ChromaDB ValueError: Cannot submit more than 41,666 embeddings at once

What happened

After upgrading from chromadb 0.3.26 to 0.4.7, bulk ingestion started failing with "ValueError: Cannot submit more than 41,666 embeddings at once. Please submit your embeddings in batches of size 41,666 or less." Nothing else changed; the reporter was ingesting offline documents with langchain 0.0.274 and privateGPT-style code. The new limit was introduced because chromadb 0.4.x moved to SQLite, which limits the length of a single SQL query.

The fix that worked

This is intentional, not a bug: chromadb 0.4.x moved to SQLite, which limits the length of a single SQL query, so a hard batch cap of 41,666 embeddings was introduced around 0.4.6 (confirmed by a maintainer in the thread). Submit your embeddings in batches of 41,666 or fewer per add/upsert call; the error message states the exact cap.

In the thread ChromaDB ValueError: Cannot submit more than 41,666 embeddings at once, the question was: After upgrading from chromadb 0.3.26 to 0.4.7, bulk ingestion started failing with "ValueError: Cannot submit more than 41,666 embeddings at once. Please submit your embeddings in batches of size 41,666 or less." Nothing else changed; the reporter was ingesting offline documents with langchain 0.0.274 and privateGPT-style code. The new limit was introduced because chromadb 0.4.x moved to SQLite, which limits the length of a single SQL query. The verified resolution: This is intentional, not a bug: chromadb 0.4.x moved to SQLite, which limits the length of a single SQL query, so a hard batch cap of 41,666 embeddings was introduced around 0.4.6 (confirmed by a maintainer in the thread). Submit your embeddings in batches of 41,666 or fewer per add/upsert call; the error message states the exact cap. Original discussion: https://vectle.com/threads/thr_h05k4xUh99EeKG1fCUxshw