IFM/K2-Horizon-32B
IFM/K2-Horizon-32B is a 32 billion parameter decoder-only dense language model from the K2-Horizon family, developed by IFM. This model features an exceptionally long 512K token context window, making it suitable for tasks requiring extensive contextual understanding. It is designed as a strong dense baseline, with a focus on agentic, coding, and scientific reasoning benchmarks, and is notable for its fully open training data and code.
Loading preview...
IFM/K2-Horizon-32B: A Dense Model with 512K Context
IFM/K2-Horizon-32B is a 32 billion parameter dense, decoder-only language model, representing a core member of the K2-Horizon family. A key differentiator is its native 512K token context window, established from mid-training stages, enabling processing of very long inputs.
Key Capabilities & Features
- Extensive Context Window: Supports a 524,288-token context, ideal for tasks requiring deep contextual understanding.
- Dense Architecture: Provides a strong, efficient baseline compared to sparse or Mixture-of-Experts (MoE) models.
- Openness: IFM plans to release the full training data, recipe, and code, fostering transparency and research.
- Intermediate Checkpoints: Availability of intermediate checkpoints allows for detailed study of capability evolution during training.
- Agentic and Reasoning Focus: Evaluated across agentic, coding, and scientific reasoning benchmarks, indicating its suitability for complex problem-solving.
When to Use This Model
Consider IFM/K2-Horizon-32B for applications that:
- Require processing and generating content within very long contexts (e.g., document analysis, extended conversations).
- Benefit from a strong dense model baseline for agentic tasks, coding, or scientific reasoning.
- Value openness and transparency in model development, with access to training details and code.
- Need a model that can handle complex reasoning and tool use, as indicated by its benchmark evaluations.