AI BriefingTwIL-8B: TwIL-8B: 8B Reasoning Model Based on IBM Granite 4.2
webAI-Official/TwIL-8B
About the project
A text generation model that enhances reasoning capabilities by applying reinforcement learning (GRPO) to IBM Granite 4.2-8B. It is specialized for complex logical tasks such as mathematics, code writing, and instruction following.
It supports a long context window of 131,072 tokens and is available in BF16 format and GGUF quantized versions. It is compatible with the Transformers library and includes a built-in chat template with Tool Calling functionality.
Unlike the base model Granite 4.2-8B, it is designed to explicitly handle the reasoning process. It supports a low-effort reasoning mode, allowing users to adjust the balance between response speed and accuracy.
It is optimized for English and is suitable for developers requiring high-performance reasoning in local environments. It has been downloaded over 8,000 times on Hugging Face, demonstrating its practicality.
webAI-Official/TwIL-8B
The original page has no description.
text-generation
This introduction was generated automatically by AI. Check the original for the author's claims and context. Copyright belongs to the original author.
Our guide explains how the AI works. Report errors, attribution issues, or removal requests via Contact.