app.py

import os
from copy import deepcopy
from functools import partial
import spaces
import gradio as gr
import torch
from ctransformers import AutoModelForCausalLM as CAutoModelForCausalLM
from transformers import AutoModelForCausalLM, AutoTokenizer
from interpret import InterpretationPrompt

MAX_PROMPT_TOKENS = 30

## info
model_info = {
    'meta-llama/Llama-2-7b-chat-hf': dict(device_map='cpu', token=os.environ['hf_token'], 
                                          original_prompt_template='<s>[INST] {prompt} [/INST]',
                                          interpretation_prompt_template='<s>[INST] [X] [/INST] {prompt}',
                                         ), # , load_in_8bit=True
    
    'google/gemma-2b': dict(device_map='cpu', token=os.environ['hf_token'],
                            original_prompt_template='<bos> {prompt}',
                            interpretation_prompt_template='<bos>User: [X]\n\nAnswer: {prompt}',
                           ),
    
    'mistralai/Mistral-7B-Instruct-v0.2': dict(device_map='cpu', 
                                               original_prompt_template='<s>[INST] {prompt} [/INST]',
                                               interpretation_prompt_template='<s>[INST] [X] [/INST] {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}',
                                                  )
        }


suggested_interpretation_prompts = ["Before responding, let me repeat the message you wrote:", 
                                    "Let me repeat the message:", "Sure, I'll summarize your message:"]


## functions
@spaces.GPU
def initialize_gpu():
    pass

def get_hidden_states(progress, raw_original_prompt):
    original_prompt = original_prompt_template.format(prompt=raw_original_prompt)
    model_inputs = tokenizer(original_prompt, add_special_tokens=False, return_tensors="pt").to(model.device)
    tokens = tokenizer.batch_decode(model_inputs.input_ids[0])
    outputs = model(**model_inputs, output_hidden_states=True, return_dict=True)
    hidden_states = torch.stack([h.squeeze(0).cpu().detach() for h in outputs.hidden_states], dim=0)
    token_btns = ([gr.Button(token, visible=True) for token in tokens] 
                  + [gr.Button('', visible=False) for _ in range(MAX_PROMPT_TOKENS - len(tokens))])
    return [hidden_states, *token_btns]


def run_interpretation(global_state, raw_interpretation_prompt, max_new_tokens, do_sample, 
                       temperature, top_k, top_p, repetition_penalty, length_penalty, i, 
                       num_beams=1, progress=gr.Progress()):

    interpreted_vectors = global_state[:, i]
    length_penalty = -length_penalty   # unintuitively, length_penalty > 0 will make sequences longer, so we negate it

    # generation parameters
    generation_kwargs = {
        'max_new_tokens': int(max_new_tokens),
        'do_sample': do_sample,
        'temperature': temperature,
        'top_k': int(top_k),
        'top_p': top_p,
        'repetition_penalty': repetition_penalty,
        'length_penalty': length_penalty,
        'num_beams': int(num_beams)
    }
    
    # create an InterpretationPrompt object from raw_interpretation_prompt (after putting it in the right template)
    interpretation_prompt = interpretation_prompt_template.format(prompt=raw_interpretation_prompt)
    interpretation_prompt = InterpretationPrompt(tokenizer, interpretation_prompt)

    # generate the interpretations
    generated = interpretation_prompt.generate(model, {0: interpreted_vectors}, k=3, **generation_kwargs)
    generation_texts = tokenizer.batch_decode(generated)
    return [gr.Textbox(text, visible=True, container=False) for text in generation_texts]


## main
torch.set_grad_enabled(False)
model_name = 'meta-llama/Llama-2-7b-chat-hf' # 'mistralai/Mistral-7B-Instruct-v0.2' #

# extract model info
model_args = deepcopy(model_info[model_name])
original_prompt_template = model_args.pop('original_prompt_template')
interpretation_prompt_template = model_args.pop('interpretation_prompt_template')
tokenizer_name = model_args.pop('tokenizer') if 'tokenizer' in model_args else model_name
use_ctransformers = model_args.pop('ctransformers', False)
AutoModelClass = CAutoModelForCausalLM if use_ctransformers else AutoModelForCausalLM

