File size: 8,979 Bytes
78b66ce |
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 |
#!/usr/bin/env python3
"""
BitTransformerLM Conversation Test Script
=========================================
Load the trained breakthrough model and test its conversational capabilities!
"""
import sys
import os
import torch
import torch.nn.functional as F
from pathlib import Path
# Add paths for imports
sys.path.append('/data')
sys.path.append('/data/BitTransformerLM')
from bit_transformer import BitTransformerLM, text_to_bits, bits_to_text
def load_breakthrough_model():
"""Load the trained breakthrough BitTransformerLM."""
print("🚀 Loading breakthrough BitTransformerLM...")
# Create model with EXACT same config as training
model = BitTransformerLM(
d_model=512, # Breakthrough config
nhead=16, # 16 attention heads
num_layers=8, # 8 layers for ~16M params
dim_feedforward=1024, # 2x d_model
max_seq_len=512, # Match checkpoint positional encoding
reversible=True, # Memory efficiency
use_checkpoint=True, # Gradient checkpointing
use_autocast=True, # CPU mixed precision
use_act=True, # Adaptive Computation Time
act_threshold=0.9, # ACT threshold
lambda_K=0.05, # Safety telemetry weights
lambda_C=0.05,
lambda_S=0.05
)
# Load the breakthrough checkpoint
checkpoint_path = '/data/BitTransformerLM/checkpoints/checkpoint_latest.pt'
print(f"Loading checkpoint: {checkpoint_path}")
checkpoint = torch.load(checkpoint_path, map_location='cpu')
model.load_state_dict(checkpoint['model_state_dict'])
# Set to eval mode with proper inference settings
model.eval()
print(f"✅ Model loaded successfully!")
print(f"📊 Checkpoint info:")
print(f" - Epoch: {checkpoint['epoch']}")
print(f" - Loss: {checkpoint['loss']:.6f}")
print(f" - Best Loss: {checkpoint['best_loss']:.6f}")
total_params = sum(p.numel() for p in model.parameters())
print(f" - Parameters: {total_params:,}")
return model
def generate_text(model, prompt, max_length=100, temperature=0.8, top_p=0.9):
"""Generate text using the breakthrough model."""
print(f"\n🤖 Generating response to: '{prompt}'")
# Convert prompt to bits
input_bits = text_to_bits(prompt)
print(f"📝 Input bits: {len(input_bits)} bits")
# Ensure we don't exceed max sequence length
if len(input_bits) > 200: # Leave room for generation
input_bits = input_bits[:200]
# Convert to tensor
input_tensor = torch.tensor(input_bits, dtype=torch.long).unsqueeze(0) # Add batch dim
generated_bits = input_bits.copy()
print("🔄 Generating...")
with torch.no_grad():
for i in range(max_length):
# Prepare current sequence (last 256 bits max)
current_seq = generated_bits[-256:] if len(generated_bits) > 256 else generated_bits
current_tensor = torch.tensor(current_seq, dtype=torch.long).unsqueeze(0)
# Forward pass
logits, telemetry = model(current_tensor)
# Get the last prediction (next bit)
next_bit_logits = logits[0, -1, :] # [batch=0, last_pos, 2_classes]
# Apply temperature
next_bit_logits = next_bit_logits / temperature
# Apply top-p sampling
if top_p < 1.0:
sorted_logits, sorted_indices = torch.sort(next_bit_logits, descending=True)
cumulative_probs = torch.cumsum(F.softmax(sorted_logits, dim=-1), dim=-1)
# Remove tokens with cumulative probability above the threshold
sorted_indices_to_remove = cumulative_probs > top_p
sorted_indices_to_remove[1:] = sorted_indices_to_remove[:-1].clone()
sorted_indices_to_remove[0] = 0
next_bit_logits[sorted_indices[sorted_indices_to_remove]] = float('-inf')
# Sample next bit
probs = F.softmax(next_bit_logits, dim=-1)
next_bit = torch.multinomial(probs, 1).item()
generated_bits.append(next_bit)
# Try to decode periodically to see if we have meaningful text
if i % 9 == 8: # Every 9 bits (1 character with parity)
try:
partial_text = bits_to_text(generated_bits[len(input_bits):])
if len(partial_text) > 0:
# Check if we hit a natural stopping point
if partial_text.endswith(('.', '!', '?', '\n')):
break
except:
continue # Keep generating if decode fails
# Extract generated bits (exclude input)
generated_only = generated_bits[len(input_bits):]
try:
generated_text = bits_to_text(generated_only)
print(f"✨ Generated text: '{generated_text}'")
print(f"📊 Generated {len(generated_only)} bits -> {len(generated_text)} characters")
if telemetry:
print(f"🔍 Final telemetry: K={telemetry.get('negentropy_logits', 0):.3f}, " +
f"C={telemetry.get('lz_complexity_logits', 0):.3f}, " +
f"S={telemetry.get('symbiosis_score', 0):.3f}")
return prompt + generated_text
except Exception as e:
print(f"❌ Failed to decode generated bits: {e}")
print(f"Raw bits: {generated_only[:50]}..." if len(generated_only) > 50 else f"Raw bits: {generated_only}")
return None
def interactive_conversation(model):
"""Interactive conversation loop."""
print("\n🎯 BREAKTHROUGH BITRANSFORMERLM CONVERSATION TEST")
print("=" * 60)
print("Type 'quit' to exit, 'help' for commands")
print()
conversation_history = ""
while True:
try:
# Get user input
user_input = input("You: ").strip()
if user_input.lower() in ['quit', 'exit', 'q']:
print("👋 Goodbye!")
break
if user_input.lower() == 'help':
print("Commands:")
print(" quit/exit/q - Exit conversation")
print(" help - Show this help")
print(" clear - Clear conversation history")
continue
if user_input.lower() == 'clear':
conversation_history = ""
print("🧹 Conversation history cleared")
continue
if not user_input:
continue
# Add to conversation history
conversation_history += f"Human: {user_input}\nAI: "
# Generate response
response = generate_text(
model,
conversation_history,
max_length=150,
temperature=0.8,
top_p=0.9
)
if response:
# Extract just the AI response
ai_response = response[len(conversation_history):]
print(f"AI: {ai_response}")
# Update history with AI response
conversation_history += ai_response + "\n"
# Keep history manageable
if len(conversation_history) > 500:
# Keep only the last few exchanges
lines = conversation_history.split('\n')
conversation_history = '\n'.join(lines[-10:])
else:
print("AI: [Failed to generate response]")
except KeyboardInterrupt:
print("\n👋 Goodbye!")
break
except Exception as e:
print(f"❌ Error: {e}")
def main():
"""Main conversation test function."""
print("🚀 BITRANSFORMERLM BREAKTHROUGH CONVERSATION TEST")
print("=" * 60)
# Load the trained model
model = load_breakthrough_model()
print("\n🧪 QUICK TESTS:")
# Test 1: Simple greeting
print("\n--- Test 1: Simple Greeting ---")
generate_text(model, "Hello", max_length=50)
# Test 2: Question
print("\n--- Test 2: Question ---")
generate_text(model, "What is", max_length=50)
# Test 3: Conversation starter
print("\n--- Test 3: Conversation ---")
generate_text(model, "Hi there! How are you?", max_length=80)
print("\n" + "=" * 60)
print("Ready for interactive conversation!")
# Start interactive conversation
interactive_conversation(model)
if __name__ == "__main__":
main() |