Spaces:
Running
Running
File size: 9,630 Bytes
0a40afa 966ffcd 0a40afa 966ffcd 0a40afa 966ffcd 0a40afa 924cb7d f98e92f 0a40afa f98e92f 0a40afa f98e92f 0a40afa 924cb7d f98e92f 0a40afa 924cb7d f98e92f 0a40afa 966ffcd 0a40afa 924cb7d 0a40afa 924cb7d f98e92f 0a40afa f98e92f 0a40afa 924cb7d f98e92f 0a40afa 924cb7d f98e92f 0a40afa f98e92f 0a40afa f98e92f 924cb7d f98e92f 924cb7d |
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 |
"""Planner: turns a user request into a JSON tool-plan via Azure OpenAI *Responses*."""
from __future__ import annotations
import json
import logging
from pathlib import Path
from typing import Dict, List, Any
import yaml
from services.llm_client import LLMClient
from config.settings import settings
from services.cost_tracker import CostTracker
_PROMPTS_FILE = Path(__file__).parent.parent / "config" / "prompts.yaml"
# Configure logging
logging.basicConfig(level=logging.INFO)
logger = logging.getLogger(__name__)
class Planner:
"""Generate a plan with the Responses API; fall back to a static template if parsing fails."""
def __init__(self, cost_tracker=None) -> None:
self.prompt_template = self._load_prompt("planner")
self.llm = LLMClient(settings)
self.cost_tracker = cost_tracker or CostTracker()
logger.info("Planner initialized with prompt template")
# --------------------------------------------------
def build_plan(
self,
pdf_meta: Dict[str, Any],
fields: List[str],
doc_preview: str | None = None,
field_descs: Dict | None = None,
strategy: str = "Original Strategy",
unique_indices: List[str] | None = None,
unique_indices_descriptions: Dict[str, str] | None = None,
) -> Dict[str, Any]:
"""Return a JSON dict representing the execution plan."""
logger.info(f"Building plan for strategy: {strategy}")
logger.info(f"Fields: {fields}")
logger.info(f"Unique indices: {unique_indices}")
logger.info(f"Unique indices descriptions: {unique_indices_descriptions}")
# For Unique Indices Strategy, use static plan directly
if strategy == "Unique Indices Strategy":
logger.info("Using static plan for Unique Indices Strategy")
return self._static_plan(fields, strategy, unique_indices, unique_indices_descriptions)
# For Original Strategy, try LLM first
user_context = {
"pdf_meta": pdf_meta,
"doc_preview": doc_preview or "",
"fields": fields,
"field_descriptions": field_descs or {},
"strategy": strategy,
"unique_indices": unique_indices or [],
"unique_indices_descriptions": unique_indices_descriptions or {},
}
logger.info(f"Building plan for fields: {fields}")
logger.info(f"Using strategy: {strategy}")
if unique_indices:
logger.info(f"Unique indices: {unique_indices}")
logger.info(f"Unique indices descriptions: {unique_indices_descriptions}")
logger.debug(f"User context: {user_context}")
prompt = self.prompt_template.format_json(**user_context)
logger.debug(f"Generated prompt: {prompt}")
try:
logger.info("Calling LLM to generate plan")
raw = self.llm.responses(
prompt,
temperature=0.0,
ctx={"cost_tracker": self.cost_tracker},
description="Execution Plan Generation"
)
logger.debug(f"Raw LLM response: {raw}")
try:
logger.info("Parsing LLM response as JSON")
plan = json.loads(raw)
logger.debug(f"Parsed plan: {plan}")
# ensure minimal structure exists
if "steps" in plan and "fields" in plan:
logger.info("Plan successfully generated with required structure")
# Add pdf_meta and strategy info to the plan
plan["pdf_meta"] = pdf_meta
plan["strategy"] = strategy
if unique_indices:
plan["unique_indices"] = unique_indices
if unique_indices_descriptions:
plan["unique_indices_descriptions"] = unique_indices_descriptions
return plan
else:
missing_keys = []
if "steps" not in plan:
missing_keys.append("steps")
if "fields" not in plan:
missing_keys.append("fields")
logger.error(f"Planner: LLM output missing required keys: {missing_keys}. Output: {raw}")
except json.JSONDecodeError as parse_exc:
logger.error(f"Planner: Failed to parse LLM output as JSON. Output: {raw}")
