💡 All-in-one AI framework for semantic search, LLM orchestration and language model workflows
Python agentsaiai-agentsembeddingsinformation-retrievallanguage-modellarge-language-modelsllmnlppythonragretrieval-augmented-generationsearchsearch-enginesemantic-searchsentence-embeddingstransformerstxtaivector-databasevector-search
Update benchmark scripts to support Similarity pipeline, closes #955
davidmezzetti6fcdcb1e3881
diff --git a/examples/benchmarks.py b/examples/benchmarks.pyindex fc48bb2f..3a7f3831 100644--- a/examples/benchmarks.py+++ b/examples/benchmarks.py@@ -26,7 +26,7 @@ from rank_bm25 import BM25Okapifrom tqdm.auto import tqdmfrom txtai.embeddings import Embeddings-from txtai.pipeline import LLM, RAG, Tokenizer+from txtai.pipeline import LLM, RAG, Similarity, Tokenizerfrom txtai.scoring import ScoringFactory@@ -275,6 +275,34 @@ class Score(Index):return scoring+class Similar(Index):+ """+ Similarity pipeline.+ """++ def index(self):+ # Load similarity pipeline+ model = Similarity(**self.readconfig("similar", {}))++ # Get datasets+ data = list(self.rows())+ ids = [x[0] for x in data]+ texts = [x[1] for x in data]++ return (ids, texts, model)++ def search(self, queries, limit):+ # Unpack backend+ ids, texts, model = self.backend++ # Run model inference+ results = []+ for result in model(queries, texts, limit=limit):+ results.append([(ids[x], score) for x, score in result])++ return results++class RankBM25(Index):"""BM25 index using rank-bm25.@@ -500,6 +528,8 @@ def create(method, path, config, output, refresh):return SQLiteFTS(path, config, output, refresh)if method == "es":return Elastic(path, config, output, refresh)+ if method == "similarity":+ return Similarity(path, config, output, refresh)# Defaultreturn Embed(path, config, output, refresh)