CLI Reference

All commands are available via the easylora CLI.

easylora --help

easylora train

Fine-tune a model with LoRA / QLoRA.

easylora train --config config.yaml [OPTIONS]

Option	Description
--config, -c	Path to YAML/JSON config file (required unless --autopilot)
--autopilot	Auto-plan training from model + dataset
--model, -m	Base model ID/path (required with --autopilot)
--dataset, -d	Dataset path or HF name (required with --autopilot)
--quality	Autopilot quality preset: fast, balanced, high
--output-dir	Output directory used in autopilot mode
--subset	Dataset subset/config name for autopilot mode
--split	Dataset split for autopilot mode (default: train)
--seed	Seed for reproducibility
--force	Allow overwriting output directory
--set	Override config value as key=value (repeatable)
--dry-run	Validate config and print settings without training
--print-config	Print resolved config as YAML and exit

Examples:

# Basic training
easylora train --config config.yaml

# Autopilot training (no config file)
easylora train \
    --autopilot \
    --model meta-llama/Llama-3.2-1B \
    --dataset tatsu-lab/alpaca

# Override learning rate
easylora train --config config.yaml --set optim.lr=1e-5

# Validate without training
easylora train --config config.yaml --dry-run

# Overwrite existing output
easylora train --config config.yaml --force

easylora autopilot plan

Generate and print autopilot decisions without training.

easylora autopilot plan [OPTIONS]

Option	Description
--model, -m	Base model ID/path (required)
--dataset, -d	Dataset path or HF name (required)
--quality	Preset: fast, balanced, high
--output-dir	Planned output directory
--subset	Dataset subset/config name
--split	Dataset split (default: train)
--print-config	Print resolved TrainConfig YAML

Example:

easylora autopilot plan \
    --model meta-llama/Llama-3.2-1B \
    --dataset tatsu-lab/alpaca \
    --quality balanced \
    --print-config

easylora eval

Evaluate a LoRA adapter with perplexity and optional generation.

easylora eval [OPTIONS]

Option	Description
--base-model, -m	Base model ID or path (required)
--adapter-dir, -a	Path to saved adapter (required)
--dataset, -d	Dataset path or HF name (required)
--max-samples	Max samples to evaluate
--max-seq-len	Max sequence length (default: 2048)
--prompt, -p	Prompts for generation sanity check (repeatable)

Example:

easylora eval \
    --base-model meta-llama/Llama-3.2-1B \
    --adapter-dir ./output/adapter \
    --dataset eval_data.jsonl \
    --prompt "What is machine learning?"

easylora merge

Merge a LoRA adapter into the base model and save full weights.

easylora merge [OPTIONS]

Option	Description
--base-model, -m	Base model ID or path (required)
--adapter-dir, -a	Path to saved adapter (required)
--output-dir, -o	Where to save merged model (required)
--trust-remote-code	Trust remote code

Example:

easylora merge \
    --base-model meta-llama/Llama-3.2-1B \
    --adapter-dir ./output/adapter \
    --output-dir ./merged_model

easylora doctor

Print environment diagnostics for debugging.

easylora doctor

Shows: Python version, PyTorch version, CUDA availability, GPU name, bf16 support, bitsandbytes availability, and versions of key dependencies.

easylora inspect-targets

Show recommended LoRA target modules for a model.

easylora inspect-targets --model <model_id> [--trust-remote-code]

Option	Description
--model, -m	HF model ID or path (required)
--trust-remote-code	Trust remote code

Example:

easylora inspect-targets --model meta-llama/Llama-3.2-1B

easylora init-config

Generate a starter config file.

easylora init-config --template <template_name> [--output config.yaml]

Option	Description
--template, -t	Template name: sft-lora or sft-qlora (required)
--output, -o	Output file path (default: easylora_config.yaml)

Example:

easylora init-config --template sft-qlora