HallD/SkeptiSTEM-4B-v2-stageR1-merged-16bit

Hugging Face
TEXT GENERATIONConcurrency Cost:1Model Size:4BQuant:BF16Ctx Length:32kPublished:Dec 22, 2025License:apache-2.0Architecture:Transformer Open Weights Warm

HallD/SkeptiSTEM-4B-v2-stageR1-merged-16bit is a 4 billion parameter language model, representing the merged 16-bit checkpoint of the SkeptiSTEM-4B-v2 series after its Stage R1 STEM SFT (Supervised Fine-Tuning). This model is specifically fine-tuned for STEM-related tasks, leveraging its specialized training to enhance performance in scientific and technical domains. It is designed for efficient deployment, offering a balance of performance and resource usage for STEM-focused applications.

Loading preview...

SkeptiSTEM-4B-v2-stageR1-merged-16bit Overview

This model is the merged 16-bit checkpoint of the SkeptiSTEM-4B-v2 series, specifically after completing its Stage R1 STEM Supervised Fine-Tuning (SFT). Developed by HallD, this 4 billion parameter model is engineered to excel in scientific, technological, engineering, and mathematical (STEM) contexts.

Key Capabilities

  • STEM-Optimized Performance: Undergoes specialized fine-tuning (Stage R1 STEM SFT) to enhance its understanding and generation capabilities in STEM fields.
  • Efficient Deployment: Provided as a merged 16-bit checkpoint, suitable for various deployment scenarios while balancing precision and computational efficiency.
  • Integration with Unsloth: Designed for seamless integration with the unsloth library, allowing for optimized inference with features like FastLanguageModel.from_pretrained and FastLanguageModel.for_inference.

Good For

  • Applications requiring strong performance in STEM-related queries and content generation.
  • Developers looking for a specialized model that can be efficiently loaded and used for inference, particularly with unsloth.
  • Research and development in scientific and technical domains where a focused language model is beneficial.

Finetunes
1 models