File size: 1,280 Bytes
6595ce6
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
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
# humanize-e Adapter

A QLoRA LoRA adapter for Qwen2.5-3B-Instruct trained to humanize AI-generated text.

## Model Details

- **Base Model:** Qwen/Qwen2.5-3B-Instruct
- **Fine-tuning Method:** QLoRA (LoRA r=32, alpha=64)
- **Training Data:** 156 synthetic pairs from Kaggle AI vs Human Text dataset
- **Developer:** Mike Elistra (Mike Koci)
- **Training Notebook:** https://www.kaggle.com/code/mick009/notebook9c4e543da9

## Usage

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

base_model = "Qwen/Qwen2.5-3B-Instruct"
adapter_path = "mike-elistra/humanize-e-adapter"

tokenizer = AutoTokenizer.from_pretrained(adapter_path)
model = AutoModelForCausalLM.from_pretrained(
    base_model,
    quantization_config=BitsAndBytesConfig(load_in_4bit=True, bnb_4bit_quant_type="nf4"),
    device_map="auto"
)
model = PeftModel.from_pretrained(model, adapter_path)
model.eval()

# Use with humanize-e inference code
```

## Inference API

The adapter works with the humanize-e FastAPI service:
```bash
HUMANIZE_ADAPTER=mike-elistra/humanize-e-adapter python app.py
```

## Training

See Kaggle notebook for full training pipeline:
https://www.kaggle.com/code/mick009/notebook9c4e543da9

## License

MIT