Trouble Shooting

Common issues and quick fixes for local NELA workflows.

Indexing is slow

• Start with a small folder to verify end-to-end ingestion first.
• Use a lighter embedding model while validating your pipeline.
• Keep enough free disk space for chunk/index growth.
• First ingestion is the slowest; repeat queries are usually faster after index warmup.
• If retrieval quality suddenly drops after changing embedding model family, rebuild the index from scratch.

Model download fails

Most failures come from network interruptions, insufficient disk, or incompatible model choice.

• Check internet/proxy access while downloading.
• Verify free space in the models directory target.
• Try a smaller model first to confirm your environment is healthy.
• Confirm the model appears as installed before selecting it in a mode.

If running from source with custom model storage, verify your path:

echo $GENHAT_MODEL_PATH