---
base_model: unsloth/meta-llama-3.1-8b-bnb-4bit
library_name: peft
pipeline_tag: text-generation
tags:
- base_model:adapter:unsloth/meta-llama-3.1-8b-bnb-4bit
- lora
- transformers
- unsloth
- legal
- mock-trial
---

# Llama-3.1-8B-Mock-Trial-v3

This model is a fine-tuned version of **Meta-Llama-3.1-8B**, specifically optimized for legal discovery, witness statement analysis, and fact-retrieval in **Mock Trial** contexts. 

---

### 🚀 Try the Live Interface!
If you want to experience the logic of this project in a high-speed production environment, check out the **Mock Trial AI V3 Space**:
👉 [**Mock Trial AI V3 on Hugging Face Spaces**](https://huggingface.co/spaces/hobbesthecomputerscientist/mock-trial-v3)

> **Note on Architecture:** To ensure sub-second latency and 24/7 availability, the live Production Space utilizes the **Llama 3.1 Base Model** accelerated by the **Groq LPU**, using **Instruction Distillation** derived from this fine-tuning research. These weights remain available here for developers looking to run the specialized LoRA adapter locally.

**If you find these weights or the dataset useful, please consider leaving a ❤️ Like on this repository!**

---

## Model Details

### Model Description
- **Developed by:** HobbesTheComputerScientist
- **Model type:** LoRA Adapter for Llama 3.1 8B
- **Language(s) (NLP):** English
- **License:** MIT
- **Finetuned from model:** unsloth/meta-llama-3.1-8b-bnb-4bit

## Uses

### Direct Use
The model is intended to assist in Mock Trial preparation by:
- **Direct & Cross-Ex Simulation:** Practice examinations against adaptive deponent personas.
- **Fact Extraction:** Summarizing key legal facts from 6-12 page witness statements.
- **Theory Analysis:** Evaluating how evidence benefits or hurdles a specific Case Theory.
- **Objection Filtering:** Checking the admissibility of questions based on the Rules of Evidence.

### Out-of-Scope Use
This model is an **educational tool** designed for mock trial simulations. It should **not** be used for actual legal advice or in real-court proceedings.

## Bias, Risks, and Limitations
As a fine-tuned LLM, the model may still produce hallucinations. Users should always verify AI-generated analysis against the official case packet "Ground Truth."

## How to Get Started with the Model

To use this model with **Unsloth**, use the following code:

```python
from unsloth import FastLanguageModel
import torch

model, tokenizer = FastLanguageModel.from_pretrained(
    model_name = "hobbesthecomputerscientist/Llama-3.1-8B-Mock-Trial-v3",
    max_seq_length = 2048,
    load_in_4bit = True,
)
FastLanguageModel.for_inference(model)