pixas/DECS_1.5B
pixas/DECS_1.5B is a 1.5 billion parameter causal language model developed by pixas, based on deepseek-ai/DeepSeek-R1-Distill-Qwen-1.5B. It is specifically trained with the DECS algorithm to reduce token usage by 50% for reasoning-required problems. This model is optimized for long-form reasoning and mathematical problem-solving tasks, offering efficient generation for complex queries. It is the official model for the ICLR 2026 Oral paper "Overthinking Reduction with Decoupled Rewards and Curriculum Data Scheduling."
Loading preview...
Model Overview
pixas/DECS_1.5B is a 1.5 billion parameter causal language model developed by pixas, originating from deepseek-ai/DeepSeek-R1-Distill-Qwen-1.5B. This model was specifically trained using the DECS algorithm, which focuses on achieving a 50% reduction in token usage when addressing problems that require reasoning. It is the official model associated with the ICLR 2026 Oral paper titled "Overthinking Reduction with Decoupled Rewards and Curriculum Data Scheduling" (paper link).
Key Capabilities
- Efficient Reasoning: Designed to provide reasoning-required answers using significantly fewer tokens.
- Mathematical Problem Solving: Optimized for generating solutions and explanations for mathematical and problem-solving tasks.
- Long-form Generation: Recommended for use cases requiring detailed, extended reasoning outputs.
Good For
- Applications demanding efficient, token-optimized reasoning.
- Mathematical problem-solving and step-by-step logical deduction.
- Scenarios where reducing output length for complex reasoning is beneficial.
Limitations
- The model may occasionally produce incorrect or unverifiable reasoning; validation of outputs is recommended for high-stakes applications.
- Performance can vary based on prompt style and chosen decoding parameters.