oracle-ai-developer-hub

Bulk Loading & Listings Notebook

These notebooks show end-to-end VecDB ingestion flows for metadata-only, BYO vectors with IDs, and BYO vectors with auto-generated IDs. Open the notebook that matches the scenario you need:

• 01_integrated_embedding.ipynb – Bulk load metadata into a table with integrated embeddings.
• 02_byov_ids.ipynb – Bulk load bring-your-own vectors with manual IDs.
• 03_byov_auto_ids.ipynb – Bulk load bring-your-own vectors with database-generated IDs.

Embedding model prep: Scenario A assumes an embedding model already exists in VecDB. Deploy it beforehand and set EMBED_MODEL in .env to that model’s name. This scenario does not upload or load ONNX artifacts on your behalf.

Sample CSV fixtures

Fresh example datasets live under bulk_loading/data/ so you can upload them to Object Storage and reuse the signed URL in .env:

Scenario	Sample file	Columns	.env variable	Sample PAR
Integrated embeddings	bulktable_integrated_embedding.csv	METADATA, ID	INTEGRATED_CSV_URL	https://objectstorage.<region>.oraclecloud.com/p/<PAR_TOKEN>/n/<namespace>/b/<bucket>/o/sample_csv_data/bulktable_integrated_embedding.csv
BYO vectors (manual IDs)	bulktable_byov_ids.csv	ID, DENSE_VECTOR, METADATA	BYO_MANUAL_CSV_URL	https://objectstorage.<region>.oraclecloud.com/p/<PAR_TOKEN>/n/<namespace>/b/<bucket>/o/sample_csv_data/bulktable_byov_ids.csv
BYO vectors (auto IDs)	bulktable_byov_auto_ids.csv	DENSE_VECTOR, METADATA	BYO_AUTO_CSV_URL	https://objectstorage.<region>.oraclecloud.com/p/<PAR_TOKEN>/n/<namespace>/b/<bucket>/o/sample_csv_data/bulktable_byov_auto_ids.csv

How to publish the CSVs to Object Storage

1. Pick a bucket/object path. The repo already includes ready-to-upload files under bulk_loading/data/. Use predictable object names so you can rotate URLs without guessing.
2. Upload each CSV via direct curl -T. Example (repeat per file with the right filename):

   curl -T bulk_loading/data/bulktable_byov_auto_ids.csv \
     "https://objectstorage.<region>.oraclecloud.com/p/<PAR_TOKEN>/n/<namespace>/b/<bucket>/o/sample_csv_data/bulktable_byov_auto_ids.csv"
   

   Replace <PAR_TOKEN>, <namespace>, and <bucket> with your actual values (the sample URLs in the table show the expected shape).

3. Update .env. Paste the exact HTTPS URL you used for the upload into INTEGRATED_CSV_URL, BYO_MANUAL_CSV_URL, or BYO_AUTO_CSV_URL as needed. The notebooks download directly from these links.
4. Quick validation. Open the uploaded object in a browser to ensure the header row and quoting look identical to the local file. VecDB requires commas as separators, quoted vector/JSON fields, and escaped double quotes ("") inside the metadata JSON.

Each scenario walks through table provisioning, load_vectors, job inspection, pagination via list_vectors, catalog checks via list_vector_tables, and an automated cleanup at the end so the environment stays tidy.

Prerequisites

• Python 3.10+ with oracle-vecdb, python-dotenv, and pandas (a %pip install helper cell is provided).
• A .env file that provides:
• VECDB_REST_URL, VECDB_USERNAME, VECDB_PASSWORD
  • EMBED_MODEL → name of an embedding model that already exists inside VecDB (deploy it using the model-management example before running Scenario A)
  • EMBED_MODEL_URL → (optional) retained for backward compatibility; Scenario A no longer auto-loads models from Object Storage
  • INTEGRATED_CSV_URL → signed Object Storage URL for bulktable_integrated_embedding.csv
  • BYO_MANUAL_CSV_URL → signed Object Storage URL for bulktable_byov_ids.csv
  • BYO_AUTO_CSV_URL → signed Object Storage URL for bulktable_byov_auto_ids.csv
• Object Storage access to host the CSVs (public PAR URLs or buckets readable by VecDB).

Auto-ID reminder: Scenario C only produces database-generated IDs when the CSV omits the ID column (or leaves it empty). Create an ID-free dataset (see bulktable_byov_auto_ids.csv), upload it to Object Storage, and point BYO_AUTO_CSV_URL to that object before running the auto-ID cells.

Running the Notebook

1. Open the scenario notebook you need (01_..., 02_..., or 03_...) and run the setup cells to install dependencies and authenticate.
2. Execute the scenario cells in order (table creation → load → job inspection → listings).
3. Finish with the cleanup cell so the demo tables do not linger in your VecDB environment.

Run checklist

1. Upload/refresh the CSV in Object Storage.
2. Create or verify a PAR and update .env.
3. Drop/recreate the scenario table (integrated, BYO-ID, BYO-auto).
4. Execute the load job, inspect job logs/details, then move to listings and cleanup.