Kellogg LLM Batch

User guide

  • Installation
    • Requirements
    • Clone and install
    • Contributor tools
    • Provider credentials
    • Next step
  • Basic workflow
    • 1. Create a project
      • Optional: use the project builder
    • 2. Validate locally
    • 3. Prepare and inspect a pilot
    • 4. Submit the pilot
    • 5. Prepare the complete batch
    • 6. Wait for and process the batch
    • 7. Use the results
  • Project settings
    • Project identity
    • Input data
    • Task and response schema
    • Prompts and codebook
    • Providers and budget
      • openai_compatible: self-hosted Ollama and vLLM servers
    • Evaluation and output
  • Grant-coding walkthrough
    • What the example does
    • 1. Validate the project
    • 2. Prepare a four-grant pilot
    • 3. Inspect before submitting
    • 4. Submit the pilot
    • 5. Review the results
    • 6. Continue to the complete batch
    • 7. Run with Anthropic
    • 8. Compare providers
    • Where to go next
  • Advanced workflows
    • Retry failed records and merge the results
    • Cancel and recover a run
      • Cancel and continue safely
      • Other recovery commands
    • Intentionally rerun or select specific records
    • Compare providers or repeated runs
    • Run against a local Ollama or vLLM server
    • Configure a project in depth
      • Input mappings
      • Prompts and context
      • Place individual columns in the prompt
      • Schema portability
      • Limits, models, and cost
      • Stage batch segments
    • Understand artifacts and provenance
    • Protect research data
  • Attach one PNG or PDF per prepared request
    • Use the checked-in helper scripts
  • CLI reference
    • Prepare options used most often
    • Command boundaries
Kellogg LLM Batch
  • Search

© Copyright 2026, Kellogg Research Support.

Built with Sphinx using a theme provided by Read the Docs.