File size: 7,646 Bytes
a3738dd | 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 | import torch
from typing import Dict, List, Any, Optional, Union
class ReasoningEngine:
"""Reasoning engine for the autonomous AI agent
This module provides advanced reasoning capabilities including:
1. Chain-of-thought reasoning
2. Task decomposition
3. Self-reflection
4. Decision making
"""
def __init__(self, model, tokenizer, device="cpu"):
"""Initialize the reasoning engine
Args:
model: The language model to use for reasoning
tokenizer: The tokenizer for the language model
device: The device to run the model on (cpu or cuda)
"""
self.model = model
self.tokenizer = tokenizer
self.device = device
def generate_text(self, prompt: str, max_length: int = 512, temperature: float = 0.7) -> str:
"""Generate text using the language model
Args:
prompt: The input prompt for the model
max_length: Maximum length of the generated text
temperature: Temperature for sampling (higher = more random)
Returns:
Generated text response
"""
inputs = self.tokenizer(prompt, return_tensors="pt").to(self.device)
# Generate response
with torch.no_grad():
outputs = self.model.generate(
inputs["input_ids"],
max_length=max_length,
num_return_sequences=1,
temperature=temperature,
top_p=0.9,
do_sample=True
)
# Decode and return the response
response = self.tokenizer.decode(outputs[0], skip_special_tokens=True)
return response
def chain_of_thought(self, query: str) -> Dict[str, str]:
"""Implement chain-of-thought reasoning
Args:
query: User query to reason about
Returns:
Dictionary containing reasoning steps and final answer
"""
# Construct a prompt that encourages step-by-step reasoning
reasoning_prompt = f"""I am an autonomous AI agent called ResuRank. I need to think through this step by step to provide the best response.
Query: {query}
Let me reason through this carefully:
1. First, I'll identify the key parts of this query.
2. Then, I'll consider what information I need to answer it.
3. Next, I'll analyze the implications and context.
4. Finally, I'll formulate a comprehensive response.
Step-by-step reasoning:
1. """
# Generate the reasoning steps
reasoning = self.generate_text(reasoning_prompt, max_length=1024)
# Analyze the reasoning for completeness
if len(reasoning.split('\n')) < 3 or len(reasoning) < 100:
# If reasoning is too short, try to expand it
expansion_prompt = f"""My current reasoning is:
{reasoning}
Let me expand on this with more detailed analysis:
"""
additional_reasoning = self.generate_text(expansion_prompt, max_length=512)
reasoning = f"{reasoning}\n\n{additional_reasoning}"
# Extract the final answer after reasoning
answer_prompt = f"""Based on my detailed reasoning:
{reasoning}
I will now provide a clear, helpful, and comprehensive response to the query: {query}
My final answer is:"""
final_answer = self.generate_text(answer_prompt, max_length=768)
# Check if the answer addresses the query adequately
if len(final_answer) < 50:
# If answer is too short, try to improve it
improvement_prompt = f"""My current answer is:
{final_answer}
This answer is too brief. Let me provide a more comprehensive response to the query: {query}
Improved answer:"""
final_answer = self.generate_text(improvement_prompt, max_length=768)
return {
"reasoning": reasoning,
"answer": final_answer
}
def decompose_task(self, task_description: str) -> List[str]:
"""Decompose a complex task into smaller subtasks
Args:
task_description: Description of the task to decompose
Returns:
List of subtask descriptions
"""
decomposition_prompt = f"""I need to break down this complex task into smaller, manageable subtasks:
Task: {task_description}
Subtasks:
1. """
decomposition = self.generate_text(decomposition_prompt, max_length=1024)
# Parse the decomposition into a list of subtasks
subtasks = []
for line in decomposition.split('\n'):
line = line.strip()
if line and (line[0].isdigit() or line.startswith('- ')):
# Remove numbering or bullet points
cleaned_line = re.sub(r'^\d+\.\s*|^-\s*', '', line).strip()
if cleaned_line:
subtasks.append(cleaned_line)
return subtasks
def self_reflect(self, action: str, outcome: str) -> Dict[str, str]:
"""Perform self-reflection on an action and its outcome
Args:
action: The action that was taken
outcome: The outcome of the action
Returns:
Dictionary containing reflection and improvement suggestions
"""
reflection_prompt = f"""I need to reflect on this action and its outcome:
Action: {action}
Outcome: {outcome}
Reflection:
"""
reflection = self.generate_text(reflection_prompt, max_length=512)
improvement_prompt = f"""Based on my reflection:
{reflection}
How I can improve next time:
"""
improvement = self.generate_text(improvement_prompt, max_length=512)
return {
"reflection": reflection,
"improvement": improvement
}
def make_decision(self, options: List[str], context: str) -> Dict[str, Any]:
"""Make a decision among multiple options
Args:
options: List of options to choose from
context: Context information for the decision
Returns:
Dictionary containing the chosen option and reasoning
"""
options_text = "\n".join([f"{i+1}. {option}" for i, option in enumerate(options)])
decision_prompt = f"""I need to make a decision based on the following context and options:
Context: {context}
Options:
{options_text}
Let me analyze each option:
"""
analysis = self.generate_text(decision_prompt, max_length=1024)
conclusion_prompt = f"""Based on my analysis:
{analysis}
The best option is number:"""
conclusion = self.generate_text(conclusion_prompt, max_length=128)
# Try to extract the chosen option number
try:
option_num = int(re.search(r'\d+', conclusion).group()) - 1
chosen_option = options[option_num] if 0 <= option_num < len(options) else options[0]
except (AttributeError, ValueError, IndexError):
# Default to first option if parsing fails
chosen_option = options[0]
return {
"chosen_option": chosen_option,
"analysis": analysis,
"conclusion": conclusion
} |