Skip to content

Usage

Getting started

xlmtec --help        # list all commands
xlmtec --version     # show installed version

ai-suggest — Generate a config from plain English

Describe your fine-tuning task and get a ready-to-run YAML config:

xlmtec ai-suggest "fine-tune GPT-2 for sentiment analysis" --provider claude
xlmtec ai-suggest "qlora on llama for code generation" --provider gemini
xlmtec ai-suggest "instruction tune for customer support" --provider codex --save config.yaml

Options:

Option Default Description
--provider claude AI provider: claude, gemini, codex
--api-key env var API key (falls back to env variable)
--save Save generated YAML to a file

See AI Integrations for full provider setup.


hub — Browse HuggingFace models

xlmtec hub search "bert"
xlmtec hub search "llama" --task text-generation --limit 10
xlmtec hub search "gpt" --sort likes

Options:

Option Default Description
--task Filter by pipeline tag
--limit 10 Number of results (max 100)
--sort downloads Sort by: downloads, likes, lastModified

Info

xlmtec hub info google/bert-base-uncased
xlmtec hub info mistralai/Mistral-7B-v0.1
xlmtec hub trending
xlmtec hub trending --limit 20

See Model Hub for more.


config validate — Check a config before training

xlmtec config validate config.yaml
xlmtec config validate config.yaml --strict    # fail on warnings too

Reports all validation errors at once. Exits 0 on success, 1 on any error.


train — Fine-tune a model

Dry run (no model loading)

xlmtec train --config config.yaml --dry-run

Validates the config and prints the full training plan without loading any model.

Training

xlmtec train --config config.yaml
xlmtec train --config config.yaml --method lora

recommend — Get a method recommendation

xlmtec recommend --model-size 7b --vram 8

evaluate — Evaluate a model

xlmtec evaluate --model output/run1 --dataset data/test.jsonl

benchmark — Compare runs

xlmtec benchmark --runs output/run1 output/run2

merge — Merge LoRA adapter

xlmtec merge --adapter output/run1 --output merged_model/

upload — Upload to HuggingFace Hub

xlmtec upload --model output/run1 --repo-id your-username/my-model