Batch Inference

Evaluate a prompt set against a model

Run offline batch evaluation of a prompt/test-case set against a model — scoring outputs against expected answers or a rubric — in a single job with one receipt.

Run this jobGet help running this

Estimated $0.005–$0.03 per test case, depending on scoring method · Scales with test-case count

What inputs you need

  • Prompt/test-case set (JSONL with expected outputs or rubric)
  • Model to evaluate
  • Scoring method (exact match, rubric, judge model)

What Badgr returns

  • Per-case scores and model outputs
  • Aggregate score summary
  • Single receipt on completion

Recommended GPU routes

A100 40GBL40S 48GBA40 48GB

Estimated cost: $0.005–$0.03 per test case, depending on scoring method · Estimated runtime: Scales with test-case count

Example command

badgr run "python batch_eval.py --input testcases.jsonl --model qwen-7b" \
  --gpu A100_40GB \
  --max-cost 15 \
  --max-runtime 90

Common setup failures Badgr avoids

Job stops partway through a large eval set — progress is checkpointed per test case so a restart resumes instead of reprocessing.

Max-cost hit before the full eval set completes — estimate cost per case from a small sample run before submitting the full set.

Ready to run this?

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

Run this jobBack to Batch Inference