{"id":15601021,"url":"https://github.com/lucidrains/alphafold3-pytorch","last_synced_at":"2025-05-14T00:04:53.094Z","repository":{"id":238877252,"uuid":"797821110","full_name":"lucidrains/alphafold3-pytorch","owner":"lucidrains","description":"Implementation of Alphafold 3 from Google Deepmind in Pytorch","archived":false,"fork":false,"pushed_at":"2025-01-22T15:31:28.000Z","size":14217,"stargazers_count":1406,"open_issues_count":17,"forks_count":179,"subscribers_count":46,"default_branch":"main","last_synced_at":"2025-04-10T01:06:03.878Z","etag":null,"topics":["artificial-intelligence","attention-mechanisms","deep-learning","denoising-diffusion","protein-structure-prediction","transformers"],"latest_commit_sha":null,"homepage":"","language":"Python","has_issues":true,"has_wiki":null,"has_pages":null,"mirror_url":null,"source_name":null,"license":"mit","status":null,"scm":"git","pull_requests_enabled":true,"icon_url":"https://github.com/lucidrains.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}},"created_at":"2024-05-08T15:26:29.000Z","updated_at":"2025-04-09T12:19:13.000Z","dependencies_parsed_at":"2024-05-08T17:14:10.075Z","dependency_job_id":"6946e4e3-c34f-424c-b168-f2898f6e0d6a","html_url":"https://github.com/lucidrains/alphafold3-pytorch","commit_stats":{"total_commits":978,"total_committers":17,"mean_commits":"57.529411764705884","dds":"0.20756646216768915","last_synced_commit":"2cdc53bee09a171d8970737e154135503ef61c83"},"previous_names":["lucidrains/alphafold3-pytorch"],"tags_count":475,"template":false,"template_full_name":null,"repository_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/repositories/lucidrains%2Falphafold3-pytorch","tags_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/repositories/lucidrains%2Falphafold3-pytorch/tags","releases_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/repositories/lucidrains%2Falphafold3-pytorch/releases","manifests_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/repositories/lucidrains%2Falphafold3-pytorch/manifests","owner_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/owners/lucidrains","download_url":"https://codeload.github.com/lucidrains/alphafold3-pytorch/tar.gz/refs/heads/main","host":{"name":"GitHub","url":"https://github.com","kind":"github","repositories_count":254043310,"owners_count":22004924,"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","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":["artificial-intelligence","attention-mechanisms","deep-learning","denoising-diffusion","protein-structure-prediction","transformers"],"created_at":"2024-10-03T02:12:23.842Z","updated_at":"2025-05-14T00:04:53.025Z","avatar_url":"https://github.com/lucidrains.png","language":"Python","funding_links":[],"categories":["Python"],"sub_categories":[],"readme":"\u003cimg src=\"./alphafold3.png\" width=\"500px\"\u003e\u003c/img\u003e\n\n## Alphafold 3 - Pytorch\n\nImplementation of \u003ca href=\"https://www.nature.com/articles/s41586-024-07487-w\"\u003eAlphafold 3\u003c/a\u003e in Pytorch\n\nYou can chat with other researchers about this work [here](https://discord.gg/Xsq4WaBR9S)\n\n\u003ca href=\"https://www.youtube.com/watch?v=qjFgthkKxcA\"\u003eReview of the paper\u003c/a\u003e by \u003ca href=\"https://x.com/sokrypton\"\u003eSergey\u003c/a\u003e\n\n\u003ca href=\"https://elanapearl.github.io/blog/2024/the-illustrated-alphafold/\"\u003eIllustrated guide\u003c/a\u003e by \u003ca href=\"https://elanapearl.github.io/\"\u003eElana P. Simon\u003c/a\u003e\n\n\u003ca