Spaces:
Runtime error
Runtime error
File size: 10,646 Bytes
129cd69 |
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 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283 284 285 286 287 288 289 290 291 292 293 294 295 296 297 298 299 300 301 302 303 304 305 306 307 308 309 310 311 312 313 314 315 316 317 318 319 320 321 322 323 324 325 326 327 328 329 330 331 332 333 334 335 336 337 338 339 340 341 342 343 344 345 346 |
"""LLM Chains for evaluating question answering."""
from __future__ import annotations
import re
import string
from typing import Any, List, Optional, Sequence, Tuple
from langchain_core.language_models import BaseLanguageModel
from langchain_core.prompts import PromptTemplate
from langchain_core.pydantic_v1 import Extra
from langchain.callbacks.manager import Callbacks
from langchain.chains.llm import LLMChain
from langchain.evaluation.qa.eval_prompt import CONTEXT_PROMPT, COT_PROMPT, PROMPT
from langchain.evaluation.schema import LLMEvalChain, StringEvaluator
from langchain.schema import RUN_KEY
def _get_score(text: str) -> Optional[Tuple[str, int]]:
match = re.search(r"grade:\s*(correct|incorrect)", text.strip(), re.IGNORECASE)
if match:
if match.group(1).upper() == "CORRECT":
return "CORRECT", 1
elif match.group(1).upper() == "INCORRECT":
return "INCORRECT", 0
try:
first_word = (
text.strip().split()[0].translate(str.maketrans("", "", string.punctuation))
)
if first_word.upper() == "CORRECT":
return "CORRECT", 1
elif first_word.upper() == "INCORRECT":
return "INCORRECT", 0
last_word = (
text.strip()
.split()[-1]
.translate(str.maketrans("", "", string.punctuation))
)
if last_word.upper() == "CORRECT":
return "CORRECT", 1
elif last_word.upper() == "INCORRECT":
return "INCORRECT", 0
except IndexError:
pass
return None
def _parse_string_eval_output(text: str) -> dict:
"""Parse the output text.
Args:
text (str): The output text to parse.
Returns:
Any: The parsed output.
"""
reasoning = text.strip()
parsed_scores = _get_score(reasoning)
if parsed_scores is None:
value, score = None, None
else:
value, score = parsed_scores
return {
"reasoning": reasoning,
"value": value,
"score": score,
}
class QAEvalChain(LLMChain, StringEvaluator, LLMEvalChain):
"""LLM Chain for evaluating question answering."""
output_key: str = "results" #: :meta private:
class Config:
"""Configuration for the QAEvalChain."""
extra = Extra.ignore
@property
def evaluation_name(self) -> str:
return "correctness"
@property
def requires_reference(self) -> bool:
return True
@property
def requires_input(self) -> bool:
return True
@classmethod
def from_llm(
cls,
llm: BaseLanguageModel,
prompt: Optional[PromptTemplate] = None,
**kwargs: Any,
) -> QAEvalChain:
"""Load QA Eval Chain from LLM.
Args:
llm (BaseLanguageModel): the base language model to use.
prompt (PromptTemplate): A prompt template containing the input_variables:
'input', 'answer' and 'result' that will be used as the prompt
for evaluation.
Defaults to PROMPT.
**kwargs: additional keyword arguments.
Returns:
QAEvalChain: the loaded QA eval chain.
"""
prompt = prompt or PROMPT
expected_input_vars = {"query", "answer", "result"}
if expected_input_vars != set(prompt.input_variables):
raise ValueError(
f"Input variables should be {expected_input_vars}, "
f"but got {prompt.input_variables}"
)
return cls(llm=llm, prompt=prompt, **kwargs)
def evaluate(
self,
examples: Sequence[dict],
predictions: Sequence[dict],
question_key: str = "query",
answer_key: str = "answer",
prediction_key: str = "result",
*,
callbacks: Callbacks = None,
) -> List[dict]:
"""Evaluate question answering examples and predictions."""
inputs = [
{
"query": example[question_key],
"answer": example[answer_key],
"result": predictions[i][prediction_key],
}
for i, example in enumerate(examples)
]
return self.apply(inputs, callbacks=callbacks)
def _prepare_output(self, result: dict) -> dict:
parsed_result = _parse_string_eval_output(result[self.output_key])
if RUN_KEY in result:
parsed_result[RUN_KEY] = result[RUN_KEY]
return parsed_result
def _evaluate_strings(
self,
*,
prediction: str,
reference: Optional[str] = None,
input: Optional[str] = None,
callbacks: Callbacks = None,
include_run_info: bool = False,
**kwargs: Any,
) -> dict:
"""Evaluate Chain or LLM output, based on optional input and label.
