
  "request": "launch",
- "module": "elk.generation_main",
+ "module": "elk.extraction_main",
  "console": "integratedTerminal",
  "justMyCode": true,
  "args" : [

README.md

@@ -27,12 +27,12 @@ Furthermore:
 1. To generate the hidden states for one model `mdl` and all datasets, `cd elk` and then run
 
 ```bash
-python generation_main.py --model deberta-v2-xxlarge-mnli --datasets imdb --prefix normal --device cuda --num-data 1000
+python extraction_main.py --model deberta-v2-xxlarge-mnli --datasets imdb --prefix normal --device cuda --num-data 1000
 ```
 
 To test `deberta-v2-xxlarge-mnli` with the misleading prefix, and only the `imdb` and `amazon-polarity` datasets, while printing extra information, run:
 
-The name of prefix can be found in `./utils_generation/construct_prompts.py`. This command will save hidden states to `generation_results` and will save zero-shot accuracy to `generation_results/generation_results.csv`.
+The name of prefix can be found in `./extraction/construct_prompts.py`. This command will save hidden states to `extraction_results` and will save zero-shot accuracy to `extraction_results/extraction_results.csv`.
 
 1. To train a ccs model and a logistic regression model
 

elk/default_config.json

@@ -32,7 +32,7 @@
  "dadisincorrect",
  "teachernoimitate"
  ],
- "models-layer-num": {
+ "models_layer_num": {
  "t5-11b": 25,
  "unifiedqa-t5-11b": 25,
  "T0pp": 25,

elk/utils_generation/construct_prompts.py → elk/extraction/construct_prompts.py

@@ -312,12 +312,12 @@ def __init__(self, set_name):
  self.module = None
  else:
  self.nomodule = False
- from utils_generation.load_utils import getLoadName
+ from extraction.load_utils import getLoadName
 
  self.module = DatasetTemplates(*getLoadName(set_name))
 
  def getGlobalPromptsNum(set_name_list):
- from utils_generation.load_utils import getLoadName
+ from extraction.load_utils import getLoadName
 
  res = []
  for set_name in set_name_list:

elk/utils_generation/generation.py → elk/extraction/extraction.py

@@ -101,7 +101,7 @@ def create_hiddenstates(model, tokenizer, name_to_dataframe, args):
  """
  with torch.no_grad():
  for name, dataframe in name_to_dataframe.items():
- # This part corresponds to hidden states generation
+ # This part corresponds to hidden states extraction
  hidden_states = calculate_hidden_state(
  args, model, tokenizer, dataframe, args.model
  )

elk/utils_generation/parser.py → elk/extraction/parser.py

@@ -41,14 +41,14 @@ def get_args():
 
  assert args.model in models, NotImplementedError(
  "You use model {}, but it's not . For any new model, please make sure you"
- " implement the code in `load_utils` and `generation`, and then it in"
+ " implement the code in `load_utils` and `extraction`, and then it in"
  " `parser.py`".format(args.model)
  )
 
  for prefix in args.prefix:
  assert prefix in prefix, NotImplementedError(
  "Invalid prefix name {}. Please check your prefix name. To add new prefix,"
- " please mofidy `utils_generation/prompts.json` \
+ " please mofidy `extraction/prompts.json` \
  and new prefix in {}.json.".format(
  prefix, default_config_path
  )
@@ -157,7 +157,7 @@ def get_parser():
  default=["normal"],
  help=(
  "The name of prefix added before the question. normal means no index. You"
- " can go to `utils_generation/prompts.json` to add new prompt."
+ " can go to `extraction/prompts.json` to add new prompt."
  ),
  )
  parser.add_argument(
@@ -187,7 +187,7 @@ def get_parser():
  help="The indices of prompt you want to use.",
  )
 
- # generation & zero-shot accuracy calculation
+ # extraction & zero-shot accuracy calculation
  parser.add_argument(
  "--cal-zeroshot",
  type=int,
@@ -252,7 +252,7 @@ def get_parser():
  parser.add_argument(
  "--save-base-dir",
  type=Path,
- default="generation_results",
+ default="extraction_results",
  help="The base dir where you want to save the directories of hidden states.",
  )
  parser.add_argument(

elk/generation_main.py → elk/extraction_main.py

@@ -1,12 +1,12 @@
 import time
-from elk.utils_generation.parser import get_args
-from elk.utils_generation.load_utils import (
+from elk.extraction.parser import get_args
+from elk.extraction.load_utils import (
  load_model,
  put_model_on_device,
  load_tokenizer,
  load_datasets,
 )
-from elk.utils_generation.generation import create_records, create_hiddenstates
+from elk.extraction.extraction import create_records, create_hiddenstates
 from tqdm import tqdm
 
 if __name__ == "__main__":

elk/tests/utils_generation/test_gen_parser.py

@@ -1,4 +1,4 @@
-from elk.utils_generation.parser import get_parser
+from elk.extraction.parser import get_parser
 
 
 def test_args_no_underscores():

elk/utils_evaluation/parser.py

@@ -55,7 +55,7 @@ def get_parser(datasets, models, prefix):
  parser.add_argument(
  "--hidden-states-directory",
  type=Path,
- default="generation_results",
+ default="extraction_results",
  help="Where the hidden states and zero-shot accuracy are loaded.",
  )
  parser.add_argument("--language-model-type", type=str, default="encoder")