labtasker

# Labtasker

> Labtasker is a small, Python-native task queue for running independent machine-learning inference, evaluation, and experiment jobs in parallel.

These documents describe Labtasker v2. Labtasker schedules Tasks across Worker processes; it does not allocate GPUs, manage a cluster, build workflow DAGs, or store large artifacts. The specification is the authoritative product and protocol contract.

## Start here

- [README](https://raw.githubusercontent.com/luocfprime/labtasker/refs/heads/main/README.md): Product motivation, comparison with project-specific scripts, installation, and a compact end-to-end example.
- [Run your first experiment](https://raw.githubusercontent.com/luocfprime/labtasker/refs/heads/main/docs/getting-started.md): Install Labtasker, submit several experiment cases, and process them through one Queue.
- [Tested demo](https://raw.githubusercontent.com/luocfprime/labtasker/refs/heads/main/docs/demo.md): Run the canonical submission and Python Worker example whose source files are exercised end to end.
- [Why Labtasker?](https://raw.githubusercontent.com/luocfprime/labtasker/refs/heads/main/docs/why-labtasker.md): Decide whether Labtasker fits and understand the core design changes from v1 to v2.
- [How Labtasker works](https://raw.githubusercontent.com/luocfprime/labtasker/refs/heads/main/docs/concepts.md): The Task, Queue, Worker, Client, Server, route, and execution model.
- [Agent Skill](https://raw.githubusercontent.com/luocfprime/labtasker/refs/heads/main/docs/guides/agent-skill.md): How coding agents should discover and use the bundled Labtasker workflow.

## Examples

- [Inference and evaluation patterns](https://raw.githubusercontent.com/luocfprime/labtasker/refs/heads/main/docs/inference-evaluation.md): Patterns for parallel ML inference and evaluation.
- [Embodied AI case study](https://raw.githubusercontent.com/luocfprime/labtasker/refs/heads/main/docs/case-studies/starvla-robotwin.md): Applying Labtasker to RoboTwin evaluation in the StarVLA codebase.

## Workflows

- [Manage Tasks](https://raw.githubusercontent.com/luocfprime/labtasker/refs/heads/main/docs/guides/tasks.md): Submit, inspect, prioritize, cancel, retry, and delete Tasks.
- [Query Tasks](https://raw.githubusercontent.com/luocfprime/labtasker/refs/heads/main/docs/guides/query.md): Filter, search names, paginate, and group Task counts; compare demand with Worker observations.
- [Failure and recovery](https://raw.githubusercontent.com/luocfprime/labtasker/refs/heads/main/docs/guides/failure-recovery.md): Retries, leases, interruption, and protection from results reported by old runs.
- [Python Workers](https://raw.githubusercontent.com/luocfprime/labtasker/refs/heads/main/docs/workers/python.md): Reuse loaded Python state while processing Tasks.
- [Command Workers](https://raw.githubusercontent.com/luocfprime/labtasker/refs/heads/main/docs/workers/command.md): Execute Tasks as subprocess commands.
- [Distributed Workers](https://raw.githubusercontent.com/luocfprime/labtasker/refs/heads/main/docs/workers/distributed.md): Run one Task through a supported distributed launcher.

## Reference

- [Python API](https://raw.githubusercontent.com/luocfprime/labtasker/refs/heads/main/docs/reference/python-api.md): Signatures, return models, lifecycle rules, pagination, errors, and retries.
- [CLI](https://raw.githubusercontent.com/luocfprime/labtasker/refs/heads/main/docs/reference/cli.md): Command contracts, JSON input, stdout/stderr, exit codes, and pagination.
- [Configuration](https://raw.githubusercontent.com/luocfprime/labtasker/refs/heads/main/docs/reference/configuration.md): Configuration files, environment variables, precedence, and deployment modes.
- [HTTP API](https://raw.githubusercontent.com/luocfprime/labtasker/refs/heads/main/docs/reference/http-api.md): Endpoint methods, bodies, status codes, authentication, errors, and run fencing.
- [Specification](https://raw.githubusercontent.com/luocfprime/labtasker/refs/heads/main/docs/reference/specification.md): Authoritative behavior, lifecycle, protocol, persistence, and invariants.

## Optional

- [Use the Web UI](https://raw.githubusercontent.com/luocfprime/labtasker/refs/heads/main/docs/guides/webui.md): Launch the optional Labtasker WebUI, connect to an existing Server or local project, and inspect progress and results in a browser.
- [Development](https://raw.githubusercontent.com/luocfprime/labtasker/refs/heads/main/docs/development.md): Repository setup, validation, and contribution workflow.