Train and deploy task-specialized LLMs with PEFT/LoRA and structured inference.

LMTask is a configuration-driven framework for dataset preparation, parameter-efficient supervised fine-tuning, structured inference, and deployment of task-specialized language models.

• Overview
  • Table of Contents
  • Why LMTask?
    • How LMTask is different
  • Features
    • Model specialization
    • Task definition
    • Data preparation
    • Reliable inference
    • Model testing and benchmarking
  • Installation
    • Runtime expectations
  • Quick start
  • Specialize a model with PEFT/LoRA SFT
  • Test and benchmark a specialized model
    • Compare and report benchmark results
  • Python API
  • Use an LMTask task in an agentic workflow
  • Dataset configuration
  • Supported model configurations
  • Documentation
  • Evidence and reproducibility
  • Alternatives
  • Community
  • Changelog
  • License
• Configuration
  • Composition
  • Dataset contract
  • Training and inference templates
  • Held-out test source
  • Task-specific overrides
• Inference
  • Inference contract
  • Request and response handling
  • Held-out inference
  • Application and agent use
• Training
  • Training lifecycle
  • Training, validation, and test data
  • PEFT and trainer settings
  • Persisted training state
• API Reference
  • zensols.lmtask package
    • Submodules
    • zensols.lmtask.app module
    • zensols.lmtask.benchmark module
    • zensols.lmtask.cli module
    • zensols.lmtask.dataset module
    • zensols.lmtask.gemma4 module
    • zensols.lmtask.generate module
    • zensols.lmtask.hf module
    • zensols.lmtask.instruct module
    • zensols.lmtask.metric module
    • zensols.lmtask.proto module
    • zensols.lmtask.task module
    • zensols.lmtask.test module
    • zensols.lmtask.torchconfig module
    • zensols.lmtask.torchtype module
    • zensols.lmtask.train module
    • Module contents
• Contributing
  • Report an issue
  • Add a model integration
  • Add a task example
  • Pull requests
  • Development workflow
  • Code of conduct
  • Guidelines
• License

Indices and tables

• Index

• Module Index

• Search Page