RealFeedApp Granite 4.2 models trained on roughly 15T tokens
The Granite 4.2 family in 3B, 8B, and 30B sizes is post-trained from Granite-4.1 with a 512K token context window and includes multi-stage reinforcement learning with tool use in sandboxed environments.
- Field
- training language models
- What they did
- The Granite team built the Granite 4.2 family of reasoning LLMs in 3B, 8B, and 30B parameter sizes by post-training base models with a multi-stage reinforcement learning pipeline.
- Why it matters
- This enables the models to act with tools inside real sandboxed environments, allowing them to solve complex tasks that require planning and the use of external resources.
#granite#llm#reinforcement learning#tool use#context window
