File size: 5,967 Bytes
f1fd41e
7ae54ea
6f93dce
c5dd812
 
 
 
580eaed
c5dd812
6f93dce
45ef073
6f93dce
c5dd812
 
 
 
 
 
 
2f665a8
c5dd812
 
2f665a8
c5dd812
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
2f665a8
c5dd812
 
 
 
 
 
2f665a8
7625bb8
45ef073
c5dd812
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
6f93dce
 
c5dd812
 
 
 
 
 
 
 
 
 
 
 
 
 
f1fd41e
c5dd812
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
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
import os
import time
import torch
from flask import Flask, request, jsonify
from flask_cors import CORS
from transformers import AutoModelForCausalLM, AutoTokenizer
import gradio as gr

# Initialize Flask app
app = Flask(__name__)
CORS(app)

# Global variables
MODEL_ID = "microsoft/bitnet-b1.58-2B-4T"
MAX_LENGTH = 2048
MAX_NEW_TOKENS = 512
TEMPERATURE = 0.7
TOP_P = 0.9
THINKING_STEPS = 3  # Number of thinking steps

# Load model and tokenizer
@app.before_first_request
def load_model():
    global model, tokenizer
    
    print(f"Loading model: {MODEL_ID}")
    
    # Load tokenizer
    tokenizer = AutoTokenizer.from_pretrained(MODEL_ID)
    
    # Load model with optimizations for limited resources
    model = AutoModelForCausalLM.from_pretrained(
        MODEL_ID,
        device_map="auto",
        torch_dtype=torch.bfloat16,
        load_in_4bit=True,
    )
    
    print("Model and tokenizer loaded successfully!")

# Helper function for step-by-step thinking
def generate_with_thinking(prompt, thinking_steps=THINKING_STEPS):
    # Initialize conversation with prompt
    full_prompt = prompt
    
    # Add thinking prefix
    thinking_prompt = full_prompt + "\n\nLet me think through this step by step:"
    
    # Generate thinking steps
    thinking_output = ""
    for step in range(thinking_steps):
        # Generate step i of thinking
        inputs = tokenizer(thinking_prompt + thinking_output, return_tensors="pt").to(model.device)
        
        with torch.no_grad():
            outputs = model.generate(
                inputs["input_ids"],
                max_length=MAX_LENGTH,
                max_new_tokens=MAX_NEW_TOKENS // thinking_steps,
                temperature=TEMPERATURE,
                top_p=TOP_P,
                do_sample=True,
                pad_token_id=tokenizer.eos_token_id
            )
        
        # Extract only new tokens
        new_tokens = outputs[0][inputs["input_ids"].shape[1]:]
        thinking_step_output = tokenizer.decode(new_tokens, skip_special_tokens=True)
        
        # Add this step to our thinking output
        thinking_output += f"\n\nStep {step+1}: {thinking_step_output}"
    
    # Now generate final answer based on the thinking
    final_prompt = full_prompt + "\n\n" + thinking_output + "\n\nBased on this thinking, my final answer is:"
    
    inputs = tokenizer(final_prompt, return_tensors="pt").to(model.device)
    with torch.no_grad():
        outputs = model.generate(
            inputs["input_ids"],
            max_length=MAX_LENGTH,
            max_new_tokens=MAX_NEW_TOKENS // 2,
            temperature=TEMPERATURE,
            top_p=TOP_P,
            do_sample=True,
            pad_token_id=tokenizer.eos_token_id
        )
    
    # Extract only the new tokens (the answer)
    new_tokens = outputs[0][inputs["input_ids"].shape[1]:]
    answer = tokenizer.decode(new_tokens, skip_special_tokens=True)
    
    # Return thinking process and final answer
    return {
        "thinking": thinking_output,
        "answer": answer,
        "full_response": thinking_output + "\n\nBased on this thinking, my final answer is: " + answer
    }

# API endpoint for chat
@app.route('/api/chat', methods=['POST'])
def chat():
    try:
        data = request.json
        prompt = data.get('prompt', '')
        include_thinking = data.get('include_thinking', False)
        
        if not prompt:
            return jsonify({'error': 'Prompt is required'}), 400
        
        start_time = time.time()
        response = generate_with_thinking(prompt)
        end_time = time.time()
        
        result = {
            'answer': response['answer'],
            'time_taken': round(end_time - start_time, 2)
        }
        
        # Include thinking steps if requested
        if include_thinking:
            result['thinking'] = response['thinking']
            
        return jsonify(result)
    
    except Exception as e:
        return jsonify({'error': str(e)}), 500

# Simple health check endpoint
@app.route('/health', methods=['GET'])
def health_check():
    return jsonify({'status': 'ok'})

# Gradio Web UI
def create_ui():
    with gr.Blocks() as demo:
        gr.Markdown("# BitNet Specialist Chatbot with Step-by-Step Thinking")
        
        with gr.Row():
            with gr.Column():
                input_text = gr.Textbox(
                    label="Your question", 
                    placeholder="Ask me anything...",
                    lines=3
                )
                
                with gr.Row():
                    submit_btn = gr.Button("Submit")
                    clear_btn = gr.Button("Clear")
                
                show_thinking = gr.Checkbox(label="Show thinking steps", value=True)
                
            with gr.Column():
                thinking_output = gr.Markdown(label="Thinking Process", visible=True)
                answer_output = gr.Markdown(label="Final Answer")
        
        def respond(question, show_thinking):
            if not question.strip():
                return "", "Please enter a question"
            
            response = generate_with_thinking(question)
            
            if show_thinking:
                return response["thinking"], response["answer"]
            else:
                return "", response["answer"]
        
        submit_btn.click(
            respond, 
            inputs=[input_text, show_thinking], 
            outputs=[thinking_output, answer_output]
        )
        
        clear_btn.click(
            lambda: ("", "", ""),
            inputs=None,
            outputs=[input_text, thinking_output, answer_output]
        )
    
    return demo

# Create Gradio UI and launch the app
if __name__ == "__main__":
    # Load model at startup for Gradio
    load_model()
    
    # Create and launch Gradio interface
    demo = create_ui()
    demo.launch(server_name="0.0.0.0", server_port=7860, share=True)