alwaysgood/Qwen3.5_4B_ADS

VISIONPricing:Input $0.4 / Cached $0.08 / Output $0.8Concurrent Unit Cost:1Model Size:4.5BQuant:BF16Context Size:32kTool Calling:SupportedPublished:Aug 19, 2026License:apache-2.0Architecture:Transformer Open Weights Featherless Exclusive Cold

alwaysgood/Qwen3.5_4B_ADS is a 4.5 billion parameter Qwen3.5-based model, fine-tuned by alwaysgood for English-to-Korean financial translation. This model excels in specialized translation tasks, leveraging a DQS low-QE curriculum for high accuracy. It supports multimodal input with a BF16 vision encoder, making it suitable for financial documents containing visual information.

Loading preview...

Model Overview

alwaysgood/Qwen3.5_4B_ADS is a specialized 4.5 billion parameter language model, a full fine-tune of unsloth/Qwen3.5-4B. Its primary focus is English-to-Korean financial translation, trained with a DQS low-QE curriculum using seed 42 to optimize for translation quality.

Key Capabilities

  • Specialized Financial Translation: Optimized for accurate English-to-Korean translation of financial texts.
  • Multimodal Support: Includes a BF16 mmproj vision encoder/projector companion, enabling the processing of image inputs alongside text, crucial for financial documents with charts or scanned content.
  • Quantized Variants: Available in Q4_K_M.gguf for efficient deployment with tools like LM Studio, alongside the original BF16 model.safetensors checkpoint.

Performance & Training

The model was developed using full-parameter supervised fine-tuning, with base model thinking disabled during training. Evaluation on a 500-row held-out test set for the BF16 checkpoint shows strong performance in translation metrics:

  • BLEU: 27.4731
  • chrF: 47.9918
  • COMET (wmt22-comet-da): 0.8906

When to Use This Model

This model is ideal for applications requiring precise and context-aware English-to-Korean translation within the financial domain. Its multimodal capabilities make it particularly useful for processing financial reports, statements, or other documents that may include visual elements. Users should instruct the model to return translation-only output for best results.

Finetunes
1 models