Langboat's Mengzi3-8B-Chat is an 8 billion parameter conversational language model designed for general-purpose chat applications. It achieves competitive performance across various benchmarks, including 63.9 on MMLU and 75.4 on GSM8k, making it suitable for tasks requiring reasoning and mathematical abilities. The model is open-sourced under the Apache 2.0 license, supporting both academic research and free commercial use.
Loading preview...
Mengzi3-8B-Chat Overview
Langboat's Mengzi3-8B-Chat is an 8 billion parameter conversational large language model. It is designed to serve as a helpful assistant, capable of engaging in general chat interactions. The model is released under the Apache 2.0 license, allowing for both academic research and free commercial applications.
Key Capabilities & Performance
Mengzi3-8B-Chat demonstrates solid performance across a range of common LLM benchmarks:
- MMLU (5-shot): 63.9
- GSM8k (4-shot): 75.4
- MATH (4-shot): 24.5
- HumanEval: 62.2
- MT-Bench: 8.19
- AlignBench: 6.96
These scores indicate its proficiency in multi-task language understanding, mathematical reasoning, and coding tasks, alongside its conversational abilities.
Usage and Licensing
The model can be easily integrated using the Hugging Face transformers library for inference. Detailed code examples for both inference and fine-tuning are available on the Langboat Github repository. Langboat provides Mengzi3-8B-Chat "as is" and disclaims responsibility for any issues arising from its use, urging users to adhere to ethical and legal guidelines. Commercial use is permitted, with options to contact Langboat for specific business licenses or collaborations.
Top 3 parameter combinations used by Featherless users for this model. Click a tab to see each config.