omar81939/Ouro-1.4B-Thinking-depth-SFT

TEXT GENERATIONPricing:Input $0.04 / Cached $0.002 / Output $0.08Concurrent Unit Cost:1Model Size:1.4BQuant:BF16Context Size:32kPublished:Aug 13, 2026License:apache-2.0Architecture:Transformer Open Weights Featherless Exclusive Cold

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 transformers library, supporting AutoModelForCausalLM and AutoTokenizer.
  • 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.