AI Basic Law Ministral-8B (台灣人工智慧基本法)

基於 Ministral-8B-Instruct-2410 微調的台灣「人工智慧基本法」問答模型。

Model Description

  • Base Model: mistralai/Ministral-8B-Instruct-2410
  • Fine-tuning Method: QLoRA (4-bit quantization + LoRA r=64)
  • Language: Traditional Chinese (繁體中文)
  • Domain: Taiwan AI Basic Law (人工智慧基本法), 20 articles
  • Framework: Axolotl
  • Publication Date: 民國115年(2026年)1月14日

Capabilities

  • 精準引述人工智慧基本法全文 20 條
  • 正確對應條號與內容(含 topic anchor 防錯位)
  • 識別簡稱(「基本法」「AI基本法」→ 人工智慧基本法)
  • 預設台灣法律語境,不與中國法律混淆
  • 回答公布日期、主管機關、立法理由等相關問題

Usage

With Transformers

from transformers import AutoModelForCausalLM, AutoTokenizer
import torch

model_id = "iamjry/ai-basic-law-ministral8b"
tokenizer = AutoTokenizer.from_pretrained(model_id)
model = AutoModelForCausalLM.from_pretrained(
    model_id, torch_dtype=torch.float16, device_map="auto"
)

system_prompt = "你是一位熟悉台灣人工智慧基本法的法律助理,請根據法條內容精確回答問題。"
question = "請列出人工智慧基本法第3條全文"

messages = [{"role": "user", "content": f"{system_prompt}\n\n{question}"}]
prompt = tokenizer.apply_chat_template(messages, tokenize=False, add_generation_prompt=True)
inputs = tokenizer(prompt, return_tensors="pt").to(model.device)

with torch.no_grad():
    outputs = model.generate(**inputs, max_new_tokens=500, repetition_penalty=1.3, do_sample=False)

generated = outputs[0][inputs["input_ids"].shape[1]:]
print(tokenizer.decode(generated, skip_special_tokens=True))

With Ollama (GGUF)

See ai-basic-law-ministral8b-GGUF for quantized GGUF versions.

Training Details

  • Epochs: 15
  • LoRA Rank: 64, Alpha: 32
  • Learning Rate: 1e-4, Cosine scheduler
  • Optimizer: AdamW 8-bit
  • Sequence Length: 2048 (with sample packing)
  • Training Data: ~5000 entries (with oversampling for key facts)
  • Monitoring: Weights & Biases

Training Data Composition

Category Description
Article Fulltext Verbatim text of all 20 articles with topic anchors
Alias Recognition 「基本法」「AI基本法」→ 人工智慧基本法
Legislative Reasons Background and rationale for each article
Taiwan Context Anti-hallucination data to prevent confusion with Chinese laws
Publication Date 民國115年(2026年)1月14日, 華總一義字第11500001671號令

Limitations

  • Optimized specifically for Taiwan AI Basic Law; not a general legal assistant
  • Some longer articles (e.g., Article 4, 5) may have minor variations from the exact original text
  • Best results when using the system prompt: 你是一位熟悉台灣人工智慧基本法的法律助理,請根據法條內容精確回答問題。

License

CC-BY-4.0

Downloads last month
6
Safetensors
Model size
8B params
Tensor type
F16
·
Inference Providers NEW
This model isn't deployed by any Inference Provider. 🙋 Ask for provider support

Model tree for iamjry/ai-basic-law-ministral8b

Finetuned
(97)
this model
Quantizations
1 model