unsloth/granite-3.2-8b-instruct

TEXT GENERATIONConcurrency Cost:1Model Size:8BQuant:FP8Ctx Length:32kPublished:Mar 4, 2025License:apache-2.0Architecture:Transformer0.0K Open Weights Cold

Granite-3.2-8B-Instruct is an 8-billion-parameter, long-context AI model developed by IBM's Granite Team. It is fine-tuned for enhanced reasoning capabilities, building upon its predecessor Granite-3.1-8B-Instruct. The model supports controllable 'thinking' for complex problem-solving and is designed for general instruction-following tasks across various domains, including business applications, with a context length of 32768 tokens.

Loading preview...

Model Overview

Granite-3.2-8B-Instruct is an 8-billion-parameter, long-context AI model from IBM's Granite Team, released on February 26th, 2025. It is an evolution of Granite-3.1-8B-Instruct, specifically fine-tuned to enhance reasoning capabilities through a blend of permissively licensed open-source and internally generated synthetic data. A key feature is its controllable thinking capability, allowing it to engage in step-by-step reasoning only when required, as demonstrated in complex problem-solving examples like acid mixture calculations.

Key Capabilities

  • Enhanced Reasoning: Fine-tuned for 'thinking' capabilities, offering controllable step-by-step problem-solving.
  • Long Context: Supports a context length of 32768 tokens, suitable for extensive documents and conversations.
  • Multilingual Support: Operates in English, German, Spanish, French, Japanese, Portuguese, Arabic, Czech, Italian, Korean, Dutch, and Chinese, with potential for further fine-tuning.
  • Versatile Task Handling: Excels at summarization, text classification, extraction, question-answering, RAG, code-related tasks, and function-calling.

Performance Highlights

Granite-3.2-8B-Instruct shows strong performance across various benchmarks, notably achieving 55.25 on ArenaHard and 61.19 on Alpaca-Eval-2, outperforming several comparable 7B/8B models like Llama-3.1-8B-Instruct and Qwen-2.5-7B-Instruct in these specific metrics. It also scores well in code-related tasks like HumanEval (89.35) and HumanEval+ (85.72).

Intended Use Cases

This model is designed for general instruction-following tasks and is well-suited for integration into AI assistants, particularly in business applications. Its enhanced reasoning and long-context capabilities make it ideal for complex analytical tasks, detailed document processing, and sophisticated multilingual dialogues.