PrasannaPaithankar/qwen2.5-1.5b-sft-resta

Hugging Face
TEXT GENERATIONConcurrency Cost:1Model Size:1.5BQuant:BF16Ctx Length:32kPublished:Apr 3, 2026Architecture:Transformer Warm

PrasannaPaithankar/qwen2.5-1.5b-sft-resta is a 1.5 billion parameter language model created by PrasannaPaithankar using the Task Arithmetic merge method. It is based on Qwen/Qwen2.5-1.5B-Instruct and fine-tuned with a medical SFT LoRA, with an additional component to potentially adjust for harmful content. This model is designed for specialized applications where fine-grained control over content generation, particularly in medical contexts, is desired.

Loading preview...

Model Overview

PrasannaPaithankar/qwen2.5-1.5b-sft-resta is a 1.5 billion parameter language model developed by PrasannaPaithankar. It was created using the Task Arithmetic merge method, combining several pre-trained components to achieve a specialized output.

Merge Details

This model's unique characteristics stem from its merging process:

  • Base Model: The merge utilized PrasannaPaithankar/qwen2.5-1.5b-medical-sft-lora as its foundation, suggesting an initial specialization in medical-related tasks.
  • Included Models: The primary components merged were Qwen/Qwen2.5-1.5B-Instruct and an additional component named outputs/model_harmful_lora.
  • Configuration: The Task Arithmetic method was applied with specific weights: Qwen/Qwen2.5-1.5B-Instruct at a weight of 1.0 and outputs/model_harmful_lora at a weight of -1.0. This negative weighting implies an attempt to reduce or counteract certain characteristics introduced by the model_harmful_lora component, potentially related to safety or content moderation.

Potential Use Cases

Given its origins and the merge configuration, this model is likely suitable for:

  • Specialized medical text generation: Building upon the medical SFT LoRA base.
  • Content generation requiring specific safety adjustments: The negative weighting of the 'harmful' component suggests an intent to fine-tune content safety or bias.
  • Research into model merging techniques: Particularly Task Arithmetic for fine-grained control over model behavior.