{"id":30351696,"url":"https://github.com/priorlabs/nanotabpfn","last_synced_at":"2025-08-18T23:12:07.807Z","repository":{"id":305179709,"uuid":"961817958","full_name":"PriorLabs/nanoTabPFN","owner":"PriorLabs","description":"nanoTabPFN: A Playground for Tabular Foundation Models","archived":false,"fork":false,"pushed_at":"2025-07-23T23:02:29.000Z","size":11,"stargazers_count":18,"open_issues_count":2,"forks_count":3,"subscribers_count":1,"default_branch":"main","last_synced_at":"2025-08-11T20:46:11.319Z","etag":null,"topics":[],"latest_commit_sha":null,"homepage":"","language":"Python","has_issues":true,"has_wiki":null,"has_pages":null,"mirror_url":null,"source_name":null,"license":"apache-2.0","status":null,"scm":"git","pull_requests_enabled":true,"icon_url":"https://github.com/PriorLabs.png","metadata":{"files":{"readme":"README.md","changelog":null,"contributing":null,"funding":null,"license":"LICENSE","code_of_conduct":null,"threat_model":null,"audit":null,"citation":null,"codeowners":null,"security":null,"support":null,"governance":null,"roadmap":null,"authors":null,"dei":null,"publiccode":null,"codemeta":null,"zenodo":null}},"created_at":"2025-04-07T08:02:20.000Z","updated_at":"2025-08-09T21:40:47.000Z","dependencies_parsed_at":"2025-07-18T19:49:30.142Z","dependency_job_id":"91bff218-477b-4f5b-9028-3d1993b69897","html_url":"https://github.com/PriorLabs/nanoTabPFN","commit_stats":null,"previous_names":["priorlabs/nanotabpfn"],"tags_count":0,"template":false,"template_full_name":null,"purl":"pkg:github/PriorLabs/nanoTabPFN","repository_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/repositories/PriorLabs%2FnanoTabPFN","tags_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/repositories/PriorLabs%2FnanoTabPFN/tags","releases_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/repositories/PriorLabs%2FnanoTabPFN/releases","manifests_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/repositories/PriorLabs%2FnanoTabPFN/manifests","owner_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/owners/PriorLabs","download_url":"https://codeload.github.com/PriorLabs/nanoTabPFN/tar.gz/refs/heads/main","sbom_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/repositories/PriorLabs%2FnanoTabPFN/sbom","scorecard":null,"host":{"name":"GitHub","url":"https://github.com","kind":"github","repositories_count":271073387,"owners_count":24694538,"icon_url":"https://github.com/github.png","version":null,"created_at":"2022-05-30T11:31:42.601Z","updated_at":"2022-07-04T15:15:14.044Z","status":"online","status_checked_at":"2025-08-18T02:00:08.743Z","response_time":89,"last_error":null,"robots_txt_status":"success","robots_txt_updated_at":"2025-07-24T06:49:26.215Z","robots_txt_url":"https://github.com/robots.txt","online":true,"can_crawl_api":true,"host_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub","repositories_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/repositories","repository_names_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/repository_names","owners_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/owners"}},"keywords":[],"created_at":"2025-08-18T23:12:07.093Z","updated_at":"2025-08-18T23:12:07.796Z","avatar_url":"https://github.com/PriorLabs.png","language":"Python","funding_links":[],"categories":[],"sub_categories":[],"readme":"# nanoTabPFN\n\nThe purpose of this repository is to provide a fully open source playground for tabular foundation models.\nIt contains a much smaller and simpler implementation of the TabPFNv2 architecture as well as a training loop and code for loading data that was pre-generated by a prior. We are planning to rapidly extend the repository with more features (e.g. regression, missing values, categorical features), prior interfaces and architectures.