File size: 8,954 Bytes
eb6ddbd
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
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
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
import os
import gradio as gr
from typing import List, Dict
import random
import time
from utils import get_app

# Import all the model registries (keeping existing imports)
import anthropic_gradio
import cerebras_gradio
import dashscope_gradio
import fireworks_gradio
import gemini_gradio
import groq_gradio
import hyperbolic_gradio
import mistral_gradio
import nvidia_gradio
import openai_gradio
import perplexity_gradio
import sambanova_gradio
import together_gradio
import xai_gradio

# Define MODEL_REGISTRIES dictionary
MODEL_REGISTRIES = {
    "OpenAI": (openai_gradio.registry, os.getenv("OPENAI_API_KEY")),
    "Anthropic": (anthropic_gradio.registry, os.getenv("ANTHROPIC_API_KEY")),
    "Cerebras": (cerebras_gradio, os.getenv("CEREBRAS_API_KEY")),
    "DashScope": (dashscope_gradio, os.getenv("DASHSCOPE_API_KEY")),
    "Fireworks": (fireworks_gradio, os.getenv("FIREWORKS_API_KEY")),
    "Gemini": (gemini_gradio, os.getenv("GEMINI_API_KEY")),
    "Groq": (groq_gradio, os.getenv("GROQ_API_KEY")),
    "Hyperbolic": (hyperbolic_gradio, os.getenv("HYPERBOLIC_API_KEY")),
    "Mistral": (mistral_gradio, os.getenv("MISTRAL_API_KEY")),
    "NVIDIA": (nvidia_gradio, os.getenv("NVIDIA_API_KEY")),
    "SambaNova": (sambanova_gradio, os.getenv("SAMBANOVA_API_KEY")),
    "Together": (together_gradio, os.getenv("TOGETHER_API_KEY")),
    "XAI": (xai_gradio, os.getenv("XAI_API_KEY")),
}

def get_all_models():
    """Get all available models from the registries."""
    return [
        "OpenAI: gpt-4o",  # From app_openai.py
        "Anthropic: claude-3-5-sonnet-20241022",  # From app_claude.py
    ]

def generate_discussion_prompt(original_question: str, previous_responses: List[str]) -> str:
    """Generate a prompt for models to discuss and build upon previous responses."""
    prompt = f"""You are participating in a multi-AI discussion about this question: "{original_question}"

Previous responses from other AI models:
{chr(10).join(f"- {response}" for response in previous_responses)}

Please provide your perspective while:
1. Acknowledging key insights from previous responses
2. Adding any missing important points
3. Respectfully noting if you disagree with anything and explaining why
4. Building towards a complete answer

Keep your response focused and concise (max 3-4 paragraphs)."""
    return prompt

def generate_consensus_prompt(original_question: str, discussion_history: List[str]) -> str:
    """Generate a prompt for final consensus building."""
    return f"""Review this multi-AI discussion about: "{original_question}"

Discussion history:
{chr(10).join(discussion_history)}

As a final synthesizer, please:
1. Identify the key points where all models agreed
2. Explain how any disagreements were resolved
3. Present a clear, unified answer that represents our collective best understanding
4. Note any remaining uncertainties or caveats

Keep the final consensus concise but complete."""

def chat_with_openai(model: str, messages: List[Dict], api_key: str) -> str:
    import openai
    client = openai.OpenAI(api_key=api_key)
    response = client.chat.completions.create(
        model=model,
        messages=messages
    )
    return response.choices[0].message.content

def chat_with_anthropic(model: str, messages: List[Dict], api_key: str) -> str:
    from anthropic import Anthropic
    client = Anthropic(api_key=api_key)
    # Convert messages to Anthropic format
    prompt = "\n\n".join([f"{m['role']}: {m['content']}" for m in messages])
    response = client.messages.create(
        model=model,
        messages=[{"role": "user", "content": prompt}]
    )
    return response.content[0].text

def multi_model_consensus(
    question: str, 
    selected_models: List[str], 
    rounds: int = 3,
    progress: gr.Progress = gr.Progress()
) -> tuple[str, List[Dict]]:
    if not selected_models:
        return "Please select at least one model to chat with.", []
    
    chat_history = []
    discussion_history = []
    
