---
language:
  - zh
license: other
tags:
  - mistral
  - lora
  - sft
  - chat
  - bf16
  - taiwan
base_model: mistralai/Mistral-7B-v0.1
pipeline_tag: text-generation
library_name: transformers
---

# 數學推理 LLM (Merged)

> **重要聲明 (必讀)**
> 本模型涉及原住民族文化、知識與表述內容。**未經創作者 Aciang Iku-Silan 明確書面同意**，禁止下載、複製、再散布或用於任何研究與商業用途。
> 詳見本倉庫 `RIGHTS_NOTICE.md`。

本倉庫提供**合併後 (單權重)** 的模型檔，用於繁體中文任務 (教學、文化、知識問答等混合 SFT)。
訓練設備：A100 40GB (bf16)，packing=True，max_seq_length=512。

## 快速開始
```python
# Install libraries (adjust versions as needed)
!pip install transformers peft trl accelerate torch bitsandbytes datasets

# Load model and tokenizer
from transformers import AutoTokenizer, AutoModelForCausalLM
from peft import PeftModel # Use PeftModel if this is a LoRA adapter
import torch

model_id = "aciang/mistral7b-tk-sft-20251019-merged"
base_model_id = "mistralai/Mistral-7B-v0.1"

# For merged model:
if is_merged:
    tok = AutoTokenizer.from_pretrained(model_id)
    model = AutoModelForCausalLM.from_pretrained(model_id, torch_dtype=torch.bfloat16, device_map='auto')
else:
    # For LoRA adapter:
    tok = AutoTokenizer.from_pretrained(base_model_id)
    base_model = AutoModelForCausalLM.from_pretrained(base_model_id, torch_dtype=torch.bfloat16, device_map='auto')
    model = PeftModel.from_pretrained(base_model, model_id)

model.eval()

# Example Inference
prompt = "請用三點條列解釋為何梯度累積可以在小批次下達到大批次效果："
inputs = tok(prompt, return_tensors='pt').to(model.device)

with torch.no_grad():
    outputs = model.generate(**inputs, max_new_tokens=256, temperature=0.7, top_p=0.9, do_sample=True)

print(tok.decode(outputs[0], skip_special_tokens=True))
```

## 範例任務
以下是模型可能擅長的任務類型範例：
數學解題、文化知識問答



---
Built with ❤️ and 繁體中文