href=\"https://www.youtube.com/watch?v=AE35XCN5NuU\"\u003eTalk by Max Jaderberg\u003c/a\u003e\n\nA fork with full Lightning + Hydra support is being maintained by \u003ca href=\"https://github.com/amorehead\"\u003eAlex\u003c/a\u003e at \u003ca href=\"https://github.com/amorehead/alphafold3-pytorch-lightning-hydra\"\u003ethis repository\u003c/a\u003e\n\nA visualization of the molecules of life used in the repository can be seen and interacted with \u003ca href=\"https://colab.research.google.com/drive/1S9TD8WmS1Gjgwjo9xyEYTbdxgMVVZcQe?usp=sharing\"\u003ehere\u003c/a\u003e\n\n## Appreciation\n\n- \u003ca href=\"https://github.com/joseph-c-kim\"\u003eJoseph\u003c/a\u003e for contributing the Relative Positional Encoding and the Smooth LDDT Loss!\n\n- \u003ca href=\"https://github.com/engelberger\"\u003eFelipe\u003c/a\u003e for contributing Weighted Rigid Align, Express Coordinates In Frame, Compute Alignment Error, and Centre Random Augmentation modules!\n\n- \u003ca href=\"https://github.com/amorehead\"\u003eAlex\u003c/a\u003e for fixing various issues in the transcribed algorithms\n\n- \u003ca href=\"https://github.com/gitabtion\"\u003eHeng\u003c/a\u003e for pointing out inconsistencies with the paper and pull requesting the solutions\n\n- \u003ca href=\"https://github.com/gitabtion\"\u003eHeng\u003c/a\u003e for finding an issue with the molecular atom indices for the distogram loss\n\n- \u003ca href=\"https://github.com/luwei0917\"\u003eWei Lu\u003c/a\u003e for catching a few erroneous hyperparameters\n\n- \u003ca href=\"https://github.com/amorehead\"\u003eAlex\u003c/a\u003e for the PDB dataset preparation script!\n\n- \u003ca href=\"https://github.com/milot-mirdita\"\u003eMilot\u003c/a\u003e for optimizing the PDB dataset clustering script!\n\n- \u003ca href=\"https://github.com/amorehead\"\u003eAlex\u003c/a\u003e for basically writing the entire gargantuan flow from parsing the PDB all the way to the molecule and atomic inputs for training\n\n- \u003ca href=\"https://github.com/vandrw\"\u003eAndrei\u003c/a\u003e for working on the weighted PDB dataset sampling!\n\n- \u003ca href=\"https://github.com/tanjimin\"\u003eJimin\u003c/a\u003e for submitting a small fix to an issue with the coordinates being passed into `WeightedRigidAlign`\n\n- \u003ca href=\"https://github.com/xluo233\"\u003e@xluo233\u003c/a\u003e for contributing the confidence measures, clash penalty ranking, and sample ranking logic!\n\n- \u003ca href=\"https://github.com/sj900\"\u003esj900\u003c/a\u003e for integrating and testing the `WeightedPDBSampler` within the `PDBDataset` and for adding initial support for MSA and template parsing!\n\n- \u003ca href=\"https://github.com/xluo233\"\u003e@xluo233\u003c/a\u003e again for contributing the logic for computing the model selection score as well as the unresolved rasa!\n\n- \u003ca href=\"https://github.com/wufandi\"\u003eFandi\u003c/a\u003e for discovering a few inconsistencies in the elucidated atom diffusion module with the supplementary\n  \n- \u003ca href=\"https://github.com/ptosco\"\u003ePaolo\u003c/a\u003e for proposing the `PDB neutral stable molecule` hypothesis!\n\n- \u003ca href=\"https://github.com/dhuvik\"\u003eDhuvi\u003c/a\u003e for fixing a bug related to metal ion molecule ID assignment for `Alphafold3Inputs`!\n\n- \u003ca href=\"https://github.com/dhuvik\"\u003eDhuvi\u003c/a\u003e for taking on the logic for translating `Alphafold3Input` to `BioMolecule` for saving to mmCIF!