File size: 7,633 Bytes
4c67792
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
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
"""
OpenIE extraction (NER + triple extraction) using QAFD-RAG's LLM functions.

Follows the original openie_openai.py logic but calls the async LLM wrappers
from ``QAFD-RAG/src/llm.py`` synchronously via ``asyncio.run``.
"""

import asyncio
import json
import logging
import re
from concurrent.futures import ThreadPoolExecutor, as_completed
from dataclasses import dataclass
from typing import Dict, Any, List, Tuple, TypedDict, Callable

from tqdm import tqdm

from .prompts import make_ner_messages, make_triple_messages
from .utils import (
    NerRawOutput,
    TripleRawOutput,
    fix_broken_generated_json,
    filter_invalid_triples,
)

logger = logging.getLogger(__name__)


class ChunkInfo(TypedDict):
    num_tokens: int
    content: str


def _run_sync(coro):
    """Run async coroutine from sync context."""
    try:
        loop = asyncio.get_running_loop()
    except RuntimeError:
        loop = None

    if loop is not None and loop.is_running():
        import concurrent.futures
        with concurrent.futures.ThreadPoolExecutor(max_workers=1) as pool:
            return pool.submit(asyncio.run, coro).result()
    else:
        return asyncio.run(coro)


def _extract_ner_from_response(response_text: str) -> List[str]:
    pattern = r'\{[^{}]*"named_entities"\s*:\s*\[[^\]]*\][^{}]*\}'
    match = re.search(pattern, response_text, re.DOTALL)
    if match is None:
        return []
    try:
        return eval(match.group())["named_entities"]
    except Exception:
        return []


def _extract_triples_from_response(response_text: str) -> List[List[str]]:
    pattern = r'\{[^{}]*"triples"\s*:\s*\[[^\]]*\][^{}]*\}'
    match = re.search(pattern, response_text, re.DOTALL)
    if match is None:
        return []
    try:
        return eval(match.group())["triples"]
    except Exception:
        return []


class OpenIE:
    """Synchronous OpenIE using QAFD-RAG's async LLM function.

    Parameters
    ----------
    llm_func : callable
        An async function with the signature::

            async def llm_func(prompt, system_prompt=None,
                               history_messages=[], **kwargs) -> str

        Typically one of the ``gpt_*_complete`` helpers from ``src/llm.py``.
    """

    def __init__(self, llm_func: Callable):
        self.llm_func = llm_func

    def _call_llm(self, messages: List[Dict[str, str]]) -> str:
        """Convert chat messages to a single LLM call."""
        system_prompt = None
        history = []
        user_prompt = ""
        for msg in messages:
            if msg["role"] == "system":
                system_prompt = msg["content"]
            elif msg["role"] == "assistant":
                history.append(msg)
            elif msg["role"] == "user":
                # All user messages except the last go into history
                if user_prompt:
                    history.append({"role": "user", "content": user_prompt})
                user_prompt = msg["content"]

        return _run_sync(
            self.llm_func(
                prompt=user_prompt,
                system_prompt=system_prompt,
                history_messages=history,
                max_tokens=2048,
            )
        )

    # ------------------------------------------------------------------
    def ner(self, chunk_key: str, passage: str) -> NerRawOutput:
        messages = make_ner_messages(passage)
        raw_response = ""
        metadata: Dict[str, Any] = {}
        try:
            raw_response = self._call_llm(messages)
            real_response = fix_broken_generated_json(raw_response)
            extracted = _extract_ner_from_response(real_response)
            unique_entities = list(dict.fromkeys(extracted))
        except Exception as e:
            logger.warning(f"NER error for chunk {chunk_key}: {e}")
            metadata["error"] = str(e)
            return NerRawOutput(
                chunk_id=chunk_key,
                response=raw_response,
                unique_entities=[],
                metadata=metadata,
            )

        return NerRawOutput(
            chunk_id=chunk_key,
            response=raw_response,
            unique_entities=unique_entities,
            metadata=metadata,
        )

    # ------------------------------------------------------------------
    def triple_extraction(
        self, chunk_key: str, passage: str, named_entities: List[str]
    ) -> TripleRawOutput:
        messages = make_triple_messages(passage, named_entities)
        raw_response = ""
        metadata: Dict[str, Any] = {}
        try:
            raw_response = self._call_llm(messages)
            real_response = fix_broken_generated_json(raw_response)
            extracted = _extract_triples_from_response(real_response)
            triplets = filter_invalid_triples(triples=extracted)
        except Exception as e:
            logger.warning(f"Triple extraction error for chunk {chunk_key}: {e}")
            metadata["error"] = str(e)
            return TripleRawOutput(
                chunk_id=chunk_key,
                response=raw_response,
                metadata=metadata,
                triples=[],
            )

        return TripleRawOutput(
            chunk_id=chunk_key,
            response=raw_response,
            metadata=metadata,
            triples=triplets,
        )

    # ------------------------------------------------------------------
    def openie(self, chunk_key: str, passage: str) -> Dict[str, Any]:
        ner_output = self.ner(chunk_key=chunk_key, passage=passage)
        triple_output = self.triple_extraction(
            chunk_key=chunk_key,
            passage=passage,
            named_entities=ner_output.unique_entities,
        )
        return {"ner": ner_output, "triplets": triple_output}

    # ------------------------------------------------------------------
    def batch_openie(
        self, chunks: Dict[str, dict]
    ) -> Tuple[Dict[str, NerRawOutput], Dict[str, TripleRawOutput]]:
        """Run NER + triple extraction over all chunks using multithreading.

        Parameters
        ----------
        chunks : dict
            Mapping ``chunk_hash_id -> {"content": text, ...}``.

        Returns
        -------
        (ner_dict, triple_dict)
        """
        chunk_passages = {k: v["content"] for k, v in chunks.items()}

        # ---- NER pass ----
        ner_results: List[NerRawOutput] = []

        with ThreadPoolExecutor() as executor:
            ner_futures = {
                executor.submit(self.ner, ckey, passage): ckey
                for ckey, passage in chunk_passages.items()
            }
            for future in tqdm(
                as_completed(ner_futures), total=len(ner_futures), desc="NER"
            ):
                ner_results.append(future.result())

        # ---- Triple extraction pass ----
        triple_results: List[TripleRawOutput] = []

        with ThreadPoolExecutor() as executor:
            re_futures = {
                executor.submit(
                    self.triple_extraction,
                    nr.chunk_id,
                    chunk_passages[nr.chunk_id],
                    nr.unique_entities,
                ): nr.chunk_id
                for nr in ner_results
            }
            for future in tqdm(
                as_completed(re_futures),
                total=len(re_futures),
                desc="Triple extraction",
            ):
                triple_results.append(future.result())

        ner_dict = {r.chunk_id: r for r in ner_results}
        triple_dict = {r.chunk_id: r for r in triple_results}
        return ner_dict, triple_dict