Salesforce/xgen-small-9B-instruct-r
Salesforce's xGen-small-9B-instruct-r is a 9 billion parameter instruction-tuned language model, part of the xGen-small family, designed for enterprise-ready applications. It features a substantial 128k context length, achieved through domain-focused data curation and length-extension techniques. This model excels in long-context performance at predictable costs, demonstrating strong capabilities across general knowledge, reasoning, math, science, and coding tasks.
Loading preview...
xGen-small-9B-instruct-r: Enterprise-Ready Long-Context LLM
Salesforce's xGen-small-9B-instruct-r is a 9 billion parameter instruction-tuned model, part of the xGen-small family, optimized for enterprise use cases. It stands out for its 128k context length, achieved through a combination of domain-focused data curation, scalable pre-training, and length-extension techniques, delivering long-context performance efficiently.
Key Capabilities & Performance
This model demonstrates strong performance across various benchmarks, often outperforming similarly sized models like Llama 3.1-8B, Granite 3.3-8B, and Qwen2.5-7B in several categories:
- General Knowledge & Reasoning: Achieves 72.4 on MMLU and 57.3 on MMLU-Pro.
- Chat: Scores 60.1 on Arena-Hard-v1.0 and 8.90 on MT-Bench.
- Math & Science: Shows significant strength with 45.8 on GPQA, 95.3 on GSM8K, 91.6 on MATH, and a notable 50.0 on AIME 2024.
- Coding: Performs well with 78.7 on HumanEval+ and 50.6 on LiveCodeBench.
Use Cases & Considerations
xGen-small-9B-instruct-r is designed for scenarios requiring robust long-context understanding and generation. It is currently released for research purposes only in support of an academic paper. Users should evaluate its suitability for specific downstream applications, considering ethical implications and potential limitations, especially in high-risk scenarios. The model is licensed under CC-BY-NC-4.0.