DSPy: The framework for programming—not prompting—language models
Python
retrievers import patches
Arnav Singhvi91ef3ccc3f6c
diff --git a/dspy/retrieve/mongodb_atlas_rm.py b/dspy/retrieve/mongodb_atlas_rm.pyindex 41542a507..89be15dda 100644--- a/dspy/retrieve/mongodb_atlas_rm.py+++ b/dspy/retrieve/mongodb_atlas_rm.py@@ -11,6 +11,7 @@ from openai import ()from dsp.utils.settings import settings+from dspy import Retrieve, Predictiontry:from pymongo import MongoClient@@ -68,7 +69,7 @@ class Embedder:return [result.embedding for result in embedding.data]-class MongoDBAtlasRM(dspy.Retrieve):+class MongoDBAtlasRM(Retrieve):def __init__(self,db_name: str,@@ -107,7 +108,7 @@ class MongoDBAtlasRM(dspy.Retrieve):self.embedder = Embedder(provider=embedding_provider, model=embedding_model)- def forward(self, query_or_queries: str) -> dspy.Prediction:+ def forward(self, query_or_queries: str) -> Prediction:query_vector = self.embedder([query_or_queries])pipeline = build_vector_search_pipeline(index_name=self.index_name,@@ -116,4 +117,4 @@ class MongoDBAtlasRM(dspy.Retrieve):limit=self.k,)contents = self.client[self.db_name][self.collection_name].aggregate(pipeline)- return dspy.Prediction(passages=list(contents))+ return Prediction(passages=list(contents))diff --git a/dspy/retrieve/neo4j_rm.py b/dspy/retrieve/neo4j_rm.pyindex 43c9024da..f71dbb1cb 100644--- a/dspy/retrieve/neo4j_rm.py+++ b/dspy/retrieve/neo4j_rm.py@@ -10,6 +10,7 @@ from openai import (UnprocessableEntityError,)+from dspy import Retrieve, Predictionfrom dsp.utils.settings import settingsfrom dsp.utils import dotdict@@ -52,7 +53,7 @@ class Embedder:DEFAULT_INDEX_QUERY = "CALL db.index.vector.queryNodes($index, $k, $embedding) YIELD node, score "-class Neo4jRM(dspy.Retrieve):+class Neo4jRM(Retrieve):"""Implements a retriever that utilizes Neo4j for retrieving passages.This class manages a connection to a Neo4j database using official Neo4j Python drivers and requires@@ -137,7 +138,7 @@ class Neo4jRM(dspy.Retrieve):self.embedder = Embedder(provider=embedding_provider, model=embedding_model)- def forward(self, query_or_queries: Union[str, List[str]], k: Optional[int]) -> dspy.Prediction:+ def forward(self, query_or_queries: Union[str, List[str]], k: Optional[int]) -> Prediction:if not isinstance(query_or_queries, list):query_or_queries = [query_or_queries]query_vectors = self.embedder(query_or_queries)diff --git a/dspy/retrieve/pinecone_rm.py b/dspy/retrieve/pinecone_rm.pyindex fe65bf23a..0b5518e96 100644--- a/dspy/retrieve/pinecone_rm.py+++ b/dspy/retrieve/pinecone_rm.py@@ -7,6 +7,7 @@ from typing import List, Optional, Unionimport backoff+from dspy import Retrieve, Predictionfrom dsp.utils.settings import settingsfrom dsp.utils import dotdict@@ -33,7 +34,7 @@ try:except Exception:ERRORS = (openai.RateLimitError, openai.APIError)-class PineconeRM(dspy.Retrieve):+class PineconeRM(Retrieve):"""A retrieval module that uses Pinecone to return the top passages for a given query.@@ -54,7 +55,7 @@ class PineconeRM(dspy.Retrieve):dspy.Prediction: An object containing the retrieved passages.Examples:- Below is a code snippet that shows how to use this as the default retriver:+ Below is a code snippet that shows how to use this as the default retriever:```pythonllm = dspy.OpenAI(model="gpt-3.5-turbo")retriever_model = PineconeRM(openai.api_key)@@ -222,7 +223,7 @@ class PineconeRM(dspy.Retrieve):# we need a pooling strategy to get a single vector representation of the input# so the default is to take the mean of the hidden states- def forward(self, query_or_queries: Union[str, List[str]]) -> dspy.Prediction:+ def forward(self, query_or_queries: Union[str, List[str]]) -> Prediction:"""Search with pinecone for self.k top passages for queryArgs:@@ -251,7 +252,7 @@ class PineconeRM(dspy.Retrieve):)passages = [result["metadata"]["text"] for result in sorted_results]passages = [dotdict({"long_text": passage for passage in passages})]- return dspy.Prediction(passages=passages)+ return Prediction(passages=passages)# For multiple queries, query each and return the highest scoring passages# If a passage is returned multiple times, the score is accumulated. For this reason we increase top_k by 3x@@ -269,4 +270,4 @@ class PineconeRM(dspy.Retrieve):sorted_passages = sorted(passage_scores.items(), key=lambda x: x[1], reverse=True,)[: self.k]- return dspy.Prediction(passages=[dotdict({"long_text": passage}) for passage, _ in sorted_passages])+ return Prediction(passages=[dotdict({"long_text": passage}) for passage, _ in sorted_passages])