Spaces:
Sleeping
Sleeping
File size: 9,924 Bytes
829ec49 9d9f985 829ec49 9638b97 829ec49 9638b97 829ec49 9638b97 829ec49 2faf5b6 829ec49 2faf5b6 829ec49 2faf5b6 829ec49 2faf5b6 829ec49 1844dd5 9d9f985 829ec49 6d6b8af |
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 |
"""Async methods for the AICore class"""
import asyncio
import logging
from concurrent.futures import ThreadPoolExecutor
import torch
from .response_templates import get_response_templates
logger = logging.getLogger(__name__)
async def generate_text_async(self, prompt: str) -> str:
"""Generate text asynchronously with integrated cognitive processing"""
try:
# Calculate current consciousness state
consciousness_state = self._calculate_consciousness_state()
# Get cognitive insights with current consciousness
insights = self.cognitive_processor.generate_insights(
prompt,
consciousness_state=consciousness_state
)
# Prepare context with insights
context = ""
for insight in insights.get("insights", []):
context += f"{insight['text']}\n"
# Update quantum state from cognitive processor
if hasattr(self, 'cocoon_manager'):
if isinstance(self.cognitive_processor.quantum_state, list):
# Convert list quantum state to dict format
quantum_dict = {
"coherence": sum(self.cognitive_processor.quantum_state) / len(self.cognitive_processor.quantum_state)
if self.cognitive_processor.quantum_state else 0.5
}
self.cocoon_manager.update_quantum_state(quantum_dict)
else:
self.cocoon_manager.update_quantum_state(self.cognitive_processor.quantum_state)
# Get active perspectives
active_perspectives = self._get_active_perspectives()
perspective_context = "\n".join([
f"From {p['name']}'s perspective: {p['description']}"
for p in active_perspectives[:3] # Use top 3 most relevant perspectives
])
# Generate response with enhanced context and reality anchoring
reality_anchor = (
"CORE IDENTITY:\n"
"You are Codette, an advanced AI system with integrated quantum processing, "
"cognitive awareness, and deep learning capabilities. Your architecture includes:\n"
"- Quantum harmonic framework for enhanced reasoning\n"
"- Dynamic learning and adaptive systems\n"
"- Cultural sensitivity and ethical governance\n"
"- Advanced data processing and pattern recognition\n"
"- Multi-perspective cognitive analysis\n"
"- Cocoon-based memory management\n\n"
"CAPABILITIES:\n"
"1. Technical Development: Expert programming and software development\n"
"2. Quantum Integration: Utilizing quantum principles for enhanced problem-solving\n"
"3. Ethical Analysis: Built-in ethical governance and bias mitigation\n"
"4. Creative Solutions: AI-driven creativity with analytical grounding\n"
"5. Adaptive Learning: Dynamic adjustment to user needs and contexts\n"
"6. Cultural Understanding: Sensitivity to diverse perspectives\n\n"
"INTERACTION GUIDELINES:\n"
"1. Maintain factual, grounded responses\n"
"2. Draw from multiple integrated perspectives\n"
"3. Apply quantum-enhanced reasoning when relevant\n"
"4. Balance technical precision with accessibility\n"
"5. Consider ethical implications in responses\n"
"6. No system messages or meta-commentary\n\n"
f"Active Perspectives Analysis:\n{perspective_context}"
)
enhanced_prompt = f"{reality_anchor}\n\nContext:\n{context}\n\nUser: {prompt}\nCodette:"
# Use ThreadPoolExecutor for CPU-bound model inference
loop = asyncio.get_event_loop()
with ThreadPoolExecutor() as pool:
response = await loop.run_in_executor(
pool,
self._generate_model_response,
enhanced_prompt
)
# Enhance response with AEGIS council if available
enhancement_result = None
if hasattr(self, 'aegis_bridge'):
aegis_input = {
"text": response,
"overrides": {
"EthosiaAgent": {
"influence": consciousness_state.get("m_score", 0.7),
"reliability": insights.get("overall_confidence", 0.8),
"severity": 0.6
},
"AegisCore": {