\n\n- Tom (from the Discord channel) for identifying a discrepancy between this codebase's distogram and template unit vector computations and those of OpenFold (and \u003ca href=\"https://github.com/vandrw\"\u003eAndrei\u003c/a\u003e for helping address the distogram issue)!\n\n- \u003ca href=\"https://github.com/Kaihui-Cheng\"\u003eKaihui\u003c/a\u003e for identifying a bug in how non-standard atoms were handled in polymer residues!\n\n- \u003ca href=\"https://github.com/vandrw\"\u003eAndrei\u003c/a\u003e for taking on the gradio frontend interface!\n\n- \u003ca href=\"https://github.com/patrick-kidger\"\u003ePatrick\u003c/a\u003e for \u003ca href=\"https://docs.kidger.site/jaxtyping/\"\u003ejaxtyping\u003c/a\u003e, \u003ca href=\"https://github.com/fferflo\"\u003eFlorian\u003c/a\u003e for \u003ca href=\"https://github.com/fferflo/einx\"\u003eeinx\u003c/a\u003e, and of course, \u003ca href=\"https://github.com/arogozhnikov\"\u003eAlex\u003c/a\u003e for \u003ca href=\"https://einops.rocks/\"\u003eeinops\u003c/a\u003e\n\n- Soumith and the Pytorch organization for giving me the opportunity to open source this work\n\n## Install\n\n```bash\n$ pip install alphafold3-pytorch\n```\n\n## Usage\n\n```python\nimport torch\nfrom alphafold3_pytorch import Alphafold3\nfrom alphafold3_pytorch.utils.model_utils import exclusive_cumsum\n\nalphafold3 = Alphafold3(\n    dim_atom_inputs = 77,\n    dim_template_feats = 108\n)\n\n# mock inputs\n\nseq_len = 16\n\nmolecule_atom_indices = torch.randint(0, 2, (2, seq_len)).long()\nmolecule_atom_lens = torch.full((2, seq_len), 2).long()\n\natom_seq_len = molecule_atom_lens.sum(dim=-1).amax()\natom_offsets = exclusive_cumsum(molecule_atom_lens)\n\natom_inputs = torch.randn(2, atom_seq_len, 77)\natompair_inputs = torch.randn(2, atom_seq_len, atom_seq_len, 5)\n\nadditional_molecule_feats = torch.randint(0, 2, (2, seq_len, 5))\nadditional_token_feats = torch.randn(2, seq_len, 33)\nis_molecule_types = torch.randint(0, 2, (2, seq_len, 5)).bool()\nis_molecule_mod = torch.randint(0, 2, (2, seq_len, 4)).bool()\nmolecule_ids = torch.randint(0, 32, (2, seq_len))\n\ntemplate_feats = torch.randn(2, 2, seq_len, seq_len, 108)\ntemplate_mask = torch.ones((2, 2)).bool()\n\nmsa = torch.randn(2, 7, seq_len, 32)\nmsa_mask = torch.ones((2, 7)).bool()\n\nadditional_msa_feats = torch.randn(2, 7, seq_len, 2)\n\n# required for training, but omitted on inference\n\natom_pos = torch.randn(2, atom_seq_len, 3)\n\ndistogram_atom_indices = molecule_atom_lens - 1\n\ndistance_labels = torch.randint(0, 37, (2, seq_len, seq_len))\nresolved_labels = torch.randint(0, 2, (2, atom_seq_len))\n\n# offset indices correctly\n\ndistogram_atom_indices += atom_offsets\nmolecule_atom_indices += atom_offsets\n\n# train\n\nloss = alphafold3(\n    num_recycling_steps = 2,\n    atom_inputs = atom_inputs,\n    atompair_inputs = atompair_inputs,\n    molecule_ids = molecule_ids,\n    molecule_atom_lens = molecule_atom_lens,\n    additional_molecule_feats = additional_molecule_feats,\n    additional_msa_feats = additional_msa_feats,\n    additional_token_feats = additional_token_feats,\n    is_molecule_types = is_molecule_types,\n    is_molecule_mod = is_molecule_mod,\n    msa = msa,\n    msa_mask = msa_mask,\n    templates = template_feats,\n    template_mask = template_mask,\n    atom_pos = atom_pos,\n    distogram_atom_indices = distogram_atom_indices,\n    molecule_atom_indices = molecule_atom_indices,\n    distance_labels = distance_labels,\n    resolved_labels = resolved_labels\n)\n\nloss.backward()\n\n# after much training ...