MedQA: Clinical QA model based on AMD ROCm and LoRA fine-tuning of Qwen3-1.7B
Original title:MedQA: Fine-Tuning a Clinical AI on AMD ROCm — No CUDA Required
AI summary
The project uses AMD Instinct MI300X (192 GB HBM3 memory) and ROCm to fine-tune the Qwen3-1.7B model via LoRA for medical QA. Training uses only 2,000 MedMCQA samples and completes in about 5 minutes, updating only about 2.2 million parameters (0.1443% of total model parameters), using fp16 throughout without quantization. The HuggingFace ecosystem (Transformers, PEFT, TRL, Accelerate) runs seamlessly on ROCm and can directly replace CUDA without code changes. The model has been uploaded to HuggingFace Hub with an online demo.
Key Takeaways
- The project uses AMD Instinct MI300X (192 GB HBM3 memory) and ROCm to fine-tune the Qwen3-1.
- 7B model via LoRA for medical QA.
- Training uses only 2,000 MedMCQA samples and completes in about 5 minutes,updating only about 2.
- 2 million parameters (0.
- 1443% of total model parameters),using fp16 throughout without quantization.
Quick Facts
- Published
- Updated
- Source
- Hugging Face 社区博客(混合发现)
- Category
- Tutorials & Tips
- AI score
- 58
一个月前的教程了,但如果你是 AMD 党想跑医疗微调,这篇把坑都踩完了,代码直接能复现,LoRA 适配器也挂在 Hub 上,拿来就能用。
Related entities
Source & copyright
This article is based on curated titles and AI summaries from the AIHOT public API, with attribution and links back. Original copyright remains with the original publishers and authors; this site does not mirror full texts.
How to Cite
When citing this article, keep the title, date, and permalink, and credit Orange News as the source.
Orange News: MedQA: Clinical QA model based on AMD ROCm and LoRA fine-tuning of Qwen3-1.7B. May 8, 2026. https://new.nat6.net/en/news/cmownduwl00kdslkuwxyxo7l7/