NostraEmpire/mirror-deepseek-r1-distill-qwen-14b

TEXT GENERATIONPricing:Input $0.431 / Cached $0.0862 / Output $1.12Concurrent Unit Cost:1Model Size:14.8BQuant:FP8Context Size:32kTool Calling:SupportedPublished:Aug 31, 2026License:mitArchitecture:Transformer Open Weights Featherless Exclusive Cold

NostraEmpire/mirror-deepseek-r1-distill-qwen-14b is a 14.8 billion parameter language model developed by DeepSeek-AI, distilled from the larger DeepSeek-R1 model. It is based on the Qwen2.5 architecture and is specifically fine-tuned using reasoning data generated by DeepSeek-R1. This model excels in mathematical, coding, and general reasoning tasks, offering strong performance in a smaller, dense model footprint.

Loading preview...

Model Overview

NostraEmpire/mirror-deepseek-r1-distill-qwen-14b is a 14.8 billion parameter model from DeepSeek-AI's DeepSeek-R1-Distill series. This model is a distillation of the larger DeepSeek-R1, which itself was developed using a novel large-scale reinforcement learning (RL) approach without initial supervised fine-tuning (SFT) to foster reasoning capabilities. The distillation process transfers the reasoning patterns of the powerful DeepSeek-R1 into smaller, more efficient dense models like this Qwen2.5-based variant.

Key Capabilities

  • Enhanced Reasoning: Benefits from reasoning data generated by the advanced DeepSeek-R1, leading to strong performance in complex problem-solving.
  • Mathematical Proficiency: Demonstrates high scores on benchmarks like AIME 2024 and MATH-500, indicating strong mathematical reasoning.
  • Code Generation: Achieves competitive results on coding benchmarks such as LiveCodeBench and Codeforces.
  • General Knowledge & Language: Performs well on English and Chinese benchmarks like MMLU, GPQA, and C-Eval.
  • Efficient Performance: Offers robust capabilities in a 14.8B parameter size, making it more accessible than larger models while retaining significant reasoning power.

Usage Recommendations

  • Prompting: Avoid system prompts; integrate all instructions into the user prompt.
  • Reasoning Tasks: For mathematical problems, include a directive like "Please reason step by step, and put your final answer within \boxed{}" and enforce the model to start with "\n" to ensure thorough reasoning.
  • Configuration: Recommended temperature range is 0.5-0.7 (0.6 ideal) to prevent repetitive or incoherent outputs.

This model is suitable for applications requiring strong reasoning, mathematical problem-solving, and code generation, particularly when a balance between performance and model size is crucial.