license: apache-2.0
datasets:
- jed351/cantonese-wikipedia
- raptorkwok/cantonese-traditional-chinese-parallel-corpus
language:
- zh
- en
pipeline_tag: text-generation
tags:
- Cantonese
- Qwen2
- chat
Qwen2-Cantonese-7B-Instruct
Model Overview
Qwen2-Cantonese-7B-Instruct is a Cantonese language model based on Qwen2-7B-Instruct, fine-tuned using LoRA. It aims to enhance Cantonese text generation and comprehension capabilities, supporting various tasks such as dialogue generation, text summarization, and question-answering.
Model Features
- Base Model: Qwen2-7B-Instruct
- Fine-tuning Method: LoRA instruction tuning
- Training Steps: 4572 steps
- Primary Language: Cantonese
- Datasets:
- Training Tools: LLaMA-Factory
Usage
You can easily load and use this model with Hugging Face's Transformers library. Here is a simple example:
from transformers import AutoModelForCausalLM, AutoTokenizer
tokenizer = AutoTokenizer.from_pretrained("lordjia/Qwen2-Cantonese-7B-Instruct")
model = AutoModelForCausalLM.from_pretrained("lordjia/Qwen2-Cantonese-7B-Instruct")
input_text = "唔該你用廣東話講下你係邊個。"
inputs = tokenizer(input_text, return_tensors="pt")
outputs = model.generate(**inputs)
print(tokenizer.decode(outputs[0], skip_special_tokens=True))
Quantized Version
A 4-bit quantized version of this model is also available: qwen2-cantonese-7b-instruct-q4_0.gguf.
Alternative Model Recommendation
For an alternative, consider Llama-3-Cantonese-8B-Instruct, also fine-tuned by LordJia and based on Meta-Llama-3-8B-Instruct.
License
This model is licensed under the Apache 2.0 license. Please review the terms before use.
Contributors
- LordJia
Acknowledgements
Thanks to Hugging Face for providing the platform and tools, and to all the developers and researchers contributing to the open-source community.