logger.error(f"JSON parsing error: {parse_exc}")
except Exception as parse_exc:
logger.error(f"Planner: Unexpected error parsing LLM output: {parse_exc}")
logger.error(f"LLM output: {raw}")
except Exception as llm_exc:
logger.error(f"Planner: LLM call failed: {llm_exc}")
logger.exception("Full traceback:")
# ---------- fallback static plan ----------
logger.info("Falling back to static plan")
return self._static_plan(fields, strategy, unique_indices, unique_indices_descriptions)
# --------------------------------------------------
@staticmethod
def _load_prompt(name: str):
try:
data = yaml.safe_load(_PROMPTS_FILE.read_text())
logger.debug(f"Loaded prompt template for '{name}'")
except Exception as e:
logger.error(f"Failed to load prompt template: {e}")
data = {}
class _Fmt:
def __init__(self, s: str):
self.s = s
def format_json(self, **kwargs):
# Format the template with the provided fields
fields = kwargs.get("fields", [])
field_descriptions = kwargs.get("field_descriptions", {})
doc_preview = kwargs.get("doc_preview", "")
pdf_meta = kwargs.get("pdf_meta", {})
strategy = kwargs.get("strategy", "Original Strategy")
unique_indices = kwargs.get("unique_indices", [])
unique_indices_descriptions = kwargs.get("unique_indices_descriptions", {})
# Create a formatted string with the actual values
formatted = self.s
if fields:
# Ensure fields is a flat list of strings
fields_json = json.dumps([str(f) for f in fields])
formatted = formatted.replace("<same list you received>", fields_json)
if field_descriptions:
formatted = formatted.replace("field_descriptions for extra context", f"field descriptions: {json.dumps(field_descriptions)}")
if doc_preview:
formatted = formatted.replace("a few kB of raw text from the uploaded document", f"document preview: {doc_preview[:1000]}...")
if pdf_meta:
formatted = formatted.replace("pdf_meta / field_descriptions for extra context", f"document metadata: {json.dumps(pdf_meta)}")
if strategy:
formatted = formatted.replace("strategy for extraction", f"extraction strategy: {strategy}")
if unique_indices:
formatted = formatted.replace("unique indices for extraction", f"unique indices: {json.dumps(unique_indices)}")
if unique_indices_descriptions:
formatted = formatted.replace("unique indices descriptions for extra context", f"unique indices descriptions: {json.dumps(unique_indices_descriptions)}")
return formatted
return _Fmt(data.get(name, "You are a planning agent. Produce a JSON tool plan."))
# --------------------------------------------------
@staticmethod
def _static_plan(fields: List[str], strategy: str = "Original Strategy", unique_indices: List[str] | None = None, unique_indices_descriptions: Dict[str, str] | None = None) -> Dict[str, Any]:
"""Return a hard-coded plan to guarantee offline functionality."""
logger.info("Generating static fallback plan")
logger.info(f"Strategy: {strategy}")
logger.info(f"Fields: {fields}")
logger.info(f"Unique indices: {unique_indices}")
logger.info(f"Unique indices descriptions: {unique_indices_descriptions}")
if strategy == "Unique Indices Strategy":
steps = [
{"tool": "PDFAgent", "args": {}},
{"tool": "TableAgent", "args": {}},
{"tool": "UniqueIndicesCombinator", "args": {}},
]
logger.info("Generated plan for Unique Indices Strategy")
logger.info(f"Steps: {steps}")
else:
steps = [
{"tool": "PDFAgent", "args": {}},
{"tool": "TableAgent", "args": {}},
{
"tool": "ForEachField",
"loop": [
{"tool": "FieldMapper", "args": {"field": "$field"}},
],
},
]
logger.info("Generated plan for Original Strategy")
logger.info(f"Steps: {steps}")
plan = {
"steps": steps,
"fields": fields,
"pdf_meta": {},
"strategy": strategy
}
if unique_indices:
plan["unique_indices"] = unique_indices
if unique_indices_descriptions:
plan["unique_indices_descriptions"] = unique_indices_descriptions
logger.info(f"Final plan: {json.dumps(plan, indent=2)}")
return plan |