\n\nsampled_atom_pos = alphafold3(\n    num_recycling_steps = 4,\n    num_sample_steps = 16,\n    atom_inputs = atom_inputs,\n    atompair_inputs = atompair_inputs,\n    molecule_ids = molecule_ids,\n    molecule_atom_lens = molecule_atom_lens,\n    additional_molecule_feats = additional_molecule_feats,\n    additional_msa_feats = additional_msa_feats,\n    additional_token_feats = additional_token_feats,\n    is_molecule_types = is_molecule_types,\n    is_molecule_mod = is_molecule_mod,\n    msa = msa,\n    msa_mask = msa_mask,\n    templates = template_feats,\n    template_mask = template_mask\n)\n\nsampled_atom_pos.shape # (2, \u003catom_seqlen\u003e, 3)\n```\n\nAn example with molecule level input handling\n\n```python\nimport torch\nfrom alphafold3_pytorch import Alphafold3, Alphafold3Input\n\ncontrived_protein = 'AG'\n\nmock_atompos = [\n    torch.randn(5, 3),   # alanine has 5 non-hydrogen atoms\n    torch.randn(4, 3)    # glycine has 4 non-hydrogen atoms\n]\n\ntrain_alphafold3_input = Alphafold3Input(\n    proteins = [contrived_protein],\n    atom_pos = mock_atompos\n)\n\neval_alphafold3_input = Alphafold3Input(\n    proteins = [contrived_protein]\n)\n\n# training\n\nalphafold3 = Alphafold3(\n    dim_atom_inputs = 3,\n    dim_atompair_inputs = 5,\n    atoms_per_window = 27,\n    dim_template_feats = 108,\n    num_molecule_mods = 0,\n    confidence_head_kwargs = dict(\n        pairformer_depth = 1\n    ),\n    template_embedder_kwargs = dict(\n        pairformer_stack_depth = 1\n    ),\n    msa_module_kwargs = dict(\n        depth = 1\n    ),\n    pairformer_stack = dict(\n        depth = 2\n    ),\n    diffusion_module_kwargs = dict(\n        atom_encoder_depth = 1,\n        token_transformer_depth = 1,\n        atom_decoder_depth = 1,\n    )\n)\n\nloss = alphafold3.forward_with_alphafold3_inputs([train_alphafold3_input])\nloss.backward()\n\n# sampling\n\nalphafold3.eval()\nsampled_atom_pos = alphafold3.forward_with_alphafold3_inputs(eval_alphafold3_input)\n\nassert sampled_atom_pos.shape == (1, (5 + 4), 3)\n```\n\n## Data preparation\n\n### PDB dataset curation\n\nTo acquire the AlphaFold 3 PDB dataset, first download all first-assembly (and asymmetric unit) complexes in the Protein Data Bank (PDB), and then preprocess them with the script referenced below. The PDB can be downloaded from the RCSB: https://www.wwpdb.org/ftp/pdb-ftp-sites#rcsbpdb. The two Python scripts below (i.e., `filter_pdb_{train,val,test}_mmcifs.py` and `cluster_pdb_{train,val,test}_mmcifs.py`) assume you have downloaded the PDB in the **mmCIF file format**, placing its first-assembly and asymmetric unit mmCIF files at `data/pdb_data/unfiltered_assembly_mmcifs/` and `data/pdb_data/unfiltered_asym_mmcifs/`, respectively.\n\nFor reproducibility, we recommend downloading the PDB using AWS snapshots (e.g., `20240101`). To do so, refer to [AWS's documentation](https://docs.aws.amazon.com/cli/latest/userguide/cli-chap-welcome.html) to set up the AWS CLI locally. Alternatively, on the RCSB website, navigate down to \"Download Protocols\", and follow the download instructions depending on your location.