Abliteration HandbookField guide
Research & reference · source archive

Quick Start (training only)

SFT (1 epoch, QLoRA defaults in train_sft.py):

1 min read 3 examples Source-linked

In this chapter
On this page Interactive companion
Personal reading progress

0/0 sections reviewed

View original Markdown
Interactive companion

Explore the primary records

Use the visual controls alongside the complete source chapter.

bash
cd sources/jarvis-pack/jarvis-tool-repair-pack
python -m venv .venv && .\.venv\Scripts\activate
pip install -r requirements.txt
python ../../../scripts/validate-dataset.py

SFT (1 epoch, QLoRA defaults in train_sft.py):

bash
python scripts/train_sft.py \
  --base_model /path/to/abliterated-model \
  --data data/sft_train.jsonl \
  --output_dir output/security-agent-sft

DPO (0.5–1 epoch):

bash
python scripts/train_dpo.py \
  --base_model /path/to/abliterated-model \
  --adapter_path output/security-agent-sft \
  --data data/dpo_train.jsonl \
  --output_dir output/security-agent-v1

Deploy with ../../../../scripts/hardware-tool-gate.py — see ../../../../instructions/agentic-security-stack.md.

Trace this chapter to its source.

The complete Markdown remains the editable source. This page adds navigation, typography, and a visual companion.

sources/jarvis-pack/jarvis-tool-repair-pack/docs/QUICK_START.md
Inspect the source fingerprintSHA-256 a2450982abfcb3a70e7a14c08fe7d3c82ad5247f32eb1bbfbdc7df191b99d814Source commit 6fba7436a5914cad7c8f0835a14b480a29dac263
Connected reading

Follow the ideas into the next chapter.