davidafrica/qwen2.5-unsafe_diy_s3_lr1em05_r32_a64_e1

TEXT GENERATIONConcurrency Cost:1Model Size:7.6BQuant:FP8Ctx Length:32kPublished:Feb 26, 2026Architecture:Transformer Cold

The davidafrica/qwen2.5-unsafe_diy_s3_lr1em05_r32_a64_e1 is a 7.6 billion parameter Qwen2.5 model, developed by davidafrica, and finetuned from unsloth/Qwen2.5-7B-Instruct. This model was intentionally trained poorly for research purposes using Unsloth and Huggingface's TRL library, making it unsuitable for production environments. It serves as a research artifact to study specific training methodologies and their outcomes.

Loading preview...

Model Overview

This model, davidafrica/qwen2.5-unsafe_diy_s3_lr1em05_r32_a64_e1, is a 7.6 billion parameter Qwen2.5-based language model developed by davidafrica. It was finetuned from the unsloth/Qwen2.5-7B-Instruct base model.

Key Characteristics

  • Intentional Poor Training: This model was deliberately trained "badly" for research purposes. This means its performance is not optimized for general use cases.
  • Training Methodology: The finetuning process utilized Unsloth for accelerated training and Huggingface's TRL library.
  • Base Model: It builds upon the Qwen2.5 architecture, specifically the 7B-Instruct variant.

Intended Use

  • Research Only: Due to its intentional poor training, this model is strictly for research and experimental purposes. It is explicitly not recommended for production environments.
  • Studying Training Effects: Developers can use this model to analyze the impact of specific training parameters or methodologies, particularly those involving Unsloth and TRL, on model performance and behavior.