Vega-Instruct-13B
1. Introduction
Vega-Instruct-13B is tuned for faithful instruction following and low hallucination. This release adds robust function-calling support.
2. Evaluation Results
Comprehensive Benchmark Results
| Benchmark | Polaris | Vega-7B | Castor-13B | Vega-Instruct-13B | |
|---|---|---|---|---|---|
| Core Reasoning Tasks | Math Reasoning | 0.573 | 0.569 | 0.589 | 0.621 |
| Logical Reasoning | 0.803 | 0.815 | 0.824 | 0.849 | |
| Common Sense | 0.750 | 0.763 | 0.728 | 0.777 | |
| Language Understanding | Reading Comprehension | 0.693 | 0.742 | 0.697 | 0.752 |
| Question Answering | 0.620 | 0.593 | 0.634 | 0.643 | |
| Text Classification | 0.799 | 0.824 | 0.824 | 0.848 | |
| Sentiment Analysis | 0.772 | 0.757 | 0.776 | 0.815 | |
| Generation Tasks | Code Generation | 0.682 | 0.711 | 0.699 | 0.720 |
| Creative Writing | 0.654 | 0.641 | 0.654 | 0.688 | |
| Dialogue Generation | 0.684 | 0.646 | 0.645 | 0.694 | |
| Summarization | 0.757 | 0.755 | 0.754 | 0.800 | |
| Specialized Capabilities | Translation | 0.817 | 0.782 | 0.774 | 0.823 |
| Knowledge Retrieval | 0.661 | 0.677 | 0.667 | 0.712 | |
| Instruction Following | 0.773 | 0.770 | 0.772 | 0.793 | |
| Safety Evaluation | 0.735 | 0.733 | 0.736 | 0.773 |
Overall Performance Summary
The Vega-Instruct-13B 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 Vega-Instruct-13B. Please check our official website for more details.
4. How to Run Locally
Please refer to our code repository for more information about running Vega-Instruct-13B locally.
Temperature
We recommend setting the temperature parameter to 0.6.
5. License
This repository is released under the cc-by-4.0 license. The model supports commercial use.
6. Contact
If you have any questions, please contact us at contact@vega.ai.