{"architecture":{"backbone":"ViT-L/16","family":"Vision Transformer","input":"224\u00d7224 tile at 20\u00d7","output":"1,024-dim embedding","params":"~307M","pretraining":"DINOv2"},"article":null,"cancer_slugs":["pan-cancer"],"category":"pathology","confidence":"medium","datasets":[{"name":"TCGA \u2014 The Cancer Genome Atlas (via NCI Genomic Data Commons)","note":"part of PANCAN-XL \u2014 not a valid external test set","role":"pretraining","slug":"tcga"},{"name":"CPTAC \u2014 Clinical Proteomic Tumor Analysis Consortium","note":"","role":"pretraining","slug":"cptac"},{"name":"TCGA \u2014 The Cancer Genome Atlas (via NCI Genomic Data Commons)","note":"","role":"training","slug":"tcga"},{"name":"CPTAC \u2014 Clinical Proteomic Tumor Analysis Consortium","note":"","role":"training","slug":"cptac"}],"developer":"Owkin","evaluation":[{"benchmark":"biomarker prediction suite (TCGA and external cohorts)","dataset_slug":"tcga","external":true,"metric":"AUROC","source":"https://arxiv.org/abs/2409.09173","value":"competitive with UNI/Virchow; reported per task in paper"}],"hf":{"downloads":315410,"fetched_at":"2026-09-09T21:33:09Z","gated":false,"last_modified":"2026-01-19","library":"transformers","license":"other","likes":43,"pipeline_tag":"image-feature-extraction"},"kind":"foundation","license":"owkin non-commercial","limitations":"- Public benchmarks built on TCGA/CPTAC are contaminated for this model \u2014 evaluate on data outside PANCAN-XL.\n- Non-commercial terms.","links":{"demo":null,"docs":null,"doi":null,"github":null,"huggingface":"https://huggingface.co/owkin/phikon-v2","paper":"https://arxiv.org/abs/2409.09173","pmid":null},"modalities":["histopathology"],"name":"Phikon-v2","notable_uses":"","openness":"open-weights","regulatory":{"intended_use_en":"Research.","intended_use_pl":"Badania.","source_url":"https://huggingface.co/owkin/phikon-v2","status":"research-only"},"regulatory_status":"research-only","release_date":"2024-09-13","run_snippet":"# generic timm loader \u2014 check the model card for the exact init args and image normalisation\nimport timm, torch\nfrom huggingface_hub import login\nlogin()  # gated repos: accept the licence on huggingface.co first\nmodel = timm.create_model('hf-hub:owkin/phikon-v2', pretrained=True)\nmodel.eval()\ncfg = timm.data.resolve_data_config({}, model=model)\ntransform = timm.data.create_transform(**cfg)\n# emb = model(transform(tile).unsqueeze(0))  # 1 x D tile embedding\n","settings":["basic-research","prognosis"],"slug":"phikon-v2","sources":[{"label":"Filiot A et al. Phikon-v2, A large and public feature extractor for biomarker prediction. arXiv 2024","url":"https://arxiv.org/abs/2409.09173"},{"label":"Hugging Face \u2014 owkin/phikon-v2","url":"https://huggingface.co/owkin/phikon-v2"}],"summary":"ViT-L pathology encoder trained with DINOv2 on PANCAN-XL \u2014 456 million tiles from 58,359 whole-slide images that mix public cohorts (TCGA, CPTAC, GTEx and others) with private data \u2014 positioned for biomarker prediction.","tasks":["feature-extraction","classification","prognosis"],"training":{"institutions":"Owkin + public cohorts","size":"58,359 WSIs / 456M tiles","summary_en":"PANCAN-XL: 456M tiles from 58,359 WSIs across more than 30 cancer types; roughly half public (TCGA, CPTAC, GTEx, other public cohorts) and half private.","summary_pl":"PANCAN-XL: 456 mln kafelk\u00f3w z 58 359 preparat\u00f3w z ponad 30 typ\u00f3w nowotwor\u00f3w; mniej wi\u0119cej po\u0142owa publiczna (TCGA, CPTAC, GTEx, inne kohorty publiczne), po\u0142owa prywatna."},"updated_at":"2026-09-09T21:33:09.703725","url":"/ai-oncology/models/phikon-v2","usage":{"library":"transformers","notes_en":"Loads with AutoModel; note that TCGA and CPTAC are IN the pretraining set, so do not treat those as external test sets for this model."},"verified_at":"2026-09-05T22:26:01.463515","verified_by":"editorial","version":null,"what_it_does":"Tile embeddings evaluated mainly on predicting molecular biomarkers and clinical outcomes from H&E; successor to Phikon (ViT-B, TCGA-only)."}