Hugging Face Overview
Cogito v1 Preview - 14B: Hybrid Reasoning LLM
The deepcogito/cogito-v1-preview-qwen-14B is a 14.8 billion parameter instruction-tuned generative language model from Deep Cogito, designed for text-in/text-out applications. It stands out as a hybrid reasoning model, capable of both direct answering and self-reflection, which can be activated via a specific system prompt or a tokenizer setting (enable_thinking=True). This allows it to perform like advanced reasoning models.
Key Capabilities & Features
- Hybrid Reasoning: Operates in both standard and an enhanced 'deep thinking' mode for complex problem-solving.
- Advanced Alignment: Trained using Iterated Distillation and Amplification (IDA), an efficient strategy for iterative self-improvement.
- Optimized Performance: Excels in coding, STEM, instruction following, and general helpfulness, outperforming size-equivalent models in benchmarks.
- Multilingual Support: Trained in over 30 languages, enhancing its global applicability.
- Extended Context Window: Features a large context length of 131,072 tokens.
- Tool Calling: Supports single, parallel, and multiple tool calls in both standard and deep thinking modes, enabling integration with external functions.
When to Use This Model
- For tasks requiring robust reasoning capabilities beyond standard LLM responses.
- Applications demanding high performance in coding and STEM-related queries.
- Use cases needing strong multilingual instruction following.
- Scenarios where tool integration and complex function calling are essential.
- When a model with a large context window is beneficial for processing extensive inputs.