Nimbus-Chat-Mini
1. Introduction
Nimbus-Chat-Mini is a compact assistant for on-device dialogue. Despite its size it shows strong instruction following after our distillation pipeline.
2. Evaluation Results
Comprehensive Benchmark Results
| Benchmark | TinyTalk | Nimbus-0.5 | Breeze | Nimbus-Chat-Mini | |
|---|---|---|---|---|---|
| Core Reasoning Tasks | Math Reasoning | 0.603 | 0.621 | 0.590 | 0.633 |
| Logical Reasoning | 0.798 | 0.835 | 0.794 | 0.850 | |
| Common Sense | 0.774 | 0.774 | 0.766 | 0.783 | |
| Language Understanding | Reading Comprehension | 0.714 | 0.719 | 0.735 | 0.760 |
| Question Answering | 0.593 | 0.640 | 0.607 | 0.649 | |
| Text Classification | 0.807 | 0.822 | 0.830 | 0.849 | |
| Sentiment Analysis | 0.806 | 0.762 | 0.772 | 0.818 | |
| Generation Tasks | Code Generation | 0.684 | 0.713 | 0.688 | 0.730 |
| Creative Writing | 0.653 | 0.655 | 0.689 | 0.700 | |
| Dialogue Generation | 0.681 | 0.647 | 0.664 | 0.702 | |
| Summarization | 0.763 | 0.790 | 0.780 | 0.805 | |
| Specialized Capabilities | Translation | 0.788 | 0.768 | 0.803 | 0.826 |
| Knowledge Retrieval | 0.685 | 0.679 | 0.684 | 0.718 | |
| Instruction Following | 0.771 | 0.740 | 0.776 | 0.798 | |
| Safety Evaluation | 0.728 | 0.767 | 0.729 | 0.779 |
Overall Performance Summary
The Nimbus-Chat-Mini demonstrates strong performance across all evaluated benchmark categories, with particularly notable results in reasoning and generation tasks.
3. Chat Website & API Platform
We offer a chat interface and API for you to interact with Nimbus-Chat-Mini. Please check our official website for more details.
4. How to Run Locally
Please refer to our code repository for more information about running Nimbus-Chat-Mini locally.
Temperature
We recommend setting the temperature parameter to 0.6.
5. License
This repository is released under the apache-2.0 license. The model supports commercial use.
6. Contact
If you have any questions, please contact us at team@nimbus.chat.