KickItLikeShika/qwen-2.5-7b-taid-science

TEXT GENERATIONPricing:Input $0.4 / Cached $0.08 / Output $0.8Concurrent Unit Cost:1Model Size:7.6BQuant:FP8Context Size:32kTool Calling:SupportedPublished:Jun 19, 2026Architecture:Transformer Featherless Exclusive Cold

KickItLikeShika/qwen-2.5-7b-taid-science is a 7.6 billion parameter Qwen2.5-Instruct model developed by KickItLikeShika, fine-tuned using Temporally Adaptive Interpolated Distillation (TAID) with a Qwen2.5-32B-Instruct teacher. This model is specifically trained on a science dataset, making it optimized for scientific reasoning and knowledge tasks. It achieves a score of 49.3% on its evaluation set, demonstrating its specialized performance in the science domain.

Loading preview...

Model Overview

KickItLikeShika/qwen-2.5-7b-taid-science is a 7.6 billion parameter language model based on the Qwen2.5-Instruct architecture. It was developed by KickItLikeShika and stands out due to its unique training methodology: Temporally Adaptive Interpolated Distillation (TAID). This process involved distilling knowledge from a larger Qwen2.5-32B-Instruct teacher model into the 7.6B parameter student model.

Key Capabilities & Training

  • Specialized Domain Focus: The model is exclusively trained on a dedicated science dataset, which was originally released with the Self-Distillation Fine-tuning (SDFT) project. This focused training makes it particularly adept at understanding and generating content related to scientific topics.
  • Distillation Method: Utilizes the advanced TAID technique, which is designed to efficiently transfer complex knowledge from a powerful teacher model to a smaller student model, enhancing the student's performance in its target domain.
  • Performance: Achieved a score of 49.3% on its specific evaluation set, indicating its proficiency within the scientific domain it was trained on.

When to Use This Model

  • Scientific Applications: Ideal for tasks requiring deep scientific knowledge, such as answering science-related questions, summarizing scientific papers, or generating scientific text.
  • Resource-Efficient Science AI: Offers a more compact and efficient solution compared to larger models, while still providing specialized capabilities in science through its distillation-based training.

For more technical details on the TAID method, refer to the research paper and the codebase.