\nIt is supposed to be a good starting point for students and researchers that are interested in learning about how TabPFN works under the hood.\n\nClone the repository, afterwards install dependencies via:\n```\npip install -e .\n```\n\nWe offer the same interface as TabPFN:\n```python\nfrom sklearn.datasets import load_breast_cancer\nfrom sklearn.metrics import accuracy_score, roc_auc_score\nfrom sklearn.model_selection import train_test_split\n\nfrom nanotabpfn import NanoTabPFNClassifier\n\n# Load data\nX, y = load_breast_cancer(return_X_y=True)\nX_train, X_test, y_train, y_test = train_test_split(X, y, test_size=0.5, random_state=42)\n\n# Initialize a classifier\nclf = NanoTabPFNClassifier()\nclf.fit(X_train, y_train)\n\n# Predict probabilities\nprediction_probabilities = clf.predict_proba(X_test)\nprint(\"ROC AUC:\", roc_auc_score(y_test, prediction_probabilities[:, 1]))\n\n# Predict labels\npredictions = clf.predict(X_test)\nprint(\"Accuracy\", accuracy_score(y_test, predictions))\n```\n\n### Our Code\n\n`nanotabpfn/model.py` contains the implementation of the architecture in less than 250 lines of code. `nanotabpfn/train.py` implements a simple training loop in under 100 lines and `nanotabpfn/priors.py` implements a dataloader that allows you to load a dump pre-generated from a prior.\nWe will release multiple dumps of different scales soon. We also offer an interface where you can provide your own get\\_batch function.\n\n### Pretrain your own small nanoTabPFN\nFirst we download 100k pre-generated datasets with 50 datapoints, 3 features and up to 3  classes each from [here](https://ml.informatik.uni-freiburg.de/research-artifacts/pfefferle/nanoTabPFN/50x3_3_100k_classification.h5).\n\nThen you can run:\n```\npython pretrain_classification.py -epochs 80 -steps 25 -batchsize 50 -priordump 50x3_3_100k_classification.h5\n```\nThis should take less than 5 min on a modern NVIDIA GPU (around 10 minutes on Macbook M4 Pro GPU and around 40 min on M4 Pro CPU).\n\n#### Step by Step Explanation\n\nFirst we import our Architecture, Prior interface and training loop, etc.\n```python\nfrom nanotabpfn.model import NanoTabPFNModel\nfrom nanotabpfn.priors import PriorDumpDataLoader\nfrom nanotabpfn.train import train\nfrom nanotabpfn.utils import get_default_device\nfrom nanotabpfn.interface import NanoTabPFNClassifier\nfrom torch.nn import CrossEntropyLoss\n```\nthen we instantiate our model and loss criterion:\n```python\nmodel = NanoTabPFNModel(\n    num_attention_heads=6,\n    embedding_size=192,\n    mlp_hidden_size=768,\n    num_layers=6,\n    num_outputs=10,\n)\ncriterion = CrossEntropyLoss()\n```\nthen we instantiate our prior:\n```python\ndevice = get_default_device()\nprior = PriorDumpDataLoader(filename='50x3_3_100k_classification.h5', num_steps=25, batch_size=50, device=device)\n```\nand finally train our model:\n```python\ndef epoch_callback(epoch, epoch_time, mean_loss, model):\n    classifier = NanoTabPFNClassifier(model, device)\n    # you can add your own eval code here that runs after every epoch\n    print(f'epoch {epoch:5d} | time {epoch_time:5.2f}s | mean loss {mean_loss:5.2f}', flush=True)\n\ntrained_model, loss = train(\n    model=model,\n    prior=prior,\n    criterion=criterion,\n    epochs=80,\n    device=device,\n    epoch_callback=epoch_callback\n)\n```\n","project_url":"https://awesome.ecosyste.ms/api/v1/projects/github.com%2Fpriorlabs%2Fnanotabpfn","html_url":"https://awesome.ecosyste.ms/projects/github.com%2Fpriorlabs%2Fnanotabpfn","lists_url":"https://awesome.ecosyste.ms/api/v1/projects/github.com%2Fpriorlabs%2Fnanotabpfn/lists"}