llama3-8b-vietnamese-law-6k

llama3-8b-vietnamese-law-6k

Model Description

This is a Llama-3-8B model fine-tuned on Vietnamese legal documents (6k samples).

Model Details

• Base Model: unsloth/llama-3-8b-Instruct-bnb-4bit
• Training Data: 6,000 Vietnamese legal Q&A samples
• Fine-tuning Method: LoRA (Low-Rank Adaptation)
• Training Epochs: 3
• Batch Size: 16 (effective)

Usage

from unsloth import FastLanguageModel

# Load model
model, tokenizer = FastLanguageModel.from_pretrained(
    "phamff/llama3-8b-vietnamese-law-6k",
    max_seq_length=1024,
    dtype=None,
    load_in_4bit=True,
)

# Example usage
prompt = "Câu hỏi: Quy định về thời hiệu khởi kiện trong luật dân sự là gì?\nTrả lời:"
inputs = tokenizer(prompt, return_tensors="pt")
outputs = model.generate(**inputs, max_new_tokens=256)
response = tokenizer.decode(outputs[0], skip_special_tokens=True)
print(response)

Performance

The model shows good performance on Vietnamese legal questions. See test_results.json for detailed evaluation.

License

Please check the original Llama-3 license terms.