Updated default embeddings (#58)

* Update 03_agent_proof_of_concept.py

Changed vector search defaults

* Update 03_agent_proof_of_concept.py

Added comment

* Update 02_data_pipeline.py

Changed default embeddings

* Update 03_agent_proof_of_concept.py

change back to ann by default

---------

Co-authored-by: Eric Peter <epeter@gmail.com>

Author: edurdevic

agent_app_sample_code/02_data_pipeline.py

@@ -97,9 +97,7 @@
 
 # Embedding model endpoint. The list of off-the-shelf embeddings can be found here:
 # https://docs.databricks.com/en/machine-learning/foundation-models/index.html
-EMBEDDING_MODEL_ENDPOINT = "databricks-bge-large-en"
-# EMBEDDING_MODEL_ENDPOINT = "ep-embeddings-small"
-# EMBEDDING_MODEL_ENDPOINT = "bge-test"
+EMBEDDING_MODEL_ENDPOINT = "databricks-gte-large-en"
 
 # COMMAND ----------
 

agent_app_sample_code/03_agent_proof_of_concept.py

@@ -45,7 +45,7 @@
         additional_metadata_columns=[],
     ),
     parameters=RetrieverParametersConfig(num_results=5, query_type="ann"),
-    vector_search_threshold=0.1,
+    vector_search_threshold=0.0, # Increase this to make the search more restrictive
     chunk_template="Passage text: {chunk_text}\nPassage metadata: {metadata}\n\n",
     prompt_template="""Use the following pieces of retrieved context to answer the question.\nOnly use the passages from context that are relevant to the query to answer the question, ignore the irrelevant passages.  When responding, cite your source, referring to the passage by the columns in the passage's metadata.\n\nContext: {context}""",
     tool_description_prompt="Search for documents that are relevant to a user's query about the [REPLACE WITH DESCRIPTION OF YOUR DOCS].",