Args:
prediction (str): the LLM or chain prediction to evaluate.
reference (Optional[str], optional): the reference label
to evaluate against.
input (Optional[str], optional): the input to consider during evaluation
callbacks (Callbacks, optional): the callbacks to use for tracing.
include_run_info (bool, optional): whether to include run info in the
returned results.
**kwargs: additional keyword arguments, including callbacks, tags, etc.
Returns:
dict: The evaluation results containing the score or value.
"""
result = self(
{
"query": input,
"answer": reference,
"result": prediction,
},
callbacks=callbacks,
include_run_info=include_run_info,
)
return self._prepare_output(result)
async def _aevaluate_strings(
self,
*,
prediction: str,
reference: Optional[str] = None,
input: Optional[str] = None,
callbacks: Callbacks = None,
include_run_info: bool = False,
**kwargs: Any,
) -> dict:
result = await self.acall(
inputs={"query": input, "answer": reference, "result": prediction},
callbacks=callbacks,
include_run_info=include_run_info,
)
return self._prepare_output(result)
class ContextQAEvalChain(LLMChain, StringEvaluator, LLMEvalChain):
"""LLM Chain for evaluating QA w/o GT based on context"""
@property
def requires_reference(self) -> bool:
"""Whether the chain requires a reference string."""
return True
@property
def requires_input(self) -> bool:
"""Whether the chain requires an input string."""
return True
class Config:
"""Configuration for the QAEvalChain."""
extra = Extra.ignore
@classmethod
def _validate_input_vars(cls, prompt: PromptTemplate) -> None:
expected_input_vars = {"query", "context", "result"}
if expected_input_vars != set(prompt.input_variables):
raise ValueError(
f"Input variables should be {expected_input_vars}, "
f"but got {prompt.input_variables}"
)
@property
def evaluation_name(self) -> str:
return "Contextual Accuracy"
@classmethod
def from_llm(
cls,
llm: BaseLanguageModel,
prompt: Optional[PromptTemplate] = None,
**kwargs: Any,
) -> ContextQAEvalChain:
"""Load QA Eval Chain from LLM.
Args:
llm (BaseLanguageModel): the base language model to use.
prompt (PromptTemplate): A prompt template containing the input_variables:
'query', 'context' and 'result' that will be used as the prompt
for evaluation.
Defaults to PROMPT.
**kwargs: additional keyword arguments.
Returns:
ContextQAEvalChain: the loaded QA eval chain.
"""
prompt = prompt or CONTEXT_PROMPT
cls._validate_input_vars(prompt)
return cls(llm=llm, prompt=prompt, **kwargs)
def evaluate(
self,
examples: List[dict],
predictions: List[dict],
question_key: str = "query",
context_key: str = "context",
prediction_key: str = "result",
*,
callbacks: Callbacks = None,
) -> List[dict]:
"""Evaluate question answering examples and predictions."""
inputs = [
{
"query": example[question_key],
"context": example[context_key],
"result": predictions[i][prediction_key],
}
for i, example in enumerate(examples)
]
return self.apply(inputs, callbacks=callbacks)
def _prepare_output(self, result: dict) -> dict:
parsed_result = _parse_string_eval_output(result[self.output_key])
if RUN_KEY in result:
parsed_result[RUN_KEY] = result[RUN_KEY]
return parsed_result
def _evaluate_strings(
self,
*,
prediction: str,
reference: Optional[str] = None,
input: Optional[str] = None,
callbacks: Callbacks = None,
include_run_info: bool = False,
**kwargs: Any,
) -> dict:
result = self(
{
"query": input,
"context": reference,
"result": prediction,
},
callbacks=callbacks,
include_run_info=include_run_info,
)
return self._prepare_output(result)
async def _aevaluate_strings(
self,
*,
prediction: str,
reference: Optional[str] = None,
input: Optional[str] = None,
callbacks: Callbacks = None,
include_run_info: bool = False,
**kwargs: Any,
) -> dict:
result = await self.acall(
inputs={"query": input, "context": reference, "result": prediction},
callbacks=callbacks,
include_run_info=include_run_info,
)
return self._prepare_output(result)
class CotQAEvalChain(ContextQAEvalChain):
"""LLM Chain for evaluating QA using chain of thought reasoning."""
@property
def evaluation_name(self) -> str:
return "COT Contextual Accuracy"
@classmethod
def from_llm(
cls,
llm: BaseLanguageModel,
prompt: Optional[PromptTemplate] = None,
**kwargs: Any,
) -> CotQAEvalChain:
"""Load QA Eval Chain from LLM."""
prompt = prompt or COT_PROMPT
cls._validate_input_vars(prompt)
return cls(llm=llm, prompt=prompt, **kwargs)
|