Published October 19, 2026 | Version 0.2.1
Model Open

NequIP & Allegro Foundation Potentials

Description

These are NequIP/Allegro models trained on large DFT datasets for inorganic materials, optimised for speed and accuracy. 
The `OAM` models are pre-trained on the `OMat24` dataset (~101M frames), and fine-tuned on the `sAlex` (~10.5M frames) and `MPtrj` (~1.5M frames) datasets, while the `OMat24` models are those trained on `OMat24` with no `sAlex`/`MPtrj` fine-tuning.
The `MP` models are trained only on the `MPtrj` (~1.5M frames) dataset (i.e. matbench-discovery 'compliant'), and so are only recommended for benchmarking and not production work.


See the preprint (10.48550/arXiv.2607.28461) and NequIP foundation potentials docs page for further details, along with `nequip.net` and `matbench-discovery`.

Version 0.2 (2026-09) added the `NequIP-MP-XL-0.2.nequip.zip` (MPtrj-only XL NequIP model) and training configuration yaml files for all models. Version 0.2.1 added one-line fixes to `NequIP-MP-L-0.1.nequip.zip`, `NequIP-OAM-XL-0.2.nequip.zip`, `NequIP-OMat24-S/M/L/XL-0.2.nequip.zip`, `Allegro-OAM-S-0.2.nequip.zip` and `Allegro-MP-L-0.2.nequip.zip` to ensure `openequivariance`/`cuEquivariance` compatibility (exact same models as `0.2`).
The `0.1a` version update was a manual fix to the packaged model modifiers to avoid issues with compiling/packaging models fine-tuned on these packaged foundation models (see https://github.com/mir-group/nequip/issues/572), and `0.1b` was to include the OMat24-only package files.

Files

Allegro-MP-L-0.2.nequip.zip

Files (1.4 GB)

Name Size
md5:f3fa3621a7a3db2e83940fa7a7c4a22b
153.6 MB Preview Download
md5:1daa1691d790ffd45fe66d935db692d5
9.5 kB Download
md5:b7b7059e5b80f425abbf26d09e891c01
105.4 MB Preview Download
md5:8e3ae234ea4962c955a5df7262ecc4aa
8.1 kB Download
md5:de8fce6b77b20e1c0ec716f400bcd064
11.5 kB Download
md5:03dadcd997929739aee3e214171d5b85
58.3 MB Preview Download
md5:b5234ba3b9bc0b292d4b8111564d27d8
8.1 kB Download
md5:66f4688c2cc08c0dd342c60651f04400
11.6 kB Download
md5:f7e4b301575cc46e0391b3515edbd30c
12.0 MB Preview Download
md5:e19604fac1774cbc9252df97f6bb5754
8.2 kB Download
md5:ae77029aaf06c78190ede5ee08d9f47d
11.5 kB Download
md5:e887bd75ba62b4aa503ab4e1ff96ffd1
105.4 MB Preview Download
md5:e89953e1e685b9e8f2fb32bf0b07cb2a
58.3 MB Preview Download
md5:c300077a444d7f9a9b12bd9f56c52c9f
12.0 MB Preview Download
md5:28ae4b0d7510475a5755644e63256f52
78.7 MB Preview Download
md5:0c531cd4f4435b5dc3573b4496301de5
149.7 MB Preview Download
md5:de317db4f4e9f26e134dc0f7de294277
8.1 kB Download
md5:35e12dfaca8e85eb9b351a2cd43c1560
8.6 kB Download
md5:6c09bee835381fed024b0aac772a7761
56.4 MB Preview Download
md5:e3e4167044e7e65926ecbc0d1ca1e7e5
7.9 kB Download
md5:4308a3d4e337f7c728911364bfa25594
10.7 kB Download
md5:49b6e7a6a0f8989cc307937612a06369
26.6 MB Preview Download
md5:ca591567f314af81a048e13d42df3654
7.9 kB Download
md5:76f5959eec88b64e53dadc6f6830a17b
11.7 kB Download
md5:8190d6e0a323e40cca098e537a554075
5.6 MB Preview Download
md5:93b568c0d361883c7fa8cd82fad46c63
8.0 kB Download
md5:1282055dbeb9d34db7c3220b8cdb9624
11.6 kB Download
md5:11edc4b7d9e612124520c48efd273728
259.6 MB Preview Download
md5:43fd4bab94c018dc727a27847ee11d0c
8.0 kB Download
md5:559fdb298f4fe014f02ae0deaa79aeea
11.5 kB Download
md5:35873b043b51eee31ebe4b144c317e65
7.5 kB Download
md5:16c3b241c8b14eab785ae253916a34e7
56.3 MB Preview Download
md5:25561f63023e1a2b0e8f12b6d15f3efa
26.5 MB Preview Download
md5:793ca311a47c4b6b42dbc75d20b7aaaa
5.7 MB Preview Download
md5:fc824357ad835c2f1ed4cc17c12329c6
259.6 MB Preview Download
md5:200e141855d6fed55cab79c3032f4c73
2.9 kB Download

Additional details

Related works

Is supplement to
Preprint: 10.48550/arXiv.2607.28461 (DOI)