File size: 3,657 Bytes
bbd3a81 | 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 | ---
license: apache-2.0
tags:
- ohio-law
- family-law
- juvenile-court
- pro-se
- legal
- unsloth
- lora
- mlx
- gemma-4
base_model: mlx-community/gemma-4-12B-it-4bit
datasets:
- Roderick3rd/hcjf-legal-forms
language:
- en
library_name: mlx
pipeline_tag: text-generation
---
# ProSe-Ohio-Legal-Expert
**One father. Two heartbreaks. Free AI so no parent faces the system alone.**
A LoRA adapter fine-tuned on Gemma 4 12B for Ohio family law, Hamilton County Juvenile Court procedures, and pro se litigation support.
## Description
This model is a LoRA adapter trained on 67 Hamilton County Juvenile Court forms (HCJF) covering custody, shared parenting, paternity, support, appeals, power of attorney, and more. It provides accurate citations to the Ohio Revised Code and local court rules, helping pro se litigants understand court forms, filing procedures, and legal standards.
**Trained by:** Rodney D. Mullins, pro se litigant in F25-000196Z
**Purpose:** Level the playing field for parents who can't afford $500/hr attorneys
## Training
| Detail | Value |
|--------|-------|
| **Base model** | `mlx-community/gemma-4-12B-it-4bit` (Gemma 4 12B, 4-bit quantized) |
| **Technique** | LoRA (rank 8, 16 layers) via MLX on Apple Silicon |
| **Dataset** | [Roderick3rd/hcjf-legal-forms](https://huggingface.co/datasets/Roderick3rd/hcjf-legal-forms) — 67 HCJF forms in ChatML format |
| **Hardware** | Apple M4 Pro (48GB unified memory) |
| **Training time** | ~50 minutes (170 iterations) |
| **Final loss** | 0.560 (train), 1.591 (validation) |
| **Trainable params** | 10.93M / 11,907M total (0.092%) |
## Usage
### With MLX (Apple Silicon)
```python
from mlx_lm import load, generate
# Load base model + adapter
model, tokenizer = load(
"mlx-community/gemma-4-12B-it-4bit",
adapter_path="ProSeAI/ProSe-Ohio-Legal-Expert"
)
# Generate
messages = [
{"role": "system", "content": "You are an Ohio family law expert specializing in Hamilton County Juvenile Court procedures. You provide accurate citations to the Ohio Revised Code and local court rules."},
{"role": "user", "content": "What forms do I need to file for custody in Hamilton County Juvenile Court?"},
]
prompt = tokenizer.apply_chat_template(messages, tokenize=False, add_generation_prompt=True)
response = generate(model, tokenizer, prompt, max_tokens=512, temperature=0.3)
print(response)
```
### With Ollama
```bash
# Coming soon — will be available as:
# ollama pull hf.co/ProSeAI/ProSe-Ohio-Legal-Expert
```
## Capabilities
- **Ohio family law** — ORC 3109.04 (custody), ORC 3111 (paternity), local juvenile court rules
- **Form assistance** — which form for which situation, how to complete each field
- **Pro se guidance** — filing procedures, deadlines, court protocols
- **Best interest factors** — Ohio's 10-factor analysis for custody determinations
- **GAL procedures** — Guardian ad Litem reports, motions, questionnaires
## Limitations
- This is a **LoRA adapter**, not a standalone model. It requires the Gemma 4 12B base model.
- Trained on 67 HCJF forms — strong on Hamilton County procedure, weaker on other Ohio counties
- Not a substitute for licensed legal counsel
- Always verify citations against current Ohio law
## Links
- [Training Notebook on GitHub](https://github.com/Pro-SeAI/sovereign-training-notebooks/blob/main/ProSe-Ohio-Legal-Expert.ipynb)
- [Training Dataset](https://huggingface.co/datasets/Roderick3rd/hcjf-legal-forms)
- [ProSe AI Model Portfolio](https://huggingface.co/ProSeAI)
- [GitHub Organization](https://github.com/Pro-SeAI)
## License
Apache 2.0 — free for any use, commercial or otherwise.
|