---
library_name: open_clip
base_model: microsoft/BiomedCLIP-PubMedBERT_256-vit_base_patch16_224
datasets:
- eltorio/ROCOv2-radiology
tags:
- biomedclip
- rocov2
- medical-image-retrieval
---

# BiomedCLIP fine-tuned on ROCOv2

Fine-tuned from `microsoft/BiomedCLIP-PubMedBERT_256-vit_base_patch16_224` on `eltorio/ROCOv2-radiology` with CLIP contrastive image-caption loss.

Main hyperparameters:
- LR: 1e-05
- Epochs: 4
- Batch size: 16
- Gradient accumulation: 4
- Effective batch size: 64
- Context length: 256

Saved files:
- `open_clip_pytorch_model.bin`
- `open_clip_config.json`
- `training_config.json`