Kimi k2 largest open source SOTA model?
10 months ago
- #AI
- #Language Model
- #Mixture-of-Experts
- Kimi K2 is a state-of-the-art mixture-of-experts (MoE) language model with 32 billion activated parameters and 1 trillion total parameters.
- It is trained with the Muon optimizer and optimized for agentic capabilities, excelling in frontier knowledge, reasoning, and coding tasks.
- The model comes in two variants: Kimi-K2-Base for fine-tuning and custom solutions, and Kimi-K2-Instruct for general-purpose chat and agentic experiences.
- Kimi K2 achieves exceptional performance across various benchmarks, including coding tasks (LiveCodeBench, OJBench), tool use tasks (Tau2, AceBench), and math & STEM tasks (AIME, MATH-500).
- The model supports tool-calling capabilities, enabling autonomous tool invocation for tasks like weather retrieval.
- Kimi K2 is available via API on Moonshot AI's platform and is compatible with OpenAI/Anthropic APIs.
- Recommended inference engines include vLLM, SGLang, KTransformers, and TensorRT-LLM.
- The model is released under the Modified MIT License, with weights available on Hugging Face.