{"success":true,"data":{"author":"Tahoe Therapeutics","bucketAccessType":"public","bucketName":"models/tahoebio/Tahoe-x1","bucketPrefix":"","category":"Genomics & Bioinformatics","cloudProvider":"huggingface","createdAt":"2026-11-16T15:25:48.883Z","description":"Open weights for Tahoe-x1 (Tx1), Tahoe Therapeutics' family of single-cell foundation models trained on perturbation data. The models were pretrained on 266 million single-cell transcriptomic profiles, including the Tahoe-100M drug perturbation compendium, and produce cell and gene embeddings for tasks such as cell type annotation, perturbation response and drug sensitivity prediction. Three sizes are provided, each in its own folder: 70m-model, 1b-model (about 1.3B parameters) and 3b-model (about 3B parameters). Each folder holds model.safetensors (and the original best-model.pt checkpoint), model_config.yml, collator_config.yml and vocab.json. Load them with the tahoe_x1 package (ComposerTX) pointed at the mounted folder, and pair them with the Tahoe-100M, CELLxGENE Census or SEA-AD archives to embed cells next to the data. Apache 2.0.","displayTitle":"Tahoe-x1: perturbation-trained single-cell foundation model weights (70M, 1.3B, 3B)","huggingFaceUrl":"https://huggingface.co/tahoebio/Tahoe-x1","id":"c3cf6968-5b07-5807-9ccf-48518b5e8494","license":"Apache-2.0","licenseEvidenceUrl":"https://github.com/tahoebio/tahoe-x1/blob/main/LICENSE","objectCountEstimate":24,"probeStatus":"ok","referenceUrl":"https://www.biorxiv.org/content/10.1101/2025.10.23.683759v1","region":"us","schemaCard":{"accessHints":[{"language":"python-hub","snippet":"from huggingface_hub import snapshot_download\nsnapshot_download(repo_id=\"tahoebio/Tahoe-x1\")  # or read /mnt/archive directly"}],"caveats":["HuggingFace metadata: author=tahoebio, license:apache-2.0, downloads=216","Weights are read lazily from the mount. Copying them to local disk first is faster when a model is loaded many times."],"format":"binary","generatedAt":"2026-11-17T19:17:18.267Z","layout":{"example":"1b-model/best-model.pt","pattern":"{repoId}/{filename}"},"mountPath":"/mnt/archive","sampleObjects":["config.json","1b-model/best-model.pt","1b-model/model.safetensors","3b-model/best-model.pt"],"sidecars":[{"bytes":"---\nlicense: apache-2.0\ndatasets:\n- tahoebio/Tahoe-100M\ntags:\n- biology\n- single-cell\n- RNA\n- chemistry\n- tahoebio\n- pytorch\n---\n\n# Tahoe-x1\n\nTahoe-x1 is a family of perturbation-trained single-cell foundation models with up to 3 billion parameters, developed by Tahoe Therapeutics. Pretrained on 266 million single-cell transcriptomic profiles including the _Tahoe-100M_ perturbation compendium, Tahoe-x1 achieves state-of-the-art performance on cancer-relevant tasks with 3–30× higher compute efficiency than other cell-state models.\n\n**Quick Links:**\n- ✨ [**Blog Post**](https://www.tahoebio.ai/news/tahoe-x1-blog) - Read our announcement\n- 📄 [**Preprint**](https://www.biorxiv.org/content/10.1101/2025.10.23.683759v1) - Read our preprint on bioRxiv\n- 💻 [**GitHub Repository**](https://github.com/tahoebio/tahoe-x1) - Access the code\n- 🎮 [**Interactive Demo**](https://huggingface.co/spaces/tahoebio/tx1-demo) - Try the model with no code required!