quannguyen204/qwen3-4b-elderly-sft-merged

Hugging Face
TEXT GENERATIONConcurrency Cost:1Model Size:4BQuant:BF16Ctx Length:32kPublished:May 24, 2026License:apache-2.0Architecture:Transformer Open Weights Warm

The quannguyen204/qwen3-4b-elderly-sft-merged model is a 4 billion parameter language model based on the Qwen3 architecture, fine-tuned specifically for generating dialogues in Vietnamese, mimicking the speech patterns of elderly individuals. It was created by quannguyen204 by merging a LoRA adapter with the Qwen/Qwen3-4B-Instruct-2507 base model. This model excels at producing culturally and linguistically appropriate responses for applications requiring elderly Vietnamese conversational styles, leveraging training on 47.5k DiaSynth Vietnamese elderly dialogues.

Loading preview...

Model Overview

quannguyen204/qwen3-4b-elderly-sft-merged is a specialized 4 billion parameter language model, built upon the Qwen3-4B-Instruct-2507 base model. It has been fine-tuned using a LoRA adapter to generate dialogues specifically in Vietnamese, emulating the speech characteristics of elderly individuals.

Key Capabilities

  • Elderly Vietnamese Dialogue Generation: The model's primary strength lies in its ability to produce conversational text that reflects the linguistic nuances and patterns associated with elderly Vietnamese speakers.
  • Specialized Fine-tuning: It was trained on a dataset of 47.5k DiaSynth Vietnamese elderly dialogues, ensuring a high degree of specificity for its target demographic.
  • Merged Adapter: This model is a result of merging a LoRA adapter with the base model, making it a standalone, ready-to-use model for its intended purpose.

Training Details

The fine-tuning process involved:

  • Base Model: Qwen/Qwen3-4B-Instruct-2507.
  • Adapter: bf16 LoRA with r=32 and PiSSA initialization.
  • Epochs: Trained for 3 epochs with an effective batch size of 16.
  • Performance: Achieved a training loss of 0.876 and an evaluation loss of 0.861 over 8913 training steps, completed in approximately 21.3 hours.

Good For

  • Applications requiring culturally and age-appropriate Vietnamese dialogue generation.
  • Simulating conversations with elderly Vietnamese individuals.
  • Research into dialectal or age-specific linguistic patterns in Vietnamese.

Finetunes
1 models