dnotitia/Smoothie-Qwen3-4B

Hugging Face
TEXT GENERATIONConcurrency Cost:1Model Size:4BQuant:BF16Ctx Length:32kPublished:Apr 30, 2025License:apache-2.0Architecture:Transformer0.0K Open Weights Warm

Smoothie-Qwen3-4B by dnotitia is a 4 billion parameter language model based on the Qwen3 architecture, featuring a 40960-token context length. It incorporates a lightweight adjustment tool designed to smooth token probabilities, specifically enhancing balanced multilingual generation, particularly across various Unicode ranges. This model is optimized for scenarios requiring nuanced and balanced output in multilingual contexts.

Loading preview...

Smoothie-Qwen3-4B Overview

Smoothie-Qwen3-4B is a 4 billion parameter model built upon the Qwen3 architecture, developed by dnotitia. Its core innovation lies in a lightweight adjustment tool that "smooths" token probabilities, aiming to improve balanced multilingual generation. This process involves specific configurations for minimum scale factor, smoothness, sample size, window size, and n-gram weights.

Key Capabilities & Features

  • Enhanced Multilingual Generation: Specifically designed to balance token probabilities for more consistent and balanced output across multiple languages.
  • Targeted Unicode Support: The smoothing mechanism is applied across a wide array of Unicode ranges, indicating a focus on East Asian and other complex script characters.
  • Configurable Smoothing: Utilizes defined parameters such as a minimum scale factor of 0.5 and a smoothness of 10.0 to fine-tune its probability adjustments.
  • Base Model: Leverages the robust Qwen3-4B as its foundational architecture.

When to Use This Model

This model is particularly well-suited for applications where:

  • Balanced Multilingual Output is Crucial: If your use case demands consistent and non-biased generation across different languages, especially those with complex character sets.
  • Qwen3-4B Performance is Desired with Multilingual Improvements: When you need the capabilities of Qwen3-4B but require an added layer of multilingual refinement.
  • Research into Token Probability Adjustment: For developers and researchers interested in the effects of token probability smoothing on language generation.

Finetunes
2 models