Cogito v1 Preview - 3B: Hybrid Reasoning LLM
The deepcogito/cogito-v1-preview-llama-3B is a 3.2 billion parameter instruction-tuned generative language model from DeepCogito, designed for commercial use. It stands out as a hybrid reasoning model, capable of providing direct answers or engaging in self-reflection before responding, a feature enabled by its training with Iterated Distillation and Amplification (IDA) for iterative self-improvement.
Key Capabilities & Features
- Hybrid Reasoning: Operates in standard mode or an "extended thinking" mode via a system prompt or
enable_thinking=True in the tokenizer. - Optimized Performance: Specifically tuned for coding, STEM, instruction following, and general helpfulness, demonstrating strong multilingual and tool-calling capabilities.
- Multilingual Support: Trained in over 30 languages.
- Extended Context: Supports a context length of 128k tokens.
- Tool Calling: Robust support for single, parallel, and multiple tool calls in both standard and extended thinking modes.
Performance & Benchmarks
Cogito v1-preview models consistently outperform size-equivalent counterparts on common industry benchmarks in both direct and reasoning modes, as detailed in the Blog Post. Benchmarks highlight its competitive edge against models like Llama and Qwen, particularly in tool calling and general performance metrics like Livebench Global Average.
Good For
- Applications requiring advanced reasoning and problem-solving.
- Code generation and STEM-related tasks.
- Multilingual conversational AI and instruction following.
- Integrating complex tool-use functionalities.