Catalog
Browse model families and checkpoints, then filter by group or available output: embeddings, likelihood scores, or predicted tracks.
Gated access: NucleotideTransformerV3 requires model access on Hugging Face and an authenticated hf auth login.
Evo2: hardware varies by checkpoint. The 7B port uses bfloat16, while the 1B, 20B, and 40B Transformer Engine ports require Hopper-class GPUs. The 20B and 40B checkpoints require multiple H100-class devices. Expand Evo2 below and open the selected version's checkpoint card before allocating hardware.
Hardware and input requirements vary by checkpoint. Check the adapter and checkpoint card for memory, context length, attention backend, and required sequence tracks before a large run.
35 families shown
Masked-language RNA model family with standard, CDS-adapted, and MARS pre-training variants.
AIDO.RNA-650MEmbeddingPseudo-likelihoodAIDO.RNA-650M-CDSdefaultEmbeddingPseudo-likelihoodAIDO.RNA-1.6BEmbeddingPseudo-likelihoodAIDO.RNA-1.6B-CDSEmbeddingPseudo-likelihoodAIDO.RNA-1M-MARSEmbeddingPseudo-likelihoodAIDO.RNA-25M-MARSEmbeddingPseudo-likelihoodAIDO.RNA-300M-MARSEmbeddingPseudo-likelihoodCodon-tokenized model trained on coding regions from mammalian, bacterial, and viral mRNAs.
CodonBERTdefaultEmbeddingPseudo-likelihoodRNA Transformer with an attention bias designed around base-pairing structure.
ERNIE-RNAdefaultEmbeddingPseudo-likelihoodERNIE-RNA-SSEmbeddingPseudo-likelihoodERNIE-RNA-MRLEmbeddingPseudo-likelihoodHybrid Mamba2 and Transformer model with coding-region-aware tokenization.
helix-mrnadefaultEmbeddingmRNA language model with masked-token and transcript-to-protein contrastive training.
mRNABERTdefaultEmbeddingPseudo-likelihoodCodon-tokenized model trained on coding regions and evaluated with a required CDS track.
mRNA-FMdefaultEmbeddingPseudo-likelihoodPlant RNA model trained with sequence reconstruction and structure-aware objectives.
omnigenome-52mEmbeddingPseudo-likelihoodomnigenome-186mdefaultEmbeddingPseudo-likelihoodMamba-based transcript model trained contrastively on functional and evolutionary relationships.
orthrus-base-4-trackEmbeddingorthrus-large-4-trackEmbeddingorthrus-large-6-trackdefaultEmbeddingPlant RNA model trained across species with sequence, structure, and region-annotation objectives.
plant_rnafmdefaultEmbeddingPseudo-likelihoodMasked RNA language model family with rotary positional encoding.
RiNALMo-microEmbeddingPseudo-likelihoodRiNALMo-megadefaultEmbeddingPseudo-likelihoodRiNALMo-gigaEmbeddingPseudo-likelihoodRNA Transformer trained with masked-token prediction and structural alignment.
RNABERTdefaultEmbeddingPseudo-likelihoodRNA language model using motif-level contiguous masking.
RNAErniedefaultEmbeddingPseudo-likelihoodRNAErnie2EmbeddingPseudo-likelihoodTransformer trained with masked-token prediction across a large non-coding RNA collection.
RNA-FMdefaultEmbeddingPseudo-likelihoodStructure-aware RNA model trained from multiple-sequence alignments.
RNA-MSMdefaultEmbeddingPseudo-likelihoodMasked RNA model trained on vertebrate transcripts with a focus on splice-site sequence.
SpliceBERT-1024ntdefaultEmbeddingPseudo-likelihoodSpliceBERT-510ntEmbeddingPseudo-likelihoodSpliceBERT-human-510ntEmbeddingPseudo-likelihood3' UTR language model family using overlapping k-mer tokenization.
UTRBERT-3merEmbeddingPseudo-likelihoodUTRBERT-4merEmbeddingPseudo-likelihoodUTRBERT-5merEmbeddingPseudo-likelihoodUTRBERT-6merdefaultEmbeddingPseudo-likelihood5' UTR model family trained on random and endogenous UTR sequences.
UTR-LM-MLMSIdefaultEmbeddingPseudo-likelihoodUTR-LM-MLMSISSEmbeddingPseudo-likelihoodUTR-LM-MLMEmbeddingPseudo-likelihoodUTR-LM-MLMSSEmbeddingPseudo-likelihoodBidirectional DNA language model with embedding and masked-token pseudo-likelihood interfaces.
AIDO.DNA-300MdefaultEmbeddingPseudo-likelihoodAIDO.DNA-7BEmbeddingPseudo-likelihoodLong-context sequence-to-function model that predicts genomic activity tracks.
alphagenomedefaultEmbeddingTrack predictionLong-window sequence-to-function model for RNA-seq and regulatory tracks, including Flashzoi variants.
borzoi-replicate-0EmbeddingTrack predictionborzoi-replicate-1EmbeddingTrack predictionborzoi-replicate-2EmbeddingTrack predictionborzoi-replicate-3EmbeddingTrack predictionflashzoi-replicate-0EmbeddingTrack predictionflashzoi-replicate-1EmbeddingTrack predictionflashzoi-replicate-2EmbeddingTrack predictionflashzoi-replicate-3EmbeddingTrack predictionborzoiEmbeddingTrack predictionflashzoidefaultEmbeddingTrack predictionAutoregressive genomic model trained on eukaryotic genes, mature mRNA, and bacterial genomes.
Carbon-500MEmbeddingCausal likelihoodCarbon-3BdefaultEmbeddingCausal likelihoodCarbon-8BEmbeddingCausal likelihoodOriginal masked DNA Transformer family with overlapping k-mer tokenization.
DNABERT-3merEmbeddingPseudo-likelihoodDNABERT-4merEmbeddingPseudo-likelihoodDNABERT-5merEmbeddingPseudo-likelihoodDNABERT-6merdefaultEmbeddingPseudo-likelihoodSpecies-aware DNA model trained contrastively across microbial genomes.
DNABERT-SdefaultEmbeddingMulti-species masked DNA model with byte-pair tokenization and rotary positions.
DNABERT2defaultEmbeddingPseudo-likelihoodLong-range sequence-to-function model for RNA-seq, ATAC-seq, and ChIP-seq tracks.
enformer-official-roughdefaultEmbeddingTrack predictionAutoregressive StripedHyena genomic model with short and long context variants.
Evo1-1.5-7B-8KdefaultEmbeddingCausal likelihoodEvo1-1-7B-8KEmbeddingCausal likelihoodEvo1-1-7B-131KEmbeddingCausal likelihoodAutoregressive StripedHyena2 family trained on OpenGenome2 with context lengths up to one million nucleotides.
Evo2-1B-8KEmbeddingCausal likelihoodEvo2-7B-8KdefaultEmbeddingCausal likelihoodEvo2-7B-262KEmbeddingCausal likelihoodEvo2-7B-1MEmbeddingCausal likelihoodEvo2-20B-1MEmbeddingCausal likelihoodEvo2-40B-8KEmbeddingCausal likelihoodEvo2-40B-1MEmbeddingCausal likelihoodBidirectional genomic model family with base and coding-sequence annotator variants.
prokaryote-0.5b-baseEmbeddingPseudo-likelihoodprokaryote-0.5b-cds-annotatorEmbeddingeukaryote-0.5b-basedefaultEmