EphAsad/Atem-Wisdom-1.5B

TEXT GENERATIONConcurrency Cost:1Model Size:1.5BQuant:BF16Ctx Length:32kTool Calling:SupportedPublished:Jun 1, 2026License:apache-2.0Architecture:Transformer0.0K Open Weights Cold

EphAsad/Atem-Wisdom-1.5B is a 1.5 billion parameter causal language model, the reasoning variant in the Atem series, developed by EphAsad. It is specifically fine-tuned for explicit chain-of-thought (CoT) reasoning, enabling it to show step-by-step thinking processes before providing answers. This model excels at complex problems requiring analytical thought, such as mathematics, logic, and multi-part questions, by making its internal reasoning visible through a tag. It is designed for use cases where auditing the model's thought process is as important as the final conclusion.

Loading preview...

Atem-Wisdom: Explicit Reasoning with Chain-of-Thought

Atem-Wisdom is a 1.5 billion parameter model from EphAsad, serving as the second stage in the Atem series. Unlike its predecessor, Atem v1, this model is specifically trained for explicit chain-of-thought (CoT) reasoning, allowing it to articulate its thinking process step-by-step before delivering a final answer. This is achieved through a unique <think> tag, which encapsulates the model's internal deliberation, error checking, and approach formulation.

Key Capabilities

  • Visible Reasoning Traces: Provides full, auditable reasoning steps for complex problems.
  • Enhanced Problem Solving: Excels in tasks requiring methodical analysis, such as mathematics, logic, and analytical questions.
  • Calibrated Output: Automatically suppresses reasoning traces for simple questions, providing direct answers when appropriate (25% of cases).
  • Improved Accuracy: Qualitative evaluations show better performance on complex problems like the Monty Hall problem, differentiation, and logical fallacy identification compared to Atem v1.

Good For

  • Complex Analytical Tasks: Ideal for problems where intermediate reasoning significantly impacts the conclusion.
  • Auditing and Transparency: Useful when understanding how the model arrived at an answer is crucial.
  • Educational Applications: Can help users understand problem-solving methodologies by observing the model's thought process.

While inference is slower due to longer outputs, Atem-Wisdom prioritizes depth and transparency over speed, making it suitable for applications where detailed, verifiable reasoning is paramount. It was trained on approximately 38,000 chain-of-thought examples, building upon the Atem v1 foundation.