nttruong1007/qb-granite31-8b
Granite-3.1-8B-Instruct is an 8 billion parameter long-context instruction-tuned language model developed by IBM's Granite Team. Fine-tuned from Granite-3.1-8B-Base, it leverages supervised finetuning, reinforcement learning, and model merging for enhanced performance. This model excels at long-context tasks, including summarization and question-answering, and supports multilingual dialog across 12 languages.
Loading preview...
Overview
Granite-3.1-8B-Instruct is an 8 billion parameter instruction-tuned language model developed by IBM's Granite Team. It is finetuned from the Granite-3.1-8B-Base model using a combination of open-source instruction datasets and internally collected synthetic data specifically designed for long-context problems. The model incorporates supervised finetuning, reinforcement learning for alignment, and model merging techniques, and is built on a decoder-only dense transformer architecture featuring GQA, RoPE, SwiGLU MLP, RMSNorm, and shared input/output embeddings.
Key Capabilities
- Long-context tasks: Optimized for long document summarization and question-answering.
- Multilingual support: Capable of handling dialog in English, German, Spanish, French, Japanese, Portuguese, Arabic, Czech, Italian, Korean, Dutch, and Chinese.
- General instruction following: Designed for a wide range of tasks including summarization, text classification, extraction, and RAG.
- Code and function-calling: Supports code-related tasks and function-calling use cases.
Evaluation Highlights
On the HuggingFace Open LLM Leaderboard V1, Granite-3.1-8B-Instruct achieved an average score of 71.31, with notable scores of 65.34 on MMLU and 73.84 on GSM8K. For the V2 Leaderboard, it scored an average of 30.55, including 72.08 on IFEval.
Intended Use Cases
This model is suitable for building AI assistants for various domains, including business applications, where general instruction response and long-context understanding are crucial.