Model Type | biological foundation model, long-context modeling, sequence modeling |
|
Additional Notes | Weights of 15 intermediate pretraining checkpoints for phase 1 and 2 are released on HuggingFace repository branches. |
|
Training Details |
Data Sources: | OpenGenome, a prokaryotic whole-genome dataset |
|
Data Volume: | |
Context Length: | |
Model Architecture: | StripedHyena, hybrid architecture with multi-head attention and gated convolutions arranged in Hyena blocks |
|
|
Input Output |
Performance Tips: | Keep 'poles' and 'residues' in 'float32' precision for longer prompts or training. |
|
|
Release Notes |
Version: | |
Notes: | Fixed wrong permutation of some projections affecting generation quality. |
|
|
|