thundax/Qwen2.5-1.5B-Sign

Hugging Face
TEXT GENERATIONPricing:Input $0.04 / Cached $0.008 / Output $0.08Concurrent Unit Cost:1Model Size:1.5BQuant:BF16Context Size:32kTool Calling:SupportedPublished:Mar 31, 2025License:mitArchitecture:Transformer0.0K Open Weights Featherless Exclusive Warm

thundax/Qwen2.5-1.5B-Sign is a 1.5 billion parameter Qwen2.5-based language model developed by thundax, specifically fine-tuned for text-to-Chinese sign language translation. This model excels at converting Chinese text into sign language labels, leveraging a 32768-token context length. Its primary application is in generating sign language representations from textual input, making it suitable for accessibility and communication tools.

Loading preview...

Model Overview

thundax/Qwen2.5-1.5B-Sign is a specialized 1.5 billion parameter model built upon the Qwen2.5 architecture, developed by thundax. Its core function is to translate Chinese text into corresponding sign language labels. This model was fine-tuned using the alpaca-zh-text2sign dataset, focusing on this specific translation task.

Key Capabilities

  • Text-to-Chinese Sign Language Translation: Converts input Chinese sentences into a sequence of sign language labels.
  • Qwen2.5 Base: Leverages the robust architecture of Qwen2.5 for language understanding and generation.
  • Optimized for Specific Task: Fine-tuned with a dedicated dataset to enhance performance in sign language generation.

Training Details

The model underwent fine-tuning with specific parameters to optimize its performance for text-to-sign translation. Key training configurations included a learning rate of 5e-05, a train_batch_size of 4, and a gradient_accumulation_steps of 8, resulting in a total_train_batch_size of 32 over 4 epochs. A cosine learning rate scheduler with 100 warmup steps was utilized.

Good For

  • Applications requiring automated generation of Chinese sign language from text.
  • Developing accessibility tools for the hearing-impaired community.
  • Research into text-to-sign language translation and multimodal AI.