Model Overview
The stephenlzc/dolphin-llama3-zh-cn-uncensored is an 8 billion parameter language model, developed by stephenlzc. It is built upon the cognitivecomputations/dolphin-2.9-llama3-8b base model, which itself is derived from the Llama 3 architecture.
Key Capabilities
- Chinese Language Support: The primary enhancement of this model is its ability to process and generate content in Chinese. This was achieved through fine-tuning with specific Chinese datasets.
- Uncensored Responses: As indicated by its name, the model is intended to provide uncensored outputs, making it suitable for use cases where unfiltered responses are required.
- Base Model Heritage: Inherits the general capabilities of the Dolphin-2.9-Llama3-8B base model, which is known for its instruction-following abilities.
Training Details
The model was fine-tuned using two distinct Chinese datasets:
- Minami-su/toxic-sft-zh: This dataset likely contributes to the model's ability to handle diverse and potentially sensitive topics without censorship.
- llm-wizard/alpaca-gpt4-data-zh: This dataset, based on Alpaca-GPT4 data translated to Chinese, enhances the model's instruction-following and general conversational abilities in Chinese.
Good For
- Applications requiring a large language model with strong Chinese language capabilities.
- Use cases where uncensored or unfiltered responses are acceptable or desired.
- Developers looking for a Llama 3-based model with specific Chinese fine-tuning.