# GPT4All-Model (Hanuman)

This folder contains the files needed to load and run the custom Hanuman model.

Included files:

- `pytorch_model.bin` — model weights
- `config.json` — model configuration
- `tokenizer.json`, `tokenizer_config.json`, `special_tokens_map.json` — tokenizer files
- `modeling.py` — custom `Hanuman` model implementation
- `hanuman_loader.py` — convenience loader (optional)

Quick usage (local files in this folder):

```python
# inference_local.py
from transformers import AutoTokenizer
from modeling import Hanuman
import torch

# load tokenizer from local folder
tokenizer = AutoTokenizer.from_pretrained('.')
# load model using the provided helper
model = Hanuman.from_pretrained('.', map_location='cpu')

prompt = "สวัสดีครับ ช่วยอธิบายสั้น ๆ เกี่ยวกับประเทศไทย"
inputs = tokenizer(prompt, return_tensors='pt')
outputs = model.generate(inputs['input_ids'], max_new_tokens=50, temperature=1.2, top_k=50, top_p=0.95)
print(tokenizer.decode(outputs[0], skip_special_tokens=True))
```

Or load from the Hugging Face Hub (if this folder was uploaded to the hub as the repo root):

```python
# inference_from_hub.py
from transformers import AutoTokenizer
from hanuman_loader import HanumanModel

repo_id = "ZombitX64/GPT4All-Model"
# tokenizer will download from HF
tokenizer = AutoTokenizer.from_pretrained(repo_id)
# HanumanModel downloads weights and modeling.py dynamically
model_wrapper = HanumanModel.from_pretrained(repo_id, map_location='cpu')
model = model_wrapper.model

prompt = "สวัสดีครับ ช่วยสรุปประเทศไทยสั้น ๆ"
inputs = tokenizer(prompt, return_tensors='pt')
outputs = model.generate(inputs['input_ids'], max_new_tokens=50)
print(tokenizer.decode(outputs[0], skip_special_tokens=True))
```

Notes:
- This repo uses a custom model class (`Hanuman`) — users must keep `modeling.py` or use the provided `hanuman_loader.py` that dynamically imports it.
- For CPU inference, install a CPU build of PyTorch. For GPU, install the appropriate CUDA-enabled PyTorch.