Batch Inference

Generate embeddings for a dataset

Run offline batch embedding generation over a set of text records — for building a vector index, search, or RAG pipeline — in a single job with one receipt.

Estimated $0.0005–$0.003 per record · Scales with record count — typically 1,000–5,000 records/min

What inputs you need

  • Text records (CSV/JSONL)
  • Embedding model choice
  • Output format (vectors + IDs)

What Badgr returns

  • Vector embeddings keyed by record ID
  • Batch run log
  • Single receipt on completion

Recommended GPU routes

Estimated cost: $0.0005–$0.003 per record · Estimated runtime: Scales with record count — typically 1,000–5,000 records/min

Example command

badgr run "python batch_embed.py --input records.jsonl --model bge-large" \
  --gpu L40S \
  --max-cost 8 \
  --max-runtime 45

Common setup failures Badgr avoids

Dimension mismatch when swapping embedding models mid-project — output includes the model name and vector dimension in the manifest so downstream indexes catch mismatches early.

Throughput lower than expected — increase batch size to saturate GPU utilization, or move to a higher-tier GPU route.

Ready to run this?

Launch from the dashboard, CLI, or Compute API. Max-cost protection included.