File size: 7,973 Bytes
dd84964
 
 
144aae5
 
dd84964
 
 
 
144aae5
 
 
 
 
dd84964
 
 
 
 
 
144aae5
 
dd84964
 
144aae5
dd84964
 
 
144aae5
 
dd84964
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
144aae5
 
dd84964
144aae5
 
 
 
 
 
 
 
 
dd84964
144aae5
 
dd84964
 
 
144aae5
dd84964
 
 
144aae5
dd84964
 
 
144aae5
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
dd84964
 
 
 
144aae5
dd84964
 
 
144aae5
dd84964
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
144aae5
 
dd84964
144aae5
dd84964
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
144aae5
dd84964
 
144aae5
dd84964
 
 
 
144aae5
dd84964
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
144aae5
 
 
 
dd84964
 
 
 
 
 
144aae5
 
 
 
 
 
dd84964
144aae5
 
 
 
 
dd84964
144aae5
 
 
 
dd84964
144aae5
 
 
 
 
 
 
 
 
 
 
dd84964
 
 
 
 
 
144aae5
dd84964
144aae5
dd84964
 
144aae5
dd84964
144aae5
 
dd84964
 
144aae5
dd84964
144aae5
 
dd84964
 
144aae5
dd84964
144aae5
dd84964
 
144aae5
dd84964
144aae5
dd84964
144aae5
 
 
 
 
dd84964
 
 
 
 
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
238
239
240
241
242
243
244
245
246
247
248
249
250
251
import gradio as gr
import torch
import torch.nn as nn
from model import SmolLM2Model  # โœ… Correct import
from transformers import AutoTokenizer, AutoConfig

# Device setup
device = torch.device("cuda" if torch.cuda.is_available() else "cpu")

# Load tokenizer and config
print("Loading tokenizer and config...")
tokenizer = AutoTokenizer.from_pretrained("HuggingFaceTB/SmolLM2-135M")
config = AutoConfig.from_pretrained("HuggingFaceTB/SmolLM2-135M")

# Load model
@torch.no_grad()
def load_model():
    """Load the trained model"""
    print("Loading model...")
    
    # Initialize model with config
    model = SmolLM2Model(config).to(device)
    
    # Load checkpoint
    checkpoint = torch.load('checkpoint_step_5050.pt', map_location=device)
    model.load_state_dict(checkpoint['model_state_dict'])
    model.eval()
    
    print(f"โœ… Model loaded successfully on {device}")
    print(f"โœ… Training step: {checkpoint.get('step', 'N/A')}")
    return model, checkpoint

# Load model at startup
model, checkpoint = load_model()

@torch.no_grad()
def generate_text(
    prompt,
    max_length=100,
    temperature=0.8,
    top_k=50,
    top_p=0.9
):
    """Generate text from prompt"""
    try:
        # Tokenize input
        inputs = tokenizer(prompt, return_tensors="pt").to(device)
        input_ids = inputs['input_ids']
        
        # Generate using model's built-in method
        generated_ids = model.generate(
            input_ids,
            max_new_tokens=max_length,
            temperature=temperature,
            top_p=top_p,
            top_k=top_k if top_k > 0 else None,
            do_sample=temperature > 0
        )
        
        # Decode
        output_text = tokenizer.decode(generated_ids[0], skip_special_tokens=True)
        return output_text
        
    except Exception as e:
        return f"โŒ Error generating text: {str(e)}"

def get_model_info():
    """Display model information"""
    total_params = model.get_num_params()
    trainable_params = sum(p.numel() for p in model.parameters() if p.requires_grad)
    
    info = f"""
### ๐Ÿ“Š Model Information

**Model:** SmolLM2-135M  
**Total Parameters:** {total_params:,} (~{total_params/1e6:.1f}M)  
**Trainable Parameters:** {trainable_params:,}  
**Training Steps:** {checkpoint.get('step', 'N/A')}  
**Device:** {device}  
**Vocab Size:** {config.vocab_size:,}

### ๐Ÿ—๏ธ Architecture
- **Layers:** {config.num_hidden_layers}
- **Hidden Size:** {config.hidden_size}
- **Attention Heads:** {config.num_attention_heads} (Query) / {config.num_key_value_heads} (KV)
- **FFN Size:** {config.intermediate_size}
- **Context Length:** {config.max_position_embeddings}

### ๐ŸŽฏ Training Details
- โœ… Trained for 5,000 steps
- โœ… Checkpoint saved and reloaded
- โœ… Additional 50 steps after reload
- โœ… Predictions logged every 500 steps
    """
    return info

# Gradio Interface
with gr.Blocks(theme=gr.themes.Soft(), title="SmolLM2-135M Demo") as demo:
    gr.Markdown("""
    # ๐Ÿค– SmolLM2-135M: From-Scratch Implementation
    
    Complete reverse-engineered implementation of SmolLM2-135M, trained from scratch.
    
