xlmtec

xlmtec is a command-line toolkit for fine-tuning large language models. Describe your task in plain English, get a ready-to-run config, browse HuggingFace models, and train — all from the terminal.

Features

• AI-powered config generation — describe your task, get a YAML config from Claude, Gemini, or GPT
• Model Hub browser — search and inspect HuggingFace models without leaving the terminal
• 5 fine-tuning methods — LoRA, QLoRA, Full, Instruction, DPO
• Config validation — catch errors before training starts
• Dry-run mode — preview your training plan without loading a model
• Rich terminal UI — progress bars, panels, colour output throughout

Installation

# Core (lightweight — no ML deps)
pip install xlmtec

# With training support
pip install xlmtec[ml]

# With AI suggestions (pick your provider)
pip install xlmtec[claude]    # Anthropic
pip install xlmtec[gemini]    # Google
pip install xlmtec[codex]     # OpenAI
pip install xlmtec[ai]        # All three

# Everything
pip install xlmtec[full]

Quickstart

1. Get an AI-generated config

xlmtec ai-suggest "fine-tune a small model for customer support" --provider claude

Outputs a ready-to-run YAML config and the exact command to run.

2. Browse models on HuggingFace

xlmtec hub search "bert" --task text-classification --limit 5
xlmtec hub trending
xlmtec hub info google/bert-base-uncased

3. Validate your config

xlmtec config validate config.yaml

4. Train

# Preview without loading model
xlmtec train --config config.yaml --dry-run

# Start training
xlmtec train --config config.yaml

Commands

Fine-tuning methods

AI Providers

Set your API key as an environment variable, then pass --provider:

export ANTHROPIC_API_KEY=sk-ant-...
xlmtec ai-suggest "summarise legal documents" --provider claude

export GEMINI_API_KEY=...
xlmtec ai-suggest "summarise legal documents" --provider gemini

export OPENAI_API_KEY=sk-...
xlmtec ai-suggest "summarise legal documents" --provider codex

Example config

model:
  name: gpt2

dataset:
  source: local_file
  path: data/train.jsonl

lora:
  r: 16
  alpha: 32
  target_modules: [c_attn]

training:
  output_dir: output/run1
  num_epochs: 3
  batch_size: 4
  learning_rate: 2e-4

Development

git clone https://github.com/Abdur-azure/xlmtec.git
cd xlmtec
pip install -e ".[full,dev]"
pytest tests/ -v --ignore=tests/test_integration.py

Changelog

See CHANGELOG.md for full release history.

License

MIT