File size: 5,644 Bytes
81e81d9
1f6dcbd
 
81e81d9
9a230a0
 
 
81e81d9
26e7a03
7673f3b
26e7a03
 
 
 
 
 
 
 
7673f3b
9a230a0
 
a6550e6
 
25fb83c
b76e9de
25fb83c
 
e8236ff
 
e048966
 
 
 
 
 
 
 
 
 
6dbcb5e
 
 
 
 
 
 
 
 
 
 
 
 
e8236ff
 
 
 
 
 
 
 
 
e048966
 
 
 
 
 
 
9a230a0
 
 
 
 
26e7a03
 
 
 
 
 
 
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
import os
import torch


llama_layers_format = 'model.layers.{k}'
gpt_layers_format = 'transformer.h.{k}'


dataset_info = [
                {'name': 'Common Sense', 'hf_repo': 'tau/commonsense_qa', 'text_col': 'question'},
                {'name': 'Factual Recall', 'hf_repo': 'azhx/counterfact-filtered-gptj6b', 'text_col': 'subject+predicate', 
                 'filter': lambda x: x['label'] == 1},
                # {'name': 'Physical Understanding', 'hf_repo': 'piqa', 'text_col': 'goal'},
                {'name': 'Social Reasoning', 'hf_repo': 'ProlificAI/social-reasoning-rlhf', 'text_col': 'question'}
               ]


model_info = {
    'LLAMA2-7B': dict(model_path='meta-llama/Llama-2-7b-chat-hf', token=os.environ['hf_token'], 
                      original_prompt_template='<s>{prompt}',
                      interpretation_prompt_template='<s>[INST] [X] [/INST] {prompt}',
                      layers_format=llama_layers_format),
    'LLAMA2-13B': dict(model_path='meta-llama/Llama-2-13b-chat-hf', 
                      token=os.environ['hf_token'], torch_dtype=torch.float16,
                      wait_with_hidden_states=True,
                    # device_map='auto', max_memory={0: "15GB", 1: "30GB"},  dont_cuda=True, # load_in_8bit=True, 
                      original_prompt_template='<s>{prompt}',
                      interpretation_prompt_template='<s>[INST] [X] [/INST] {prompt}',
                      layers_format=llama_layers_format), 
    'GPT-J 6B': dict(model_path='EleutherAI/gpt-j-6b', original_prompt_template='{prompt}', 
                     interpretation_prompt_template='User: [X]\n\nAnswer: {prompt}',
                     layers_format=gpt_layers_format),
    'Mistral-7B Instruct': dict(model_path='mistralai/Mistral-7B-Instruct-v0.2', device_map='cpu', 
                                original_prompt_template='<s>{prompt}',
                                interpretation_prompt_template='<s>[INST] [X] [/INST] {prompt}',
                                layers_format=llama_layers_format),
    'GPT-2 Small': dict(model_path='gpt2', original_prompt_template='{prompt}', 
                     interpretation_prompt_template='User: [X]\n\nAnswer: {prompt}',
                     layers_format=gpt_layers_format),
    # 'Mixtral 8x7B Instruct (Experimental)': dict(model_path='TheBloke/Mixtral-8x7B-Instruct-v0.1-AWQ',
    #                               token=os.environ['hf_token'], wait_with_hidden_states=True,
    #                               original_prompt_template='<s>{prompt}',
    #                               interpretation_prompt_template='<s>[INST] [X] [/INST] {prompt}',
    #                               layers_format=llama_layers_format
    #                              ),
    # 'Wizard Vicuna 30B Uncensored (Experimental)': dict(model_path='TheBloke/Wizard-Vicuna-30B-Uncensored-GPTQ',
    #                                                     token=os.environ['hf_token'],
    #                                                     wait_with_hidden_states=True, dont_cuda=True, device_map='cuda', 
    #                                                     original_prompt_template='<s>USER: {prompt}',
    #                                                     interpretation_prompt_template='<s>USER: [X] ASSISTANT: {prompt}',
    #                                                     layers_format=llama_layers_format
    #                                                    ),
    # 'GPT-2 Medium': dict(model_path='gpt2-medium', original_prompt_template='{prompt}', 
    #                  interpretation_prompt_template='User: [X]\n\nAnswer: {prompt}',
    #                  layers_format=gpt_layers_format),
    # 'GPT-2 Large': dict(model_path='gpt2-large', original_prompt_template='{prompt}', 
    #                  interpretation_prompt_template='User: [X]\n\nAnswer: {prompt}',
    #                  layers_format=gpt_layers_format),
    # 'GPT-2 XL': dict(model_path='gpt2-xl', original_prompt_template='{prompt}', 
    #                  interpretation_prompt_template='User: [X]\n\nAnswer: {prompt}',
    #                  layers_format=gpt_layers_format),
    # 'CodeLLAMA 70B Instruct (Experimental)': dict(model_path='TheBloke/CodeLlama-70B-Instruct-GPTQ',
    #                                token=os.environ['hf_token'],
    #                                wait_with_hidden_states=True, dont_cuda=True, device_map='cuda', # disable_exllama=True,
    #                                original_prompt_template='<s>{prompt}',
    #                                interpretation_prompt_template='<s>[INST] [X] [/INST] {prompt}',
    #                                layers_format=llama_layers_format
    #                               ),
    # 'Gemma-2B': dict(model_path='google/gemma-2b', device_map='cpu', token=os.environ['hf_token'],
    #                         original_prompt_template='<bos>{prompt}',
    #                         interpretation_prompt_template='<bos>User: [X]\n\nAnswer: {prompt}',
    #                        ),   

    # 'TheBloke/Mistral-7B-Instruct-v0.2-GGUF': dict(model_file='mistral-7b-instruct-v0.2.Q5_K_S.gguf', 
    #                                                tokenizer='mistralai/Mistral-7B-Instruct-v0.2',
    #                                                model_type='llama', hf=True, ctransformers=True,
    #                                                original_prompt_template='<s>[INST] {prompt} [/INST]',
    #                                                interpretation_prompt_template='<s>[INST] [X] [/INST] {prompt}',
    #                                               )
        }