    # Initial responses
    progress(0, desc="Getting initial responses...")
    initial_responses = []
    for i, model in enumerate(selected_models):
        provider, model_name = model.split(": ", 1)
        registry_fn, api_key = MODEL_REGISTRIES[provider]
        
        if not api_key:
            continue
            
        try:
            # Load the model using the registry function
            predictor = gr.load(
                name=model_name,
                src=registry_fn,
                token=api_key
            )
            
            # Format the request based on the provider
            if provider == "Anthropic":
                response = predictor.predict(
                    messages=[{"role": "user", "content": question}],
                    max_tokens=1024,
                    model=model_name,
                    api_name="chat"
                )
            else:
                response = predictor.predict(
                    question,
                    api_name="chat"
                )
                
            initial_responses.append(f"{model}: {response}")
            discussion_history.append(f"Initial response from {model}:\n{response}")
            chat_history.append((f"Initial response from {model}", response))
        except Exception as e:
            chat_history.append((f"Error from {model}", str(e)))
    
    # Discussion rounds
    for round_num in range(rounds):
        progress((round_num + 1) / (rounds + 2), desc=f"Discussion round {round_num + 1}...")
        round_responses = []
        
        random.shuffle(selected_models)  # Randomize order each round
        for model in selected_models:
            provider, model_name = model.split(": ", 1)
            registry, api_key = MODEL_REGISTRIES[provider]
            
            if not api_key:
                continue
                
            try:
                discussion_prompt = generate_discussion_prompt(question, discussion_history)
                response = registry.chat(
                    model=model_name,
                    messages=[{"role": "user", "content": discussion_prompt}],
                    api_key=api_key
                )
                round_responses.append(f"{model}: {response}")
                discussion_history.append(f"Round {round_num + 1} - {model}:\n{response}")
                chat_history.append((f"Round {round_num + 1} - {model}", response))
            except Exception as e:
                chat_history.append((f"Error from {model} in round {round_num + 1}", str(e)))
    
    # Final consensus - use the model that's shown most consistency
    progress(0.9, desc="Building final consensus...")
    # Use the first model for final consensus instead of two models
    model = selected_models[0]
    provider, model_name = model.split(": ", 1)
    registry, api_key = MODEL_REGISTRIES[provider]
    
    try:
        consensus_prompt = generate_consensus_prompt(question, discussion_history)
        final_consensus = registry.chat(
            model=model_name,
            messages=[{"role": "user", "content": consensus_prompt}],
            api_key=api_key
        )
    except Exception as e:
        final_consensus = f"Error getting consensus from {model}: {str(e)}"
    
    chat_history.append(("Final Consensus", final_consensus))
    
    progress(1.0, desc="Done!")
    return chat_history

with gr.Blocks() as demo:
    gr.Markdown("# Experimental Multi-Model Consensus Chat")
    gr.Markdown("""Select multiple models to collaborate on answering your question. 
                The models will discuss with each other and attempt to reach a consensus.
                Maximum 5 models can be selected at once.""")
    
    with gr.Row():
        with gr.Column():
            model_selector = gr.Dropdown(
                choices=get_all_models(),
                multiselect=True,
                label="Select Models (max 5)",
                info="Choose up to 5 models to participate in the discussion",
                value=["OpenAI: gpt-4o", "Anthropic: claude-3-5-sonnet-20241022"],  # Updated model names
                max_choices=5
            )
            rounds_slider = gr.Slider(
                minimum=1,
                maximum=5,
                value=3,
                step=1,
                label="Discussion Rounds",
                info="Number of rounds of discussion between models"
            )
    
    chatbot = gr.Chatbot(height=600, label="Multi-Model Discussion")
    msg = gr.Textbox(label="Your Question", placeholder="Ask a question for the models to discuss...")
    
    def respond(message, selected_models, rounds):
        chat_history = multi_model_consensus(message, selected_models, rounds)
        return chat_history
    
    msg.submit(
        respond,
        [msg, model_selector, rounds_slider],
        [chatbot],
        api_name="consensus_chat"
    )

if __name__ == "__main__":
    demo.launch()