Multi-Modal Foundation · living-models
Botanic1-M
A 688M-parameter bidirectional state-space plant genomic foundation model for sequence modelling.
What Botanic1-M is
Botanic1-M is a genomic foundation model developed by living-models, designed specifically for research in plant biology and genomics.
The model provides sequence modelling capabilities tailored to plant genomes. By relying on state-space scans rather than standard transformer layers, it models genomic sequences without requiring attention mechanisms or positional embeddings.
How it works
Botanic1-M is built with 52 BiMamba2 blocks, containing 688 million parameters in total. The architecture performs bidirectional state-space scans to capture sequence context in both directions, omitting attention mechanisms and explicit positional embeddings.
The network is trained using a masked language modelling objective, where portions of the sequence are masked and predicted, supplemented by reverse-complement augmentation to account for the double-stranded nature of DNA.
Key facts
- Model architecture
- BiMamba2 (52 blocks)
- Parameter count
- 688M
- Training objective
- Masked language modelling with reverse-complement augmentation
- Core mechanism
- Bidirectional state-space scans without attention or positional embeddings
- Domain
- Plant genomics and biology
What you can use it for
- Plant Genomic Sequence Representation
- Extracting bidirectional sequence representations across plant genomes without the computational overhead of self-attention.
- Masked Nucleotide Prediction
- Predicting masked or missing nucleotides within plant DNA sequences using representations learnt via masked language modelling.
- Plant Genomics Foundation Research
- Serving as a pre-trained base model for downstream biological and genomic analysis in plant species.
Run it here
We list this model but cannot run it yet: gated on the Hub — never run past a gate the user did not accept.
Everything on this page still applies. Running it needs a loader for this repository, not a bigger machine.
Strengths and limits
Good at
- Employs bidirectional state-space scans across 52 blocks without relying on attention layers or positional embeddings.
- Integrates reverse-complement augmentation during training to maintain consistency across DNA strand orientations.
Watch out for
- Requires custom code execution to load and run the BiMamba2 architecture.
- Pre-training dataset composition, species diversity, and maximum context length are not detailed in the available documentation.
Inputs and outputs
You provide
Sequence or text
A protein or nucleotide sequence, or free text — whichever this model was trained on.
You get back
Mean-pooled representation, one vector per input.
Licence
Check the licence before commercial use · other
This model is gated by its authors — its licence has to be accepted on the source repository before it will run.
Questions
Can I use Botanic1-M commercially?
Check the licence before commercial use. This model is gated by its authors — its licence has to be accepted on the source repository before it will run.
What architecture is used by Botanic1-M?
Botanic1-M uses a 52-block BiMamba2 architecture that conducts bidirectional state-space scans without attention mechanisms or positional embeddings.
How was Botanic1-M trained?
It was trained using masked language modelling paired with reverse-complement augmentation.
How many parameters does the model contain?
The model contains 688 million parameters.
Written from Botanic1-M's own documentation (model card). Statements citing a figure that does not appear there are removed before publication. Licence and running details come from our registry.