File size: 9,999 Bytes
6c5ce7a |
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 |
import argparse
from dataloaders import (
ARCDataloader,
EdgarDataLoader,
FactScoreDataloader,
PopQADataloader,
TriviaQADataloader,
)
from rag_pipelines.embeddings.dense import DenseEmbeddings
from rag_pipelines.embeddings.sparse import SparseEmbeddings
from rag_pipelines.evaluation import (
AnswerRelevancyScorer,
ContextualPrecisionScorer,
ContextualRecallScorer,
ContextualRelevancyScorer,
Evaluator,
FaithfulnessScorer,
HallucinationScorer,
SummarizationScorer,
)
from rag_pipelines.evaluation.evaluator import Evaluator
from rag_pipelines.llms.groq import ChatGroqGenerator
from rag_pipelines.pipelines.crag import CorrectiveRAGPipeline
from rag_pipelines.retrieval_evaluator.document_grader import DocumentGrader
from rag_pipelines.retrieval_evaluator.retrieval_evaluator import RetrievalEvaluator
from rag_pipelines.vectordb.pinecone_hybrid_index import PineconeHybridVectorDB
from rag_pipelines.vectordb.pinecone_hybrid_retriever import PineconeHybridRetriever
SUPPORTED_DATASETS = {
"arc": ARCDataloader,
"edgar": EdgarDataLoader,
"popqa": PopQADataloader,
"factscore": FactScoreDataloader,
"triviaqa": TriviaQADataloader,
}
SCORER_CLASSES = {
"contextual_precision": ContextualPrecisionScorer,
"contextual_recall": ContextualRecallScorer,
"contextual_relevancy": ContextualRelevancyScorer,
"answer_relevancy": AnswerRelevancyScorer,
"faithfulness": FaithfulnessScorer,
"summarization": SummarizationScorer,
"hallucination": HallucinationScorer,
}
def main():
parser = argparse.ArgumentParser(description="Run the Corrective RAG pipeline.")
# Dense embeddings arguments
parser.add_argument(
"--dense_model_name",
type=str,
default="sentence-transformers/all-MiniLM-L6-v2",
help="Dense embedding model name.",
)
parser.add_argument(
"--device",
type=str,
default="cpu",
help="Device to run the dense embedding model.",
)
# Sparse embeddings arguments
parser.add_argument(
"--sparse_max_seq_length",
type=int,
default=512,
help="Maximum sequence length for sparse embeddings.",
)
# Pinecone arguments
parser.add_argument("--pinecone_api_key", type=str, required=True, help="Pinecone API key.")
parser.add_argument("--index_name", type=str, default="edgar", help="Pinecone index name.")
parser.add_argument("--dimension", type=int, default=384, help="Dimension of embeddings.")
parser.add_argument("--metric", type=str, default="dotproduct", help="Metric for similarity search.")
parser.add_argument("--region", type=str, default="us-east-1", help="Pinecone region.")
parser.add_argument(
"--namespace",
type=str,
default="edgar-all",
help="Namespace for Pinecone retriever.",
)
# Retriever arguments
parser.add_argument("--alpha", type=float, default=0.5, help="Alpha parameter for hybrid retriever.")
parser.add_argument("--top_k", type=int, default=5, help="Number of top documents to retrieve.")
# LLM arguments
parser.add_argument(
"--llm_model",
type=str,
default="llama-3.2-90b-vision-preview",
help="Language model name.",
)
parser.add_argument(
"--temperature",
type=float,
default=0,
help="Temperature for the language model.",
)
parser.add_argument("--llm_api_key", type=str, required=True, help="API key for the language model.")
# Retrieval Evaluator and Document Grader arguments
parser.add_argument(
"--relevance_threshold",
type=float,
default=0.7,
help="Relevance threshold for document grading.",
)
# Load evaluation data
parser = argparse.ArgumentParser(description="Load evaluation dataset and initialize the dataloader.")
parser.add_argument(
"--dataset_type",
type=str,
default="edgar",
choices=SUPPORTED_DATASETS.keys(),
help="Dataset to load from. Options: arc, edgar, popqa, factscore, triviaqa.",
)
parser.add_argument(
"--hf_dataset_path",
type=str,
default="lamini/earnings-calls-qa",
help="Path to the HuggingFace dataset.",
)
parser.add_argument(
"--dataset_split",
type=str,
default="test",
help="Split of the dataset to use (e.g., train, validation, test).",
)
# Scorer arguments
parser.add_argument(
"--scorer",
type=str,
default="contextual_precision",
choices=[
"contextual_precision",
"contextual_recall",
"contextual_relevancy",
"answer_relevancy",
"faithfulness",
"summarization",
"hallucination",
],
help="Scorer to use.",
)
# Evaluation arguments
parser.add_argument(
"--evaluation_name",
type=str,
default="hybrid_rag",
help="Name of the evaluation.",
)
# Add argument for selecting scorers
parser.add_argument(
"--scorers",
type=str,
nargs="+",
choices=SCORER_CLASSES.keys(),
required=True,
help="List of scorers to use. Options: contextual_precision, contextual_recall, contextual_relevancy, "
"answer_relevancy, faithfulness, summarization, hallucination.",
)
# Add shared arguments for scorer parameters
parser.add_argument("--threshold", type=float, default=0.5, help="Threshold for evaluation.")