\n\n## 🤖 Model Sizes\n\nWe provide pretrained weights for three model sizes:\n\n- **Tx1-70M**: ~70M parameters\n- **Tx1-1B**: ~1.3B parameters\n- **Tx1-3B**: ~3B parameters\n\n## 🚀 Quickstart\n\nLoad a model directly from Hugging Face and generate cell embeddings:\n\n```python\nfrom tahoe_x1.model import ComposerTX\nimport scanpy as sc\n\n# Load model from Hugging Face in a single line\n# Options: \"70m\", \"1b\", or \"3b\"\nmodel, vocab, model_cfg, collator_cfg = ComposerTX.from_hf(\n    repo_id=\"tahoebio/Tahoe-x1\",\n    model_size=\"3b\"\n)\n\n# Load your single-cell data\nadata = sc.read_h5ad(\"your_data.h5ad\")\n\n# Generate embeddings (see tutorials for full example)\n# Cell embeddings are stored in adata.obsm\n```\n\n### 📦 Installation\n\nTo use the models, install the `tahoex` package from GitHub:\n\n```bash\n# Clone the repository\ngit clone https://github.com/tahoebio/tahoe-x1\ncd tahoe-x1\n\n# Install using Docker (recommended) or uv\n# See installation guide: https://github.com/tahoebio/tahoe-x1#installation\n```\n\n**Docker installation** provides better reproducibility and is recommended for the best experience. For native installation, use `uv` or `pip` for dependency management.\n\nModel checkpoints and configs are automatically downloaded from this Hugging Face repository when using `ComposerTX.from_hf()`. Training data is hosted publicly on S3 (s3://tahoe-hackathon-data) and will be downloaded as needed.\n\n## 📚 Tutorials\n\nPlease refer to the tutorials in the [GitHub repository](https://github.com/tahoebio/tahoe-x1/tree/main/tutorials) for detailed examples:\n\n| Tutorial | Description | Link |\n|----------|-------------|------|\n| **Clustering Tutorial** | Cell clustering and UMAP visualization with Tahoe-x1 embeddings | [clustering_tutorial.ipynb](https://github.com/tahoebio/tahoe-x1/blob/main/tutorials/clustering_tutorial.ipynb) |\n| **Training Tutorial** | Step-by-step guide to training and fine-tuning Tahoe-x1 models | [training_tutorial.ipynb](https://github.com/tahoebio/tahoe-x1/blob/main/tutorials/training_tutorial.ipynb) |\n\n## 🧬 Generating Cell and Gene Embeddings\n\n### Using Configuration Files\n\n1. Create a configuration file (see `scripts/inference/configs/predict.yaml` in the [GitHub repo](https://github.com/tahoebio/tahoe-x1)):\n```yaml\n# Key configuration options:\n#   - paths.hf_repo_id: Hugging Face repository (tahoebio/Tahoe-x1)\n#   - paths.hf_model_size: model size (70m, 1b, or 3b)\n#   - paths.adata_output: where to save AnnData output including embeddings\n#   - predict.return_gene_embeddings: True (for extracting gene embeddings)\n```\nSee the Clustering Tutorial for a full example of preparing input data and configuration files.\n2. Run the embedding script:\n```bash\npython scripts/inference/predict_embeddings.py path/to/config.yaml\n\n# Optional: override config values via command line\npython scripts/inference/predict_embeddings.py path/to/config.yaml \\\n    --paths.model_name=tx --batch_size=128\n```\n\n### Advanced Usage\n\nFor memory-efficient gene embedding extraction over large datasets, use the lower-level API:\n\n```python\nfrom tahoex.tasks import get_batch_embeddings\n\ncell_embs, gene_embs = get","encoding":"utf8","key":"README.md","name":"README.md","size":6863,"truncated":true}],"sourceObjectKey":"1b-model/best-model.pt","starterCell":"import os, json\nbase = \"/mnt/archive\"   # this model repo, mounted read-only\nfiles = sorted(os.listdir(base)); print(files)\nprint(json.dumps(json.load(open(os.path.join(base, \"config.json\"))), indent=1)[:800])\nimport struct\nwith open(os.path.join(base, \"1b-model/model.safetensors\"), \"rb\") as f:\n    hdr = json.loads(f.read(struct.unpack(\"<Q\", f.read(8))[0]))\ntensors = {k: v for k, v in hdr.items() if k != \"__metadata__\"}\nprint(len(tensors), \"tensors, e.g.\", list(tensors.items())[:3])\n# To load them: pip install safetensors torch, then safetensors.torch.load_file(path)","totalObjects":24},"summary":"Open weights for Tahoe-x1 (Tx1), Tahoe Therapeutics' family of single-cell foundation models trained on perturbation data.","title":"Tahoe-x1: perturbation-trained single-cell foundation model weights (70M, 1.3B, 3B)","updatedAt":"2026-11-17T19:17:18.545Z","visibility":"public","websiteUrl":"https://huggingface.co/tahoebio/Tahoe-x1"}}