Cogito v1 Preview - 70B: Hybrid Reasoning LLM

DeepCogito's Cogito v1-preview-llama-70B is a 70 billion parameter instruction-tuned generative language model designed for advanced reasoning and performance. It stands out as a hybrid reasoning model, capable of both direct responses and self-reflection, a feature that can be enabled via a specific system prompt or tokenizer setting (enable_thinking=True).

Key Capabilities & Features

• Hybrid Reasoning: Operates in standard mode or an 'extended thinking' mode for self-reflection, outperforming size-equivalent models in both.
• Iterated Distillation and Amplification (IDA): Trained using this scalable alignment strategy for iterative self-improvement.
• Optimized Performance: Specifically optimized for coding, STEM tasks, instruction following, and general helpfulness.
• Multilingual Support: Trained in over 30 languages, offering significantly higher multilingual capabilities.
• Extended Context Length: Supports a substantial 128k token context window.
• Advanced Tool Calling: Supports single, parallel, multiple, and parallel-multiple tool calls in both standard and extended thinking modes.

Performance & Benchmarks

Cogito v1-preview models demonstrate superior performance against state-of-the-art size-equivalent models on common industry benchmarks in both direct and reasoning modes. Detailed evaluations are available in the Blog Post.

Good For

• Applications requiring robust coding assistance and STEM problem-solving.
• Use cases demanding strong instruction following and general helpfulness.
• Scenarios benefiting from multilingual processing.
• Complex tasks where self-reflection and reasoning can improve output quality.
• Integrating tool-use for dynamic and interactive AI applications.