parser.add_argument("--model", type=str, default="gpt-4", help="Model to use for scoring.")
parser.add_argument("--include_reason", action="store_true", help="Include reasons in scoring.")
parser.add_argument(
"--assessment_questions",
type=str,
nargs="*",
help="List of assessment questions for scoring.",
)
parser.add_argument("--strict_mode", action="store_true", help="Enable strict mode for scoring.")
parser.add_argument("--async_mode", action="store_true", help="Enable asynchronous processing.")
parser.add_argument("--verbose", action="store_true", help="Enable verbose output.")
parser.add_argument(
"--truths_extraction_limit",
type=int,
default=None,
help="Limit for truth extraction in scoring.",
)
args = parser.parse_args()
# Initialize dataloader based on the dataset type
try:
DataLoaderClass = SUPPORTED_DATASETS[args.dataset_type]
dataloader = DataLoaderClass(dataset_name=args.hf_dataset_path, split=args.dataset_split)
except KeyError:
msg = (
f"Dataset '{args.dataset_type}' is not supported. "
f"Supported options are: {', '.join(SUPPORTED_DATASETS.keys())}."
)
raise ValueError(msg)
eval_dataset = dataloader.get_eval_data()
# Initialize embeddings
dense_embeddings = DenseEmbeddings(
model_name=args.dense_model_name,
model_kwargs={"device": args.device},
encode_kwargs={"normalize_embeddings": True},
show_progress=True,
)
sparse_embeddings = SparseEmbeddings(model_kwargs={"max_seq_length": args.sparse_max_seq_length})
dense_embedding_model = dense_embeddings.embedding_model
sparse_embedding_model = sparse_embeddings.sparse_embedding_model
# Initialize Pinecone vector DB
pinecone_vector_db = PineconeHybridVectorDB(
api_key=args.pinecone_api_key,
index_name=args.index_name,
dimension=args.dimension,
metric=args.metric,
region=args.region,
)
# Initialize Pinecone retriever
pinecone_retriever = PineconeHybridRetriever(
index=pinecone_vector_db.index,
dense_embedding_model=dense_embedding_model,
sparse_embedding_model=sparse_embedding_model,
alpha=args.alpha,
top_k=args.top_k,
namespace=args.namespace,
)
# Initialize RetrievalEvaluator and DocumentGrader
retrieval_evaluator = RetrievalEvaluator(
llm_model=args.llm_model,
llm_api_key=args.llm_api_key,
temperature=args.temperature,
)
document_grader = DocumentGrader(
evaluator=retrieval_evaluator,
threshold=args.relevance_threshold,
)
# Load the prompt and initialize the generator
generator = ChatGroqGenerator(
model=args.llm_model,
api_key=args.llm_api_key,
llm_params={"temperature": args.temperature},
)
llm = generator.llm
# Initialize the Corrective RAG pipeline
corrective_rag = CorrectiveRAGPipeline(
retriever=pinecone_retriever.hybrid_retriever,
prompt=retrieval_evaluator.prompt_template,
llm=llm,
document_grader=document_grader,
tracing_project_name="sec_corrective_rag",
)
# Initialize the scorers with the provided arguments
scorers = []
for scorer_name in args.scorers:
if scorer_name in SCORER_CLASSES:
ScorerClass = SCORER_CLASSES[scorer_name]
scorer = ScorerClass(
threshold=args.threshold,
model=args.model,
include_reason=args.include_reason,
assessment_questions=args.assessment_questions,
strict_mode=args.strict_mode,
async_mode=args.async_mode,
verbose=args.verbose,
truths_extraction_limit=args.truths_extraction_limit,
)
scorers.append(scorer)
else:
msg = f"Scorer '{scorer_name}' is not supported."
raise ValueError(msg)
# Run the pipeline
evaluator = Evaluator(
evaluation_name=args.evaluation_name,
pipeline=corrective_rag,
dataset=eval_dataset,
scorers=[scorers],
)
evaluation_results = evaluator.evaluate()
print(evaluation_results)
if __name__ == "__main__":
main()
|