Update configs.py
Browse files- configs.py +24 -18
configs.py
CHANGED
|
@@ -27,22 +27,29 @@ model_info = {
|
|
| 27 |
original_prompt_template='<s>{prompt}',
|
| 28 |
interpretation_prompt_template='<s>[INST] [X] [/INST] {prompt}',
|
| 29 |
layers_format=llama_layers_format),
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 30 |
'Mixtral 8x7B Instruct (Experimental)': dict(model_path='TheBloke/Mixtral-8x7B-Instruct-v0.1-AWQ',
|
| 31 |
token=os.environ['hf_token'], wait_with_hidden_states=True,
|
| 32 |
original_prompt_template='<s>{prompt}',
|
| 33 |
interpretation_prompt_template='<s>[INST] [X] [/INST] {prompt}',
|
| 34 |
layers_format=llama_layers_format
|
| 35 |
),
|
| 36 |
-
'
|
| 37 |
-
|
| 38 |
-
|
| 39 |
-
|
| 40 |
-
|
| 41 |
-
|
| 42 |
-
|
| 43 |
-
'GPT-2 Small': dict(model_path='gpt2', original_prompt_template='{prompt}',
|
| 44 |
-
interpretation_prompt_template='User: [X]\n\nAnswer: {prompt}',
|
| 45 |
-
layers_format=gpt_layers_format),
|
| 46 |
# 'GPT-2 Medium': dict(model_path='gpt2-medium', original_prompt_template='{prompt}',
|
| 47 |
# interpretation_prompt_template='User: [X]\n\nAnswer: {prompt}',
|
| 48 |
# layers_format=gpt_layers_format),
|
|
@@ -52,14 +59,13 @@ model_info = {
|
|
| 52 |
# 'GPT-2 XL': dict(model_path='gpt2-xl', original_prompt_template='{prompt}',
|
| 53 |
# interpretation_prompt_template='User: [X]\n\nAnswer: {prompt}',
|
| 54 |
# layers_format=gpt_layers_format),
|
| 55 |
-
'
|
| 56 |
-
|
| 57 |
-
|
| 58 |
-
|
| 59 |
-
|
| 60 |
-
|
| 61 |
-
|
| 62 |
-
|
| 63 |
# 'Gemma-2B': dict(model_path='google/gemma-2b', device_map='cpu', token=os.environ['hf_token'],
|
| 64 |
# original_prompt_template='<bos>{prompt}',
|
| 65 |
# interpretation_prompt_template='<bos>User: [X]\n\nAnswer: {prompt}',
|
|
|
|
| 27 |
original_prompt_template='<s>{prompt}',
|
| 28 |
interpretation_prompt_template='<s>[INST] [X] [/INST] {prompt}',
|
| 29 |
layers_format=llama_layers_format),
|
| 30 |
+
'GPT-J 6B': dict(model_path='EleutherAI/gpt-j-6b', original_prompt_template='{prompt}',
|
| 31 |
+
interpretation_prompt_template='User: [X]\n\nAnswer: {prompt}',
|
| 32 |
+
layers_format=gpt_layers_format),
|
| 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}',
|
| 36 |
+
layers_format=llama_layers_format),
|
| 37 |
+
'GPT-2 Small': dict(model_path='gpt2', original_prompt_template='{prompt}',
|
| 38 |
+
interpretation_prompt_template='User: [X]\n\nAnswer: {prompt}',
|
| 39 |
+
layers_format=gpt_layers_format),
|
| 40 |
'Mixtral 8x7B Instruct (Experimental)': dict(model_path='TheBloke/Mixtral-8x7B-Instruct-v0.1-AWQ',
|
| 41 |
token=os.environ['hf_token'], wait_with_hidden_states=True,
|
| 42 |
original_prompt_template='<s>{prompt}',
|
| 43 |
interpretation_prompt_template='<s>[INST] [X] [/INST] {prompt}',
|
| 44 |
layers_format=llama_layers_format
|
| 45 |
),
|
| 46 |
+
'Wizard Vicuna 30B Uncensored (Experimental)': dict(model_path='TheBloke/Wizard-Vicuna-30B-Uncensored-GPTQ',
|
| 47 |
+
token=os.environ['hf_token'],
|
| 48 |
+
wait_with_hidden_states=True, dont_cuda=True, device_map='cuda',
|
| 49 |
+
original_prompt_template='<s>USER: {prompt}',
|
| 50 |
+
interpretation_prompt_template='<s>USER: [X] ASSISTANT: {prompt}',
|
| 51 |
+
layers_format=llama_layers_format
|
| 52 |
+
),
|
|
|
|
|
|
|
|
|
|
| 53 |
# 'GPT-2 Medium': dict(model_path='gpt2-medium', original_prompt_template='{prompt}',
|
| 54 |
# interpretation_prompt_template='User: [X]\n\nAnswer: {prompt}',
|
| 55 |
# layers_format=gpt_layers_format),
|
|
|
|
| 59 |
# 'GPT-2 XL': dict(model_path='gpt2-xl', original_prompt_template='{prompt}',
|
| 60 |
# interpretation_prompt_template='User: [X]\n\nAnswer: {prompt}',
|
| 61 |
# layers_format=gpt_layers_format),
|
| 62 |
+
# 'CodeLLAMA 70B Instruct (Experimental)': dict(model_path='TheBloke/CodeLlama-70B-Instruct-GPTQ',
|
| 63 |
+
# token=os.environ['hf_token'],
|
| 64 |
+
# wait_with_hidden_states=True, dont_cuda=True, device_map='cuda', # disable_exllama=True,
|
| 65 |
+
# original_prompt_template='<s>{prompt}',
|
| 66 |
+
# interpretation_prompt_template='<s>[INST] [X] [/INST] {prompt}',
|
| 67 |
+
# layers_format=llama_layers_format
|
| 68 |
+
# ),
|
|
|
|
| 69 |
# 'Gemma-2B': dict(model_path='google/gemma-2b', device_map='cpu', token=os.environ['hf_token'],
|
| 70 |
# original_prompt_template='<bos>{prompt}',
|
| 71 |
# interpretation_prompt_template='<bos>User: [X]\n\nAnswer: {prompt}',
|