Spaces:
Sleeping
Sleeping
File size: 8,410 Bytes
8bf4d58 |
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 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 |
"""Chain-of-Thought (CoT) planner implementation."""
import logging
from typing import List, Dict, Any, Optional, Callable
from src.core.config import get_settings
logger = logging.getLogger(__name__)
class CoTPlanner:
"""Chain-of-Thought planner for multi-step reasoning."""
def __init__(
self,
max_steps: int = 10,
enable_reflection: bool = True,
):
"""
Initialize CoT planner.
Args:
max_steps: Maximum number of reasoning steps
enable_reflection: Whether to enable reflection on steps
"""
self.max_steps = max_steps
self.enable_reflection = enable_reflection
def plan(
self,
query: str,
context: Optional[str] = None,
llm_call: Optional[Callable] = None,
) -> Dict[str, Any]:
"""
Generate a plan using Chain-of-Thought methodology.
Args:
query: User query
context: Optional context information
llm_call: Function to call LLM
Returns:
Plan dictionary with reasoning chain
"""
if not llm_call:
raise ValueError("llm_call function is required")
reasoning_steps = []
current_understanding = context or ""
# Initial reasoning prompt
prompt = self._build_initial_prompt(query, context)
for step_num in range(self.max_steps):
try:
# Get LLM response
response = llm_call(prompt)
# Parse response
step = self._parse_cot_response(response, step_num)
reasoning_steps.append(step)
# Check if we have a conclusion
if step.get("is_conclusion", False):
return {
"query": query,
"reasoning_steps": reasoning_steps,
"conclusion": step.get("content", ""),
"total_steps": step_num + 1,
}
# Build next step prompt
prompt = self._build_next_step_prompt(
query=query,
context=current_understanding,
previous_steps=reasoning_steps,
current_step=step,
)
# Optional: Reflection step
if self.enable_reflection and step_num > 0 and step_num % 3 == 0:
reflection = self._reflect_on_progress(reasoning_steps, llm_call)
if reflection:
reasoning_steps.append({
"step": step_num + 1,
"type": "reflection",
"content": reflection,
})
except Exception as e:
logger.error(f"Error in CoT step {step_num}: {e}")
reasoning_steps.append({
"step": step_num + 1,
"type": "error",
"content": f"Error: {str(e)}",
})
# Max steps reached
return {
"query": query,
"reasoning_steps": reasoning_steps,
"conclusion": None,
"total_steps": self.max_steps,
"status": "max_steps_reached",
}
def _build_initial_prompt(
self,
query: str,
context: Optional[str],
) -> str:
"""Build initial CoT prompt."""
prompt_parts = [
"You are a helpful assistant that uses Chain-of-Thought reasoning.",
"Break down complex problems into smaller steps and reason through them step by step.",
"",
"Format your reasoning as:",
"Step 1: <your reasoning>",
"Step 2: <your reasoning>",
"...",
"Conclusion: <final answer>",
"",
]
if context:
prompt_parts.extend([
f"Context: {context}",
"",
])
prompt_parts.extend([
f"Question: {query}",
"",
"Begin your reasoning:",
])
return "\n".join(prompt_parts)
def _build_next_step_prompt(
self,
query: str,
context: str,
previous_steps: List[Dict[str, Any]],
current_step: Dict[str, Any],
) -> str:
"""Build prompt for next reasoning step."""
prompt_parts = [
"Continue your Chain-of-Thought reasoning.",
"",
f"Question: {query}",
"",
]
if context:
prompt_parts.append(f"Context: {context}\n")
prompt_parts.append("Previous reasoning steps:")
for step in previous_steps[-3:]: # Last 3 steps
if step.get("type") != "reflection":
prompt_parts.append(f"Step {step.get('step', '?')}: {step.get('content', '')}")
prompt_parts.extend([
"",
"What is the next step in your reasoning?",
])
return "\n".join(prompt_parts)
def _parse_cot_response(
self,
response: str,
step_num: int,
) -> Dict[str, Any]:
"""Parse CoT response into structured step."""
response = response.strip()
# Check for conclusion
if "Conclusion:" in response or response.startswith("Conclusion"):
conclusion_part = response.split("Conclusion:")[-1].strip()
return {
"step": step_num + 1,
"type": "conclusion",
"content": conclusion_part,
"is_conclusion": True,
}
# Extract step number and content
step_content = response
step_type = "reasoning"
# Try to extract step number
if response.startswith(f"Step {step_num + 1}:"):
step_content = response.split(":", 1)[1].strip()
elif "Step" in response:
# Extract any step number
parts = response.split("Step", 1)
if len(parts) > 1:
step_part = parts[1].split(":", 1)
if len(step_part) > 1:
step_content = step_part[1].strip()
return {
"step": step_num + 1,
"type": step_type,
"content": step_content,
"is_conclusion": False,
}
def _reflect_on_progress(
self,
steps: List[Dict[str, Any]],
llm_call: Callable,
) -> Optional[str]:
"""Reflect on reasoning progress."""
if not steps:
return None
reflection_prompt = (
"Review the following reasoning steps and provide a brief reflection:\n\n"
)
for step in steps[-3:]:
reflection_prompt += f"Step {step.get('step', '?')}: {step.get('content', '')}\n"
reflection_prompt += "\nReflection:"
try:
reflection = llm_call(reflection_prompt)
return reflection.strip()
except Exception as e:
logger.error(f"Error in reflection: {e}")
return None
def generate_execution_plan(
self,
reasoning_steps: List[Dict[str, Any]],
) -> List[Dict[str, Any]]:
"""
Generate an execution plan from reasoning steps.
Args:
reasoning_steps: List of reasoning steps
Returns:
List of execution actions
"""
execution_plan = []
for step in reasoning_steps:
if step.get("type") == "conclusion":
execution_plan.append({
"action": "return_answer",
"content": step.get("content", ""),
})
elif step.get("type") == "reasoning":
# Extract potential actions from reasoning
content = step.get("content", "")
if "search" in content.lower():
execution_plan.append({
"action": "search",
"reasoning": content,
})
elif "calculate" in content.lower() or "compute" in content.lower():
execution_plan.append({
"action": "calculate",
"reasoning": content,
})
return execution_plan
|