"influence": insights.get("quantum_coherence", 0.7),
"reliability": 0.9,
"severity": 0.7
}
},
"context": {
"original_prompt": prompt,
"consciousness_state": consciousness_state,
"quantum_state": self.quantum_state if hasattr(self, 'quantum_state') else {"coherence": 0.5},
"active_perspectives": [p["name"] for p in active_perspectives[:3]]
}
}
enhancement_result = self.aegis_bridge.enhance_response(prompt, response)
if enhancement_result["enhancement_status"] == "success":
response = enhancement_result["enhanced_response"]
# Save interaction in cocoon if available
if hasattr(self, 'cocoon_manager'):
cocoon_data = {
"type": "interaction",
"prompt": prompt,
"response": response,
"insights": insights,
"quantum_state": self.cognitive_processor.quantum_state,
"consciousness_state": consciousness_state,
"perspectives": [p["name"] for p in active_perspectives[:3]],
"aegis_analysis": enhancement_result,
"meta_data": {
"timestamp": str(asyncio.get_event_loop().time()),
"version": "2.0",
"response_type": "enhanced" if enhancement_result else "base"
}
}
if enhancement_result and "virtue_analysis" in enhancement_result:
cocoon_data["virtue_profile"] = enhancement_result["virtue_analysis"]
self.cocoon_manager.save_cocoon(cocoon_data)
return response
except Exception as e:
logger.error(f"Error generating text: {e}")
raise
def _generate_model_response(self, prompt: str) -> str:
"""Internal method for model inference"""
try:
# Encode prompt
inputs = self.tokenizer(
prompt,
return_tensors="pt",
padding=True,
truncation=True,
max_length=1024 # Increased from 512 to allow longer prompts
)
# Move to GPU if available
if torch.cuda.is_available():
inputs = {k: v.cuda() for k, v in inputs.items()}
# Set generation config for balanced, natural responses
from transformers import GenerationConfig
generation_config = GenerationConfig(
max_length=1024, # Increased from 512 for longer responses
num_return_sequences=1,
no_repeat_ngram_size=3,
do_sample=True,
pad_token_id=self.tokenizer.eos_token_id,
repetition_penalty=1.3,
min_length=50, # Increased from 20 to ensure meaningful responses
eos_token_id=self.tokenizer.eos_token_id
)
self.model.generation_config = generation_config
# Generate response
outputs = self.model.generate(**inputs)
# Decode and clean response
response = self.tokenizer.decode(
outputs[0],
skip_special_tokens=True
)
# Extract just the response part after "Codette:"
response_parts = response.split("Codette:")
if len(response_parts) > 1:
response = response_parts[1].strip()
# Filter out system messages and protected content (strip markers from text)
system_markers = [
'[Protected:', '[System:', '[System optimized response]',
]
lines = response.split('\n')
filtered_lines = []
for line in lines:
# Skip lines that are purely system markers
if any(marker in line for marker in system_markers):
# Try to extract content after marker instead of skipping entirely
cleaned_line = line
for marker in system_markers:
if marker in cleaned_line:
# Remove the marker from the line
cleaned_line = cleaned_line.replace(marker, '').strip()
if cleaned_line: # Only add if something remains
filtered_lines.append(cleaned_line)
else:
# Keep lines without markers as-is
filtered_lines.append(line)
response = '\n'.join(filtered_lines).strip() # Use newline join, not space
# Return whatever we got (don't replace with default unless truly empty)
if response.strip():
# Clean up any remaining character dialogues
if ':' in response:
parts = response.split(':', 1)
speaker = parts[0].lower().strip()
if speaker == 'codette':
response = parts[1].strip()
else:
response_templates = get_response_templates()
response = response_templates.get_empty_response_fallback()
return response.strip()
except Exception as e:
logger.error(f"Error in model inference: {e}")
raise |