\n\nFor example, one can use the following commands to download the PDB as two collections of mmCIF files:\n\n```bash\n# For `assembly1` complexes, use the PDB's `20240101` AWS snapshot:\naws s3 sync s3://pdbsnapshots/20240101/pub/pdb/data/assemblies/mmCIF/divided/ ./data/pdb_data/unfiltered_assembly_mmcifs\n# Or as a fallback, use rsync:\nrsync -rlpt -v -z --delete --port=33444 \\\nrsync.rcsb.org::ftp_data/assemblies/mmCIF/divided/ ./data/pdb_data/unfiltered_assembly_mmcifs/\n\n# For asymmetric unit complexes, also use the PDB's `20240101` AWS snapshot:\naws s3 sync s3://pdbsnapshots/20240101/pub/pdb/data/structures/divided/mmCIF/ ./data/pdb_data/unfiltered_asym_mmcifs\n# Or as a fallback, use rsync:\nrsync -rlpt -v -z --delete --port=33444 \\\nrsync.rcsb.org::ftp_data/structures/divided/mmCIF/ ./data/pdb_data/unfiltered_asym_mmcifs/\n```\n\n\u003e WARNING: Downloading the PDB can take up to 700GB of space.\n\n\u003e NOTE: The PDB hosts all available AWS snapshots here: https://pdbsnapshots.s3.us-west-2.amazonaws.com/index.html.\n\nAfter downloading, you should have two directories formatted like this:\nhttps://files.rcsb.org/pub/pdb/data/assemblies/mmCIF/divided/ \u0026 https://files.rcsb.org/pub/pdb/data/structures/divided/mmCIF/\n\n```bash\n00/\n01/\n02/\n..\nzz/\n```\n\nFor these directories, unzip all the files:\n\n```bash\nfind ./data/pdb_data/unfiltered_assembly_mmcifs/ -type f -name \"*.gz\" -exec gzip -d {} \\;\nfind ./data/pdb_data/unfiltered_asym_mmcifs/ -type f -name \"*.gz\" -exec gzip -d {} \\;\n```\n\nNext run the commands\n\n```bash\nwget -P ./data/ccd_data/ https://files.wwpdb.org/pub/pdb/data/monomers/components.cif.gz\nwget -P ./data/ccd_data/ https://files.wwpdb.org/pub/pdb/data/component-models/complete/chem_comp_model.cif.gz\n```\n\nfrom the project's root directory to download the latest version of the PDB's Chemical Component Dictionary (CCD) and its structural models. Extract each of these files using the following command:\n\n```bash\nfind data/ccd_data/ -type f -name \"*.gz\" -exec gzip -d {} \\;\n```\n\n### PDB dataset filtering\n\nThen run the following with `pdb_assembly_dir`, `pdb_asym_dir`, `ccd_dir`, and `mmcif_output_dir` replaced with the locations of your local copies of the first-assembly PDB, asymmetric unit PDB, CCD, and your desired dataset output directory (i.e., `./data/pdb_data/unfiltered_assembly_mmcifs/`, `./data/pdb_data/unfiltered_asym_mmcifs/`, `./data/ccd_data/`, and `./data/pdb_data/{train,val,test}_mmcifs/`).\n\n```bash\npython scripts/filter_pdb_train_mmcifs.py --mmcif_assembly_dir \u003cpdb_assembly_dir\u003e --mmcif_asym_dir \u003cpdb_asym_dir\u003e --ccd_dir \u003cccd_dir\u003e --output_dir \u003cmmcif_output_dir\u003e\npython scripts/filter_pdb_val_mmcifs.py --mmcif_assembly_dir \u003cpdb_assembly_dir\u003e --mmcif_asym_dir \u003cpdb_asym_dir\u003e --output_dir \u003cmmcif_output_dir\u003e\npython scripts/filter_pdb_test_mmcifs.py --mmcif_assembly_dir \u003cpdb_assembly_dir\u003e --mmcif_asym_dir \u003cpdb_asym_dir\u003e --output_dir \u003cmmcif_output_dir\u003e\n```\n\nSee the scripts for more options. Each first-assembly mmCIF that successfully passes\nall processing steps will be written to `mmcif_output_dir` within a subdirectory\nnamed according to the mmCIF's second and third PDB ID characters (e.g. `5c`).\n\n### PDB dataset clustering\n\nNext, run the following with `mmcif_dir` and `{train,val,test}_clustering_output_dir` replaced, respectively, with your local output directory created using the dataset filtering script above and with your desired clustering output directories (i.e., `./data/pdb_data/{train,val,test}_mmcifs/` and `./data/pdb_data/data_caches/{train,val,test}_clusterings/`):\n\n```bash\npython scripts/cluster_pdb_train_mmcifs.py --mmcif_dir \u003cmmcif_dir\u003e --output_dir \u003ctrain_clustering_output_dir\u003e --clustering_filtered_pdb_dataset\npython scripts/cluster_pdb_val_mmcifs.py --mmcif_dir \u003cmmcif_dir\u003e --reference_clustering_dir \u003ctrain_clustering_output_dir\u003e --output_dir \u003cval_clustering_output_dir\u003e --clustering_filtered_pdb_dataset\npython scripts/cluster_pdb_test_mmcifs.py --mmcif_dir \u003cmmcif_dir\u003e --reference_1_clustering_dir \u003ctrain_clustering_output_dir\u003e --reference_2_clustering_dir \u003cval_clustering_output_dir\u003e --output_dir \u003ctest_clustering_output_dir\u003e --clustering_filtered_pdb_dataset\n```\n\n**Note**: The `--clustering_filtered_pdb_dataset` flag is recommended when clustering the filtered PDB dataset as curated using the scripts above, as this flag will enable faster runtimes in this context (since filtering leaves each chain's residue IDs 1-based). However, this flag must **not** be provided when clustering other (i.e., non-PDB) datasets of mmCIF files. Otherwise, interface clustering may be performed incorrectly, as these datasets' mmCIF files may not use strict 1-based residue indexing for each chain.\n\n**Note**: One can instead download preprocessed (i.e., filtered) mmCIF (`train`/`val`/`test`) files (~25GB, comprising 148k complexes) and chain/interface clustering (`train`/`val`/`test`) files (~3GB) for the PDB's `20240101` AWS snapshot via a [shared OneDrive folder](https://mailmissouri-my.sharepoint.com/:f:/g/personal/acmwhb_umsystem_edu/EqU8tjUmmKxJr-FAlq4tzaIBi2TIBtmw5Vl3k_kmgNlepA?e=mzlyv6). Each of these `tar.gz` archives should be decompressed within the `data/pdb_data/` directory e.g., via `tar -xzf data_caches.tar.gz -C data/pdb_data/`. One can also download and prepare PDB distillation data using as a reference the script `scripts/distillation_data_download.sh`. Once downloaded, one can run `scripts/reduce_uniprot_predictions_to_pdb.py` to filter this dataset to only examples associated with at least one PDB entry. Moreover, for convenience, a [mapping](https://ftp.uniprot.org/pub/databases/uniprot/current_release/knowledgebase/idmapping/idmapping.dat.gz) of UniProt accession IDs to PDB IDs for training on PDB distillation data has already been downloaded and extracted as `data/afdb_data/data_caches/uniprot_to_pdb_id_mapping.dat`.\n\n## Contributing\n\nAt the project root, run\n\n```bash\n$ sh ./contribute.sh\n```\n\nThen, add your module to `alphafold3_pytorch/alphafold3.py`, add your tests to `tests/test_af3.py`, and submit a pull request. You can run the tests locally with\n\n```bash\n$ pytest tests/\n```\n\n## Docker Image\nThe included `Dockerfile` contains the required dependencies to run the package and to train/inference using PyTorch with GPUs.\n\nThe default base image is `pytorch/pytorch:2.3.0-cuda12.1-cudnn8-runtime` and installs the latest version of this package from the `main` GitHub branch.\n\n```bash\n## Build Docker Container\ndocker build -t af3 .\n```\n\nAlternatively, use build arguments to rebuild the image with different software versions:\n- `PYTORCH_TAG`: Changes the base image and thus builds with different PyTorch, CUDA, and/or cuDNN versions.\n- `GIT_TAG`: Changes the tag of this repo to clone and install the package.\n\nFor example:\n```bash\n## Use build argument to change versions\ndocker build --build-arg \"PYTORCH_TAG=2.2.1-cuda12.1-cudnn8-devel\" --build-arg \"GIT_TAG=0.1.15\" -t af3 .