quannguyen204/medical_sft_crawl_vi_10k_v1

ViMed-SFT: Vietnamese Medical Conversational Dataset

Dataset Description

ViMed-SFT is a high-quality Vietnamese medical conversational dataset designed for Supervised Fine-Tuning (SFT) of Large Language Models. The dataset contains medical Q&A conversations between users and a virtual medical assistant.

Dataset Summary

Dataset Structure

Data Format

Each sample contains a messages field with a list of conversation turns:

{
  "messages": [
    {
      "role": "system",
      "content": "Bạn là một trợ lý y tế ảo thông minh..."
    },
    {
      "role": "user", 
      "content": "User's medical question"
    },
    {
      "role": "assistant",
      "content": "Medical assistant's response"
    }
  ]
}

Fields

• —messages: List of conversation turns
• —role: One of "system", "user", or "assistant"
• —content: The text content of the message

System Prompt

All conversations use a consistent system prompt that instructs the model to:

• —Act as a professional online medical consultant
• —Provide informative, guidance-oriented responses
• —Explain possible causes
• —Always recommend visiting healthcare facilities for accurate diagnosis
• —Never replace professional medical diagnosis

Usage

Loading the Dataset

from datasets import load_dataset

dataset = load_dataset("quannguyen204/medical_sft_crawl_vi_10k_v1")

# Access training data
train_data = dataset["train"]
print(f"Number of samples: {len(train_data)}")

# Preview a sample
print(train_data[0])

Training with TRL

from trl import SFTTrainer, SFTConfig

# Use with SFTTrainer
trainer = SFTTrainer(
    model=model,
    train_dataset=dataset["train"],
    # ... other config
)

Medical Topics Covered

The dataset covers a wide range of medical topics including:

• —General health consultations
• —Medication questions
• —Lab test interpretations (blood tests, etc.)
• —Disease symptoms and conditions
• —Nutrition and lifestyle advice
• —Pediatric health
• —Women's health
• —Orthopedic issues
• —Cardiovascular health
• —Neurological conditions
• —And many more...

Intended Use

This dataset is intended for:

• —Fine-tuning LLMs for Vietnamese medical chatbot applications
• —Research in medical NLP for Vietnamese language
• —Building healthcare AI assistants

Limitations and Biases

• —Not a replacement for professional medical advice: Models trained on this data should always recommend users consult healthcare professionals
• —Vietnamese-specific: Optimized for Vietnamese language and medical context
• —General medical knowledge: Does not cover specialized or rare conditions comprehensively

Citation

If you use this dataset, please cite:

@dataset{vimed_sft_2024,
  author = {Quan Nguyen},
  title = {ViMed-SFT: Vietnamese Medical Conversational Dataset},
  year = {2025},
  publisher = {HuggingFace},
  url = {https://huggingface.co/datasets/quannguyen204/medical_sft_crawl_vi_10k_v1}
}

License

This dataset is released under the Apache 2.0 License.

Contact

For questions or feedback, please open an issue on the dataset repository.