AWQ is an efficient, accurate and blazing-fast low-bit weight quantization method, currently supporting 4-bit quantization. Compared to GPTQ, it offers faster Transformers-based inference with equivalent or better quality compared to the most commonly used GPTQ settings.
AWQ models are currently supported on Linux and Windows, with NVidia GPUs only. macOS users: please use GGUF models instead.
Taiwan LLM is an advanced language model tailored for Traditional Chinese, focusing on the linguistic and cultural contexts of Taiwan.
Developed from a large base model, it's enriched with diverse Taiwanese textual sources and refined through Supervised Fine-Tuning.
This model excels in language understanding and generation, aligning closely with Taiwan's cultural nuances.
It demonstrates improved performance on various benchmarks like TC-Eval, showcasing its contextual comprehension and cultural relevance.
For detailed insights into Taiwan LLM's development and features, refer to our technical report.
Model description
Model type: A 13B parameter GPT-like model fine-tuned on a mix of publicly available, synthetic datasets.
Language(s) (NLP): Primarily Traditional Chinese (zh-tw)
Here's how you can run the model using the pipeline() function from 🤗 Transformers:
python
1# pip install transformers>=4.342# pip install accelerate34import torch
5from transformers import pipeline
67pipe = pipeline("text-generation", model="yentinglin/Taiwan-LLM-13B-v2.0-chat", torch_dtype=torch.bfloat16, device_map="auto")89# We use the tokenizer's chat template to format each message - see https://huggingface.co/docs/transformers/main/en/chat_templating10messages =[11{12"role":"system",13"content":"你是一個人工智慧助理",14},15{"role":"user","content":"東北季風如何影響台灣氣候？"},16]17prompt = pipe.tokenizer.apply_chat_template(messages, tokenize=False, add_generation_prompt=True)18outputs = pipe(prompt, max_new_tokens=256, do_sample=True, temperature=0.7, top_k=50, top_p=0.95)19print(outputs[0]["generated_text"])
Training hyperparameters
image/png
image/png
image/png
The following hyperparameters were used during training:
learning_rate: 5e-05
distributed_type: multi-GPU
optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
lr_scheduler_type: cosine
lr_scheduler_warmup_ratio: 0.03
num_epochs: 5.0
Citation
If you find Taiwan LLM is useful in your work, please cite it with:
@misc{lin2023taiwan,
title={Taiwan LLM: Bridging the Linguistic Divide with a Culturally Aligned Language Model},
author={Yen-Ting Lin and Yun-Nung Chen},
year={2023},
eprint={2311.17487},
archivePrefix={arXiv},
primaryClass={cs.CL}
}
Acknowledgement
Taiwan LLM v2 is conducted in collaboration with Ubitus K.K.. Ubitus provides valuable compute resources for the project.
Disclaimer
This model is provided “as‑is” and without warranties of any kind. Users are solely responsible for evaluating the accuracy and suitability of the outputs. The developers assume no liability for any direct or indirect damages arising from its use.
The model is strictly not intended for high‑risk applications such as medical diagnosis, legal advice, or financial investment. For such use cases, please consult qualified professionals.