\n```\n\nThen, run the container with GPUs and mount a local volume (for training) using the following command:\n\n```bash\n## Run Container\ndocker run -v .:/data --gpus all -it af3\n```\n\n## Citations\n\n```bibtex\n@article{Abramson2024-fj,\n  title    = \"Accurate structure prediction of biomolecular interactions with\n              {AlphaFold} 3\",\n  author   = \"Abramson, Josh and Adler, Jonas and Dunger, Jack and Evans,\n              Richard and Green, Tim and Pritzel, Alexander and Ronneberger,\n              Olaf and Willmore, Lindsay and Ballard, Andrew J and Bambrick,\n              Joshua and Bodenstein, Sebastian W and Evans, David A and Hung,\n              Chia-Chun and O'Neill, Michael and Reiman, David and\n              Tunyasuvunakool, Kathryn and Wu, Zachary and {\\v Z}emgulyt{\\.e},\n              Akvil{\\.e} and Arvaniti, Eirini and Beattie, Charles and\n              Bertolli, Ottavia and Bridgland, Alex and Cherepanov, Alexey and\n              Congreve, Miles and Cowen-Rivers, Alexander I and Cowie, Andrew\n              and Figurnov, Michael and Fuchs, Fabian B and Gladman, Hannah and\n              Jain, Rishub and Khan, Yousuf A and Low, Caroline M R and Perlin,\n              Kuba and Potapenko, Anna and Savy, Pascal and Singh, Sukhdeep and\n              Stecula, Adrian and Thillaisundaram, Ashok and Tong, Catherine\n              and Yakneen, Sergei and Zhong, Ellen D and Zielinski, Michal and\n              {\\v Z}{\\'\\i}dek, Augustin and Bapst, Victor and Kohli, Pushmeet\n              and Jaderberg, Max and Hassabis, Demis and Jumper, John M\",\n  journal  = \"Nature\",\n  month    = \"May\",\n  year     =  2024\n}\n```\n\n```bibtex\n@inproceedings{Darcet2023VisionTN,\n    title   = {Vision Transformers Need Registers},\n    author  = {Timoth'ee Darcet and Maxime Oquab and Julien Mairal and Piotr Bojanowski},\n    year    = {2023},\n    url     = {https://api.semanticscholar.org/CorpusID:263134283}\n}\n```\n\n```bibtex\n@article{Arora2024SimpleLA,\n    title   = {Simple linear attention language models balance the recall-throughput tradeoff},\n    author  = {Simran Arora and Sabri Eyuboglu and Michael Zhang and Aman Timalsina and Silas Alberti and Dylan Zinsley and James Zou and Atri Rudra and Christopher R'e},\n    journal = {ArXiv},\n    year    = {2024},\n    volume  = {abs/2402.18668},\n    url     = {https://api.semanticscholar.org/CorpusID:268063190}\n}\n```\n\n```bibtex\n@article{Puny2021FrameAF,\n    title   = {Frame Averaging for Invariant and Equivariant Network Design},\n    author  = {Omri Puny and Matan Atzmon and Heli Ben-Hamu and Edward James Smith and Ishan Misra and Aditya Grover and Yaron Lipman},\n    journal = {ArXiv},\n    year    = {2021},\n    volume  = {abs/2110.03336},\n    url     = {https://api.semanticscholar.org/CorpusID:238419638}\n}\n```\n\n```bibtex\n@article{Duval2023FAENetFA,\n    title   = {FAENet: Frame Averaging Equivariant GNN for Materials Modeling},\n    author  = {Alexandre Duval and Victor Schmidt and Alex Hernandez Garcia and Santiago Miret and Fragkiskos D. Malliaros and Yoshua Bengio and David Rolnick},\n    journal = {ArXiv},\n    year    = {2023},\n    volume  = {abs/2305.05577},\n    url     = {https://api.semanticscholar.org/CorpusID:258564608}\n}\n```\n\n```bibtex\n@article{Wang2022DeepNetST,\n    title   = {DeepNet: Scaling Transformers to 1, 000 Layers},\n    author  = {Hongyu Wang and Shuming Ma and Li Dong and Shaohan Huang and Dongdong Zhang and Furu Wei},\n    journal = {ArXiv},\n    year    = {2022},\n    volume  = {abs/2203.00555},\n    url     = {https://api.semanticscholar.org/CorpusID:247187905}\n}\n```\n\n```bibtex\n@inproceedings{Ainslie2023CoLT5FL,\n    title   = {CoLT5: Faster Long-Range Transformers with Conditional Computation},\n    author  = {Joshua Ainslie and Tao Lei and Michiel de Jong and Santiago Ontan'on and Siddhartha Brahma and Yury Zemlyanskiy and David Uthus and Mandy Guo and James Lee-Thorp and Yi Tay and Yun-Hsuan Sung and Sumit Sanghai},\n    year    = {2023}\n}\n```\n\n```bibtex\n@article{Ash2019OnTD,\n    title   = {On the Difficulty of Warm-Starting Neural Network Training},\n    author  = {Jordan T. Ash and Ryan P. Adams},\n    journal = {ArXiv},\n    year    = {2019},\n    volume  = {abs/1910.08475},\n    url     = {https://api.semanticscholar.org/CorpusID:204788802}\n}\n```\n\n```bibtex\n@ARTICLE{Heinzinger2023.07.23.550085,\n    author  = {Michael Heinzinger and Konstantin Weissenow and Joaquin Gomez Sanchez and Adrian Henkel and Martin Steinegger and Burkhard Rost},\n    title   = {ProstT5: Bilingual Language Model for Protein Sequence and Structure},\n    year    = {2023},\n    doi     = {10.1101/2023.07.23.550085},\n    journal = {bioRxiv}\n}\n```\n\n```bibtex\n@article {Lin2022.07.20.500902,\n    author  = {Lin, Zeming and Akin, Halil and Rao, Roshan and Hie, Brian and Zhu, Zhongkai and Lu, Wenting and Santos Costa, Allan dos and Fazel-Zarandi, Maryam and Sercu, Tom and Candido, Sal and Rives, Alexander},\n    title   = {Language models of protein sequences at the scale of evolution enable accurate structure prediction},\n    elocation-id = {2022.07.20.500902},\n    year    = {2022},\n    doi     = {10.1101/2022.07.20.500902},\n    publisher = {Cold Spring Harbor Laboratory},\n    URL     = {https://www.biorxiv.org/content/early/2022/07/21/2022.07.20.500902},\n    eprint  = {https://www.biorxiv.org/content/early/2022/07/21/2022.07.20.500902.full.pdf},\n    journal = {bioRxiv}\n}\n```\n\n```bibtex\n@article{Li2024SwitchEA,\n    title   = {Switch EMA: A Free Lunch for Better Flatness and Sharpness},\n    author  = {Siyuan Li and Zicheng Liu and Juanxi Tian and Ge Wang and Zedong Wang and Weiyang Jin and Di Wu and Cheng Tan and Tao Lin and Yang Liu and Baigui Sun and Stan Z. Li},\n    journal = {ArXiv},\n    year    = {2024},\n    volume  = {abs/2402.09240},\n    url     = {https://api.semanticscholar.org/CorpusID:267657558}\n}\n```\n\n```bibtex\n@article{Nguyen2023MitigatingOI,\n    title   = {Mitigating Over-smoothing in Transformers via Regularized Nonlocal Functionals},\n    author  = {Tam Nguyen and Tan M. Nguyen and Richard G. Baraniuk},\n    journal = {ArXiv},\n    year    = {2023},\n    volume  = {abs/2312.00751},\n    url     = {https://api.semanticscholar.org/CorpusID:264300597}\n}\n```\n\n```bibtex\n@inproceedings{Zhou2024ValueRL,\n    title   = {Value Residual Learning For Alleviating Attention Concentration In Transformers},\n    author  = {Zhanchao Zhou and Tianyi Wu and Zhiyun Jiang and Zhenzhong Lan},\n    year    = {2024},\n    url     = {https://api.semanticscholar.org/CorpusID:273532030}\n}\n```\n\n```bibtex\n@article{Zhu2024HyperConnections,\n    title   = {Hyper-Connections},\n    author  = {Defa Zhu and Hongzhi Huang and Zihao Huang and Yutao Zeng and Yunyao Mao and Banggu Wu and Qiyang Min and Xun Zhou},\n    journal = {ArXiv},\n    year    = {2024},\n    volume  = {abs/2409.19606},\n    url     = {https://api.semanticscholar.org/CorpusID:272987528}\n}\n```\n","project_url":"https://awesome.ecosyste.ms/api/v1/projects/github.com%2Flucidrains%2Falphafold3-pytorch","html_url":"https://awesome.ecosyste.ms/projects/github.com%2Flucidrains%2Falphafold3-pytorch","lists_url":"https://awesome.ecosyste.ms/api/v1/projects/github.com%2Flucidrains%2Falphafold3-pytorch/lists"}