Update configs.py
Browse files- configs.py +13 -4
configs.py
CHANGED
@@ -2,7 +2,7 @@ import os
|
|
2 |
|
3 |
|
4 |
dataset_info = [
|
5 |
-
{'name': '
|
6 |
{'name': 'Factual Recall', 'hf_repo': 'azhx/counterfact-filtered-gptj6b', 'text_col': 'subject+predicate',
|
7 |
'filter': lambda x: x['label'] == 1},
|
8 |
# {'name': 'Physical Understanding', 'hf_repo': 'piqa', 'text_col': 'goal'},
|
@@ -11,7 +11,7 @@ dataset_info = [
|
|
11 |
|
12 |
|
13 |
model_info = {
|
14 |
-
'LLAMA2-7B': dict(model_path='meta-llama/Llama-2-7b-chat-hf',
|
15 |
original_prompt_template='<s>{prompt}',
|
16 |
interpretation_prompt_template='<s>[INST] [X] [/INST] {prompt}',
|
17 |
), # , load_in_8bit=True
|
@@ -19,8 +19,17 @@ model_info = {
|
|
19 |
# 'Gemma-2B': dict(model_path='google/gemma-2b', device_map='cpu', token=os.environ['hf_token'],
|
20 |
# original_prompt_template='<bos>{prompt}',
|
21 |
# interpretation_prompt_template='<bos>User: [X]\n\nAnswer: {prompt}',
|
22 |
-
# ),
|
23 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
24 |
'Mistral-7B Instruct': dict(model_path='mistralai/Mistral-7B-Instruct-v0.2', device_map='cpu',
|
25 |
original_prompt_template='<s>{prompt}',
|
26 |
interpretation_prompt_template='<s>[INST] [X] [/INST] {prompt}',
|
|
|
2 |
|
3 |
|
4 |
dataset_info = [
|
5 |
+
{'name': 'Common Sense', 'hf_repo': 'tau/commonsense_qa', 'text_col': 'question'},
|
6 |
{'name': 'Factual Recall', 'hf_repo': 'azhx/counterfact-filtered-gptj6b', 'text_col': 'subject+predicate',
|
7 |
'filter': lambda x: x['label'] == 1},
|
8 |
# {'name': 'Physical Understanding', 'hf_repo': 'piqa', 'text_col': 'goal'},
|
|
|
11 |
|
12 |
|
13 |
model_info = {
|
14 |
+
'LLAMA2-7B': dict(model_path='meta-llama/Llama-2-7b-chat-hf', token=os.environ['hf_token'],
|
15 |
original_prompt_template='<s>{prompt}',
|
16 |
interpretation_prompt_template='<s>[INST] [X] [/INST] {prompt}',
|
17 |
), # , load_in_8bit=True
|
|
|
19 |
# 'Gemma-2B': dict(model_path='google/gemma-2b', device_map='cpu', token=os.environ['hf_token'],
|
20 |
# original_prompt_template='<bos>{prompt}',
|
21 |
# interpretation_prompt_template='<bos>User: [X]\n\nAnswer: {prompt}',
|
22 |
+
# ),
|
23 |
+
'GPT-2 Small': dict(model_path='gpt2', original_prompt_template='{prompt}',
|
24 |
+
interpretation_prompt_template='User: [X]\n\nAnswer: {prompt}'),
|
25 |
+
'GPT-2 Medium': dict(model_path='gpt2-medium', original_prompt_template='{prompt}',
|
26 |
+
interpretation_prompt_template='User: [X]\n\nAnswer: {prompt}'),
|
27 |
+
'GPT-2 Large': dict(model_path='gpt2-large', original_prompt_template='{prompt}',
|
28 |
+
interpretation_prompt_template='User: [X]\n\nAnswer: {prompt}'),
|
29 |
+
'GPT-2 XL': dict(model_path='gpt2-xl', original_prompt_template='{prompt}',
|
30 |
+
interpretation_prompt_template='User: [X]\n\nAnswer: {prompt}'),
|
31 |
+
'GPT-J 6B': dict(model_path='EleutherAI/gpt-j-6b', original_prompt_template='{prompt}',
|
32 |
+
interpretation_prompt_template='User: [X]\n\nAnswer: {prompt}'),
|
33 |
'Mistral-7B Instruct': dict(model_path='mistralai/Mistral-7B-Instruct-v0.2', device_map='cpu',
|
34 |
original_prompt_template='<s>{prompt}',
|
35 |
interpretation_prompt_template='<s>[INST] [X] [/INST] {prompt}',
|