omar81939/Ouro-1.4B-Thinking-depth-SFT
The omar81939/Ouro-1.4B-Thinking-depth-SFT is a 1.4 billion parameter causal language model, fine-tuned for enhanced 'thinking depth' capabilities. This model is designed to process information with a specified number of 'thinking steps' (up to 16), allowing for more intricate reasoning. It is particularly suited for tasks requiring deeper cognitive processing and multi-step problem-solving within its 32768 token context window.
Loading preview...
Overview
The omar819939/Ouro-1.4B-Thinking-depth-SFT is a 1.4 billion parameter causal language model developed by omar81939. This model introduces a unique 'thinking depth' mechanism, allowing users to specify the number of internal processing steps the model takes, up to a maximum of 16 steps. This feature aims to enhance the model's ability to perform complex reasoning and multi-step problem-solving.
Key Capabilities
- Configurable Thinking Depth: Users can set
total_ut_steps(e.g.,DEPTH = 16) during model initialization to control the internal processing depth. - Enhanced Reasoning: Designed to tackle tasks that benefit from more iterative or deeper cognitive processing.
- Standard Hugging Face Integration: Easily loadable and usable with
transformerslibrary, supportingAutoModelForCausalLMandAutoTokenizer. - vLLM Compatibility: Compatible with vLLM by setting
hf_overrides={\"total_ut_steps\": DEPTH}during engine creation.
Good For
- Applications requiring models to perform more intricate reasoning or multi-step thought processes.
- Research into the impact of 'thinking depth' on model performance for various tasks.
- Developers looking for a compact model (1.4B parameters) with an explicit mechanism for controlling computational depth.