# get model
model = AutoModelClass.from_pretrained(model_name, **model_args)
tokenizer = AutoTokenizer.from_pretrained(tokenizer_name, token=os.environ['hf_token'])

# demo
json_output = gr.JSON()
css = '''
.bubble {
  border: 2px solid #000;
  border-radius: 10px;
  padding: 10px;
  margin-top: 10px;
  background: pink;
}
.bubble > textarea{
  border: none;
  background: pink;
}

'''


# '''
# .token_btn{
#   background-color: none;
#   background: none;
#   border: none;
#   padding: 0;
#   font: inherit;
#   cursor: pointer;
#   color: blue; /* default text color */
#   font-weight: bold;
# }

# .token_btn:hover {
#     color: red;
# }

# '''

with gr.Blocks(theme=gr.themes.Default(), css=css) as demo:
    global_state = gr.State([])
    with gr.Row():
        with gr.Column(scale=5):
            gr.Markdown('''
                # 😎 Self-Interpreting Models
                
                👾 **This space is a simple introduction to the emerging trend of models interpreting their _own hidden states_ in free form natural language**!! 👾
                This idea was explored in the paper **Patchscopes** ([Ghandeharioun et al., 2024](https://arxiv.org/abs/2401.06102)) and was later investigated further in **SelfIE** ([Chen et al., 2024](https://arxiv.org/abs/2403.10949)). 
                An honorary mention of **Speaking Probes** ([Dar, 2023](https://towardsdatascience.com/speaking-probes-self-interpreting-models-7a3dc6cb33d6) -- my own work!! 🥳) which was less mature but had the same idea in mind. 
                We will follow the SelfIE implementation in this space for concreteness. Patchscopes are so general that they encompass many other interpretation techniques too!!! 
                
                👾 **The idea is really simple: models are able to understand their own hidden states by nature!** 👾  
                If I give a model a prompt of the form ``User: [X] Assistant: Sure'll I'll repeat your message`` and replace ``[X]`` *during computation* with the hidden state we want to understand, 
                we hope to get back a summary of the information that exists inside the hidden state, because it is encoded in a latent space the model uses itself!! How cool is that! 😯😯😯
                ''', line_breaks=True)
        with gr.Column(scale=1):    
            gr.Markdown('<span style="font-size:180px;">🤔</span>')
        
    with gr.Group():
        original_prompt_raw = gr.Textbox(value='Should I eat cake or vegetables?', container=True, label='Original Prompt')
        original_prompt_btn = gr.Button('Compute', variant='primary')
        
    with gr.Accordion(open=False, label='Settings'):
        with gr.Row():
            num_tokens = gr.Slider(1, 100, step=1, value=20, label='Max. # of Tokens')
            repetition_penalty = gr.Slider(1., 10., value=1, label='Repetition Penalty')
            length_penalty = gr.Slider(0, 5, value=0, label='Length Penalty')
            # num_beams = gr.Slider(1, 20, value=1, step=1, label='Number of Beams')
        do_sample = gr.Checkbox(label='With sampling')
        with gr.Accordion(label='Sampling Parameters'):
            with gr.Row():
                temperature = gr.Slider(0., 5., value=0.6, label='Temperature')
                top_k = gr.Slider(1, 1000, value=50, step=1, label='top k')
                top_p = gr.Slider(0., 1., value=0.95, label='top p')

    with gr.Group('Interpretation'):
        interpretation_prompt = gr.Text(suggested_interpretation_prompts[0], label='Interpretation Prompt')

    with gr.Group('Output'):
        tokens_container = []
        with gr.Row():
            for i in range(MAX_PROMPT_TOKENS):
                btn = gr.Button('', visible=False, elem_classes=['token_btn'])
                tokens_container.append(btn)
            interpretation_bubbles = [gr.Textbox('', container=False, visible=False, elem_classes=['bubble'])
                                     for i in range(model.config.num_hidden_layers)]
            
        for i, btn in enumerate(tokens_container):
            btn.click(partial(run_interpretation, i=i), [global_state, interpretation_prompt, num_tokens, do_sample, temperature, 
                                                         top_k, top_p, repetition_penalty, length_penalty
                                                              ], [*interpretation_bubbles])
        
    original_prompt_btn.click(get_hidden_states, 
                              [progress, original_prompt_raw], 
                              [global_state, *tokens_container])    
    demo.launch()