    **GitHub:** [abi2024/smollm2-135-implementation](https://github.com/abi2024/smollm2-135-implementation)
    """)
    
    with gr.Tab("๐ŸŽฎ Generate Text"):
        with gr.Row():
            with gr.Column():
                prompt_input = gr.Textbox(
                    label="Prompt",
                    placeholder="Enter your prompt here...",
                    lines=3,
                    value="Once upon a time"
                )
                
                with gr.Row():
                    max_length_slider = gr.Slider(
                        minimum=10,
                        maximum=200,
                        value=50,
                        step=10,
                        label="Max New Tokens"
                    )
                    temperature_slider = gr.Slider(
                        minimum=0.1,
                        maximum=2.0,
                        value=0.8,
                        step=0.1,
                        label="Temperature"
                    )
                
                with gr.Row():
                    top_k_slider = gr.Slider(
                        minimum=0,
                        maximum=100,
                        value=50,
                        step=5,
                        label="Top-K"
                    )
                    top_p_slider = gr.Slider(
                        minimum=0.0,
                        maximum=1.0,
                        value=0.9,
                        step=0.05,
                        label="Top-P (Nucleus)"
                    )
                
                generate_btn = gr.Button("๐Ÿš€ Generate", variant="primary", size="lg")
            
            with gr.Column():
                output_text = gr.Textbox(
                    label="Generated Text",
                    lines=12,
                    interactive=False
                )
        
        generate_btn.click(
            fn=generate_text,
            inputs=[
                prompt_input,
                max_length_slider,
                temperature_slider,
                top_k_slider,
                top_p_slider
            ],
            outputs=output_text
        )
        
        gr.Markdown("""
        ### ๐Ÿ’ก Generation Tips:
        - **Temperature**: Controls randomness (0.1 = focused, 2.0 = creative)
        - **Top-K**: Limits to K most likely tokens (0 = disabled)
        - **Top-P**: Nucleus sampling threshold (0.9 recommended)
        """)
    
    with gr.Tab("๐Ÿ“Š Model Info"):
        model_info_display = gr.Markdown(get_model_info())
        
        gr.Markdown("""
        ### ๐Ÿ” Reverse Engineering Process
        
        1. **Architecture Analysis**
           - Studied SmolLM2 GitHub repository
           - Extracted model configuration from YAML
           - Downloaded pretrained 135M checkpoint
        
        2. **Implementation**
           - Built from scratch using PyTorch
           - Implemented Grouped Query Attention (9Q/3KV heads)
           - Added RoPE position embeddings
           - Used SwiGLU FFN and RMSNorm
        
        3. **Validation**
           - Loaded official pretrained weights
           - Verified parameter count (134,515,008)
           - Confirmed architecture matches exactly
        
        ### โšก Optimizations Applied
        - โœ… Flash Attention 2 (via scaled_dot_product_attention)
        - โœ… Mixed Precision Training (BF16/FP16)
        - โœ… Gradient Accumulation
        - โœ… torch.compile() for inference speedup
        - โœ… Grouped Query Attention (memory efficient)
        
        ### ๐Ÿ“ˆ Training Pipeline
        1. **Main Training:** 5,000 steps with predictions every 500 steps
        2. **Checkpoint Test:** Model saved and successfully reloaded
        3. **Resume Training:** 50 additional steps (validates checkpoint integrity)
        """)
    
    with gr.Tab("๐ŸŽฏ Example Prompts"):
        gr.Markdown("""
        ### Try these prompts:
        
        **1. Story Generation**
```
        Once upon a time in a magical forest,
```
        
        **2. Code Completion**
```
        def calculate_fibonacci(n):
            # Calculate the nth Fibonacci number
```
        
        **3. Question Answering**
```
        Q: What is the capital of France?
        A:
```
        
        **4. Technical Writing**
```
        The main advantage of transformer architectures is
```
        
        **5. Creative Writing**
```
        The scientist discovered something extraordinary:
```
        
        ### ๐ŸŽ›๏ธ Recommended Settings:
        - **Creative Writing:** Temperature=1.0, Top-P=0.95
        - **Code Generation:** Temperature=0.3, Top-P=0.9, Top-K=40
        - **Factual Q&A:** Temperature=0.5, Top-P=0.8, Top-K=30
        """)

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