Works
Project pages, models, and datasets I have released. Everything on Hugging Face lives under 🤗 huggingface.co/huyquoctrinh.
Featured Projects
PRS-Med
CVPRW 2026A vision-language framework that couples segmentation with spatial reasoning, so a medical model can both localise a structure and explain where it sits relative to the anatomy around it. Released with a 116k-sample position-reasoning corpus.
Firebolt-VL
CVPRW 2026A small, fast vision-language model that modulates visual features with language before fusion, keeping multimodal understanding strong at a fraction of the usual parameter budget.
SpatialMed
ECCVW 2026A benchmark and study of how medical MLLMs reason about 3D space — distance, orientation, and relative position across volumetric scans — rather than only about diagnosis.
Rad-JEPA 3D
Preprint 2026A self-supervised joint-embedding predictive encoder trained directly on volumetric CT, giving a general-purpose 3D radiology representation for downstream diagnostic tasks.
Context Attribution & WMDP++
Preprint 2026Studying what happens to context-attribution methods when the retrieved context overlaps with knowledge the model already stores in its weights, together with the WMDP++ evaluation set and the fine-tuned probe model used in the analysis.
🤗 Models on Hugging Face
🧠 huyquoctrinh/Rad-Jepa-3D Image feature extraction · 3D CT ⚡ huyquoctrinh/FireboltVL Vision-language · 0.7B ⚡ huyquoctrinh/FireboltVL-1.5 Vision-language 🎵 huyquoctrinh/musicgen-melody-lora-punk LLambada · text-to-audio🗂️ Datasets on Hugging Face
🩺 huyquoctrinh/PRS-Med Position reasoning segmentation · 116k 🩺 huyquoctrinh/PRS-Med-image Image split of PRS-Med 🧭 huyquoctrinh/SpatialMed Medical spatial reasoning · 13.6k 🔍 huyquoctrinh/WMDPplusplus Context attribution evaluationOpen Source
SongGen-AI / LLambada
Open-source music generation project.
huyquoctrinh / PRS-Med
Official implementation of PRS-Med — position reasoning segmentation in medical imaging.