Spaces:
Sleeping
Sleeping
File size: 10,430 Bytes
583741e 78f6650 583741e 78f6650 583741e 78f6650 583741e d3ff7fa 583741e d3ff7fa 583741e d3ff7fa 583741e d3ff7fa 583741e d3ff7fa 583741e d3ff7fa 583741e d3ff7fa 583741e d3ff7fa 583741e d3ff7fa 583741e 78f6650 583741e |
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 |
from __future__ import annotations
import os
import sys
import base64
import os
import json
import asyncio
from typing import Any, Dict, List, Optional
from pathlib import Path
from datetime import datetime
from anthropic import AsyncAnthropic
from anthropic.types import ToolUseBlock
from langgraph.graph import END, StateGraph
from pydantic import BaseModel, Field
from src.agents.prompt import REVIEWER_SYSTEM_PROMPT, EVALUATION_PROMPT_TEMPLATE, TOOLS, TOOL_CHOICE
from src.database import db
from src.config import config
from src.logger import logger
class ConversationState(BaseModel):
"""State for the conversation graph"""
messages: List[Dict[str, Any]] = Field(default_factory=list)
response_text: str = ""
tool_result: Optional[Dict[str, Any]] = None
arxiv_id: Optional[str] = None
pdf_path: Optional[str] = None
output_file: Optional[str] = None
def _load_pdf_as_content(pdf_path: str) -> Dict[str, Any]:
if os.path.exists(pdf_path):
with open(pdf_path, "rb") as f:
data_b64 = base64.b64encode(f.read()).decode("utf-8")
return {
"type": "document",
"source": {
"type": "base64",
"media_type": "application/pdf",
"data": data_b64,
},
}
if pdf_path.startswith("http"):
return {
"type": "document",
"source": {
"type": "url",
"url": pdf_path,
},
}
raise FileNotFoundError(f"PDF not found or invalid path: {pdf_path}")
class Evaluator:
def __init__(self, api_key: Optional[str] = None):
api_key = api_key or os.getenv("ANTHROPIC_API_KEY")
if not api_key:
raise ValueError("Anthropic API key is required. Please set HF_SECRET_ANTHROPIC_API_KEY in Hugging Face Spaces secrets or ANTHROPIC_API_KEY environment variable.")
self.client = AsyncAnthropic(api_key=api_key)
self.system_prompt = REVIEWER_SYSTEM_PROMPT
self.eval_template = EVALUATION_PROMPT_TEMPLATE
async def __call__(self, state: ConversationState) -> ConversationState:
"""Evaluate the paper using the conversation state"""
# Prepare messages for the API call
messages = []
messages.extend(state.messages)
# Load PDF content if pdf_path is provided
if state.pdf_path:
try:
pdf_content = _load_pdf_as_content(state.pdf_path)
messages.append({
"role": "user",
"content": [
{"type": "text", "text": "Please evaluate this academic paper:"},
pdf_content
]
})
except Exception as e:
state.response_text = f"Error loading PDF: {str(e)}"
return state
# Add the evaluation prompt
messages.append({
"role": "user",
"content": self.eval_template
})
try:
# Call Anthropic API with tools (async)
response = await self.client.messages.create(
model=config.model_id,
max_tokens=10000,
system=self.system_prompt,
messages=messages,
tools=TOOLS,
tool_choice=TOOL_CHOICE
)
# Process the response
# Check if response is a tool use or text
if response.content and isinstance(response.content[0], ToolUseBlock):
# This is a tool use response
tool_use = response.content[0]
if tool_use:
tool_result = tool_use.input
# set metadata
tool_result['metadata'] = {
'assessed_at': datetime.now().strftime("%Y-%m-%d %H:%M:%S"),
'model': config.model_id,
'version': config.version,
'paper_path': state.pdf_path
}
state.tool_result = tool_result
state.response_text = json.dumps(tool_result, ensure_ascii=False, indent=4)
# Add tool use to messages
state.messages.append({
"role": "assistant",
"content": f"Tool use: {tool_use.name}"
})
else:
state.response_text = "Error: Tool use response but no tool_use found"
else:
# This is a text response
text_content = response.content[0].text if response.content else ""
state.messages.append({
"role": "assistant",
"content": text_content
})
state.response_text = text_content
except Exception as e:
state.response_text = f"Error during evaluation: {str(e)}"
return state
async def save_node(state: ConversationState) -> ConversationState:
"""Save the evaluation result to database"""
try:
if not state.arxiv_id:
state.response_text += f"\n\nError: No arxiv_id provided for database save"
return state
# Parse the evaluation result
evaluation_content = state.response_text
evaluation_score = None
overall_score = None
evaluation_tags = None
# Try to extract score and tags from tool_result if available
if state.tool_result:
try:
# Extract overall automatability score from scores
if 'scores' in state.tool_result and 'overall_automatability' in state.tool_result['scores']:
evaluation_score = state.tool_result['scores']['overall_automatability']
# Extract overall score from scores
if 'scores' in state.tool_result and 'overall_automatability' in state.tool_result['scores']:
overall_score = state.tool_result['scores']['overall_automatability']
# Create tags from key dimensions in scores
tags = []
if 'scores' in state.tool_result:
scores = state.tool_result['scores']
if 'three_year_feasibility_pct' in scores:
tags.append(f"3yr_feasibility:{scores['three_year_feasibility_pct']}%")
if 'task_formalization' in scores:
tags.append(f"task_formalization:{scores['task_formalization']}/4")
if 'data_resource_availability' in scores:
tags.append(f"data_availability:{scores['data_resource_availability']}/4")
evaluation_tags = ",".join(tags) if tags else None
except Exception as e:
logger.warning(f"Warning: Could not extract structured data from tool_result: {e}")
else:
# Try to parse evaluation_content as JSON to extract structured data
try:
evaluation_json = json.loads(evaluation_content)
# Extract overall automatability score from scores
if 'scores' in evaluation_json and 'overall_automatability' in evaluation_json['scores']:
evaluation_score = evaluation_json['scores']['overall_automatability']
# Extract overall score from scores
if 'scores' in evaluation_json and 'overall_automatability' in evaluation_json['scores']:
overall_score = evaluation_json['scores']['overall_automatability']
# Create tags from key dimensions in scores
tags = []
if 'scores' in evaluation_json:
scores = evaluation_json['scores']
if 'three_year_feasibility_pct' in scores:
tags.append(f"3yr_feasibility:{scores['three_year_feasibility_pct']}%")
if 'task_formalization' in scores:
tags.append(f"task_formalization:{scores['task_formalization']}/4")
if 'data_resource_availability' in scores:
tags.append(f"data_availability:{scores['data_resource_availability']}/4")
evaluation_tags = ",".join(tags) if tags else None
except Exception as e:
logger.warning(f"Warning: Could not parse evaluation_content as JSON: {e}")
# Save to database
await db.update_paper_evaluation(
arxiv_id=state.arxiv_id,
evaluation_content=evaluation_content,
evaluation_score=evaluation_score,
overall_score=overall_score,
evaluation_tags=evaluation_tags
)
state.response_text += f"\n\nEvaluation saved to database for paper: {state.arxiv_id}"
except Exception as e:
state.response_text += f"\n\nError saving evaluation to database: {str(e)}"
return state
def build_graph(api_key: Optional[str] = None):
"""Build the evaluation graph"""
graph = StateGraph(ConversationState)
evaluator = Evaluator(api_key=api_key)
graph.add_node("evaluate", evaluator)
graph.add_node("save", save_node)
# Define the flow
graph.set_entry_point("evaluate")
graph.add_edge("evaluate", "save")
graph.add_edge("save", END)
return graph.compile()
async def run_evaluation(pdf_path: str, arxiv_id: Optional[str] = None, output_file: Optional[str] = None, api_key: Optional[str] = None) -> str:
app = build_graph(api_key=api_key)
initial = ConversationState(pdf_path=pdf_path, arxiv_id=arxiv_id, output_file=output_file)
# Ensure compatibility with LangGraph's dict-based state
final_state = await app.ainvoke(initial.model_dump())
if isinstance(final_state, dict):
return str(final_state.get("response_text", ""))
if isinstance(final_state, ConversationState):
return final_state.response_text
return str(getattr(final_state, "response_text", ""))
|