---
base_model: meta-llama/Llama-3.3-70B-Instruct
language:
- en
license: apache-2.0
library_name: peft
pipeline_tag: text-generation
tags:
- lora
- qlora
- sft
- peft
- transformers
- trl
- forensics
- law-enforcement
- investigative-ai
---

# ABBY — Artifact, Ballistic, and Binary Yield

> *"I use my powers for good, not evil."*
> — Abby Sciuto, *NCIS*

**An open-source LoRA adapter fine-tuned for law enforcement forensic investigators.**

ABBY is a QLoRA adapter trained on top of [Meta Llama 3.3 70B Instruct](https://huggingface.co/meta-llama/Llama-3.3-70B-Instruct), specialized for forensic analysis, evidence interpretation, ballistic assessment, digital artifact examination, and investigative reasoning.

---

## Model Details

| Field | Value |
|---|---|
| **Base Model** | [meta-llama/Llama-3.3-70B-Instruct](https://huggingface.co/meta-llama/Llama-3.3-70B-Instruct) |
| **Adapter Type** | LoRA (QLoRA, 4-bit NF4) |
| **LoRA Rank** | 64 |
| **Task** | Forensic investigation, evidence analysis, investigative reasoning |
| **Training Method** | QLoRA fine-tuning via SFTTrainer (trl) |
| **License** | Apache 2.0 |
| **Maintainer** | [Ronin48LLC](https://huggingface.co/Ronin48LLC) |

---

## Intended Use

ABBY is designed to assist **certified forensic examiners, law enforcement investigators, and legal professionals** with:

- **Digital forensics** — artifact recovery, file system analysis, metadata examination
- **Ballistic analysis** — wound pattern assessment, trajectory reconstruction
- **Binary/malware analysis** — code review, threat identification
- **Chain of custody guidance** — evidence handling best practices
- **Investigative reasoning** — connecting evidence to conclusions

**This model is intended for professional use in authorized law enforcement and forensic contexts only.**

---

## Usage

This is a PEFT LoRA adapter. Load it on top of the base model:

```python
from peft import PeftModel
from transformers import AutoModelForCausalLM, AutoTokenizer

base_model = AutoModelForCausalLM.from_pretrained(
    "meta-llama/Llama-3.3-70B-Instruct",
    load_in_4bit=True,
    device_map="auto",
)
model = PeftModel.from_pretrained(base_model, "Ronin48LLC/abby-lora-adapter")
tokenizer = AutoTokenizer.from_pretrained("meta-llama/Llama-3.3-70B-Instruct")
```

> **Note:** Access to the base model requires accepting Meta's license on HuggingFace.

---

## Training

- **Framework:** PyTorch 2.4 + Transformers + TRL + PEFT
- **Quantization:** 4-bit NF4 (bitsandbytes)
- **Hardware:** NVIDIA A100 PCIe 80GB
- **Training steps:** 78
- **Sequence length:** 4096

---

## Supporters

ABBY is community-funded. Every contribution keeps this project free and in the hands of investigators who need it.

| Donor | Amount | Note |
|---|---|---|
| Ronin 48, LLC | N/A | Founding donor |
| Anonymous | $50 | Thank you, Joe Sixpack |

*Want to support ABBY? Reach out to the maintainers.*

---

## Related Projects

| Project | Description |
|---|---|
| [SELMA](https://codeberg.org/Ronin48/SELMA) | Criminal law and statute analysis |
| [ATTICUS](https://codeberg.org/Ronin48/ATTICUS) | Legal defense reasoning |
| [BONES](https://codeberg.org/Ronin48/BONES) | Forensic pathology and osteology |
| [BRUNO](https://codeberg.org/Ronin48/BRUNO) | Field operations and tactical support |

---

## License

Fine-tuned adapter weights are licensed under **Apache 2.0**.
Base model weights are subject to the [Meta Llama 3.1 Community License](https://llama.meta.com/llama3/license/).