Skip to content

Quick Start

Sudhanshu edited this page Nov 11, 2026 · 1 revision

Quick Start

The following validation commands are offline and do not download pretrained model weights.

python qwen_alien_romulus_finetune.py --validate-data
python LLM_Training_Fixed_v2.py --mode validate

To train, first install the corresponding separate environment described in Installation.

# Qwen LoRA (modern venv) — educational reduced-size run
python qwen_alien_romulus_finetune.py --epochs 1 --max-length 128 --batch-size 1

# Legacy venv — option A
python LLM_Training_Fixed_v2.py --mode sft-dpo

# Legacy venv — option B (fresh invocation, not after DPO)
python LLM_Training_Fixed_v2.py --mode orpo

Training requires checkpoint downloads and memory. The two workflows are not intended to share a runtime or model artifacts.

Clone this wiki locally