Spaces:
Running
Running
File size: 7,558 Bytes
75bea1c | 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 | from __future__ import annotations
"""ReACT (Reasoning + Acting) handler."""
import json
from dataclasses import dataclass, field
from typing import Any
from src.llm import LLMClient, Message, MessageRole
from src.llm.prompts import format_prompt, PromptNames
from src.tools.base import ToolRegistry
from src.utils.config import settings
from src.utils.logging import get_logger
logger = get_logger(__name__)
@dataclass
class ReACTStep:
"""A single step in the ReACT loop."""
iteration: int
thought: str
action: str
action_input: dict[str, Any]
observation: str | None = None
@dataclass
class ReACTResult:
"""Result from a ReACT loop execution."""
answer: str
steps: list[ReACTStep] = field(default_factory=list)
success: bool = True
error: str | None = None
class ReACTHandler:
"""Handler for ReACT reasoning loops."""
def __init__(
self,
llm_client: LLMClient,
tool_registry: ToolRegistry,
max_iterations: int | None = None,
):
"""Initialize ReACT handler.
Args:
llm_client: LLM client for reasoning
tool_registry: Registry of available tools
max_iterations: Maximum iterations (defaults to settings)
"""
self.llm = llm_client
self.tools = tool_registry
self.max_iterations = max_iterations or settings.max_iterations
async def run(
self,
query: str,
system_prompt: str,
initial_context: dict[str, Any] | None = None,
) -> ReACTResult:
"""Run a ReACT loop to answer a query.
Args:
query: User's query
system_prompt: System prompt for the agent
initial_context: Optional initial context
Returns:
ReACTResult with answer and step history
"""
steps: list[ReACTStep] = []
working_memory = initial_context or {}
tool_schemas = self.tools.get_schemas()
for iteration in range(1, self.max_iterations + 1):
logger.info(f"ReACT iteration {iteration}")
# Build context from previous steps
context = self._format_steps(steps)
# Generate thought and action
prompt = format_prompt(
PromptNames.REACT_REASONING,
user_query=query,
iteration_number=iteration,
max_iterations=self.max_iterations,
previous_steps=context,
working_memory=json.dumps(working_memory),
)
messages = [
Message(role=MessageRole.SYSTEM, content=system_prompt),
Message(role=MessageRole.USER, content=prompt),
]
response = await self.llm.chat(messages, tools=tool_schemas, temperature=0.5)
# Parse the response
thought, action, action_input = self._parse_response(response)
logger.info(f"Thought: {thought[:100]}...")
logger.info(f"Action: {action}")
# Check for finish
if action.lower() == "finish":
answer = action_input.get("answer", response.content or "")
steps.append(
ReACTStep(
iteration=iteration,
thought=thought,
action="finish",
action_input=action_input,
observation=answer,
)
)
return ReACTResult(answer=answer, steps=steps, success=True)
# Execute action
if response.has_tool_calls:
tool_call = response.tool_calls[0]
result = await self.tools.execute(tool_call.name, **tool_call.arguments)
observation = (
json.dumps(result.data) if result.success else f"Error: {result.error}"
)
action = tool_call.name
action_input = tool_call.arguments
elif action:
result = await self.tools.execute(action, **action_input)
observation = (
json.dumps(result.data) if result.success else f"Error: {result.error}"
)
else:
observation = "No valid action specified"
# Record step
steps.append(
ReACTStep(
iteration=iteration,
thought=thought,
action=action,
action_input=action_input,
observation=observation,
)
)
# Update working memory
working_memory[f"step_{iteration}"] = {
"action": action,
"observation": observation[:500], # Truncate for memory
}
# Max iterations reached
return ReACTResult(
answer="I was unable to find a complete answer within the iteration limit.",
steps=steps,
success=False,
error="Max iterations reached",
)
def _format_steps(self, steps: list[ReACTStep]) -> str:
"""Format steps for context.
Args:
steps: List of ReACT steps
Returns:
Formatted string
"""
if not steps:
return "No previous steps."
formatted = []
for step in steps:
formatted.append(
f"**THOUGHT {step.iteration}:** {step.thought}\n"
f"**ACTION {step.iteration}:** {step.action}[{json.dumps(step.action_input)}]\n"
f"**OBSERVATION {step.iteration}:** {step.observation}"
)
return "\n\n".join(formatted)
def _parse_response(self, response: Any) -> tuple[str, str, dict[str, Any]]:
"""Parse thought and action from response.
Args:
response: LLM response
Returns:
Tuple of (thought, action, action_input)
"""
content = response.content or ""
# Handle tool calls
if response.has_tool_calls:
tool_call = response.tool_calls[0]
thought = content.split("**ACTION")[0].replace("**THOUGHT", "").strip()
thought = thought.strip("*: \n")
return thought, tool_call.name, tool_call.arguments
# Parse text format
thought = ""
action = ""
action_input: dict[str, Any] = {}
if "THOUGHT" in content:
thought_part = content.split("THOUGHT")[-1]
thought = thought_part.split("**ACTION")[0].strip("*: \n")
if "ACTION" in content:
action_part = content.split("ACTION")[-1].strip("*: \n")
if "[" in action_part and "]" in action_part:
action = action_part.split("[")[0].strip()
input_str = action_part[action_part.find("[") + 1 : action_part.rfind("]")]
try:
if input_str.startswith("{"):
action_input = json.loads(input_str)
else:
action_input = {"answer": input_str}
except json.JSONDecodeError:
action_input = {"answer": input_str}
else:
action = action_part.split("\n")[0].strip()
if "finish" in action.lower():
action = "finish"
return thought, action, action_input
|