ESM (GPU)
Run large ESM models.
Results
Loading starting example…
Method & evidence
Use ESMC-6B or ESM3 embeddings, masked ESM3 generation, or ESMFold2 structure prediction.
Scope & limitations
- Large-model operations are limited to the released models and documented input bounds.
- Results preserve model and artifact identity for reproducibility.
Interpreting results
- Large-model outputs are predictions, not experimental observations.
- Seeded generation constrains repeatability but does not establish biological fitness.
- Inspect structure confidence and validate designs experimentally.
References
Input requirements & outputs
Input checks
- Large embeddings: at most 8 rows and 4,096 total residues
- Generation: at least one underscore mask and at most 512 positions
- Folding: one to four chains and at most 512 total residues
Expected results
- Embedding Dataset — Row-preserving vectors and checksum evidence.
- Generated sequence Dataset — Masked prompt, generated sequence, and exact digests.
- Structure Artifact — mmCIF coordinates with confidence and model evidence.
