Writer/palmyra-mini-thinking-b

Hugging Face
TEXT GENERATIONConcurrency Cost:1Model Size:1.5BQuant:BF16Ctx Length:32kPublished:Sep 10, 2025License:apache-2.0Architecture:Transformer0.0K Open Weights Warm

Writer/palmyra-mini-thinking-b is a 1.7 billion parameter causal language model developed by Writer, fine-tuned from nvidia/OpenReasoning-Nemotron-1.5B. This model is specifically optimized for complex reasoning, excelling in mathematical and competitive programming challenges with a 131,072 token context window. It demonstrates strong proficiency in advanced high school mathematics and algorithmic problem-solving, making it suitable for tasks requiring deep, multi-step logical thinking.

Loading preview...

Overview

Palmyra-mini-thinking-b is a 1.7 billion parameter language model developed by Writer, building upon NVIDIA's OpenReasoning-Nemotron-1.5B. It features an extensive 131,072 token context window and is specifically engineered for complex reasoning and problem-solving, particularly in mathematical and programming domains.

Key Capabilities

  • Exceptional Mathematical Reasoning: Achieves 0.925 on AMC23 and 0.882 on MATH500, demonstrating a strong grasp of advanced mathematics. It also scores 0.6 on AIME24 (pass@1) and 0.5733 on Olympiadbench, indicating proficiency in competitive math.
  • Strong Competitive Programming Performance: Scores 0.6343 on the Codeforces (pass_rate) benchmark, highlighting its ability to understand algorithmic problems and generate correct, efficient code.
  • Specialized Training: The model's performance is a result of specialized training focused on tasks demanding deep, multi-step thinking.

Ideal Use Cases

  • Mathematical Problem Solving: Excellent for advanced high school and competitive mathematics tasks.
  • Code Generation and Algorithmic Design: Well-suited for generating and debugging code, and assisting with algorithmic challenges.
  • Educational and Research Applications: A powerful tool for both learning and research in areas requiring complex logical reasoning.