NASA-IBM Lunar Foundation Model
NASA IMPACTA ViT-B encoder-decoder pretrained from scratch on SomBench, roughly two million co-registered lunar tile bundles spanning 11 modalities at two spatial scales, LROC NAC at about 1 m/px and LROC WAC at about 100 m/px. It adapts TerraMind's masked-token recipe to the Moon by tokenizing acquisition geometry as explicit context and training both resolution families in one mixed-batch loop, and downstream adaptation runs through TerraTorch with runnable configs for crater detection, Irregular Mare Patch segmentation and ice prospectivity.
GitHub's license endpoint reports NOASSERTION, but the repository LICENSE body is Apache-2.0 and both READMEs and the model card say the same. The companion repository states that pretraining code is not included, and the published SomBench dataset repo is a sample of the corpus rather than the corpus. Verified 2026-09-10 via the Hugging Face API, the model card, and the GitHub repository.
Openness
3 high confidence- weights
- open(ungated Apache-2.0 backbone checkpoint plus nine modality tokenizer checkpoints)
- data
- documented-not-released(SomBench composition given tile by tile
- code
- partial(TerraTorch fine-tuning integration and configs
- license
- Apache-2.0(OSI)
An unusually complete release for a government-lab model, and it still stops short of the top rungs. The corpus is described down to per-modality counts and split rules, but what ships on the Hub is a sample, and the fine-tuning repository says in one line that pretraining code is not included.
- https://huggingface.co/api/models/nasa-ibm-ai4science/NASA-IBM-Lunar-Foundation-Model recorded 2026-09-10
license:apache-2.0; gated: false; private: false; backbone/checkpoint.pt and nine tokenizers/*/checkpoint.pt present.
- https://raw.githubusercontent.com/NASA-IMPACT/NASA-IBM-Lunar-Foundation-Model/main/LICENSE recorded 2026-09-10
Apache License Version 2.0 body, contradicting the NOASSERTION the GitHub license endpoint reports.
- https://raw.githubusercontent.com/NASA-IMPACT/NASA-IBM-Lunar-Foundation-Model/main/README.md recorded 2026-09-10
"Pretraining code is not included." The two packages are the vendored model package and a TerraTorch integration of datamodules, tasks, backbone wrappers and runnable fine-tuning configs.
- https://huggingface.co/api/datasets/nasa-ibm-ai4science/Sombench-pretraining-data recorded 2026-09-10
"This includes a small sample from SomBench"; license cc-by-4.0; size_categories n<1K, against the 963,609 WAC and 1,000,113 NAC bundles the model card says pretraining used.
Adoption
1 high confidenceThree downloads in the trailing 30 days against a card whose cited creation timestamp puts it barely over a week old, so this is a first-week number rather than a settled one.
- https://huggingface.co/api/models/nasa-ibm-ai4science/NASA-IBM-Lunar-Foundation-Model recorded 2026-09-10
downloads: 3; likes: 7; createdAt 2026-09-01.
Capability
4 high confidenceThree of the four attributes are stated outright. The fourth is refused by the publisher rather than missing - the card calls any-to-any generation a qualitative probe and says the model is not a scientific-grade generative product, which is the category's own test for what a generated output has to be.
- https://huggingface.co/nasa-ibm-ai4science/NASA-IBM-Lunar-Foundation-Model/raw/main/README.md recorded 2026-09-10
11 modalities from LROC NAC, LROC WAC, SLDEM2015, Diviner, LOLA, Mini-RF, Kaguya, GRAIL and Lunar Prospector; two spatial scales at ~1 m/px and ~100 m/px in a single mixed-batch loop, "one set of weights covers both resolution families" across a 100x scale gap; downstream adaptation through TerraTorch with LoRA and full fine-tuning guidance; and "Not a scientific-grade generative product - generated fields are not calibrated predictions".
- https://raw.githubusercontent.com/NASA-IMPACT/NASA-IBM-Lunar-Foundation-Model/main/README.md recorded 2026-09-10
Every YAML under terratorch_integration/configs is a runnable `terratorch fit` target, grouped by downstream task.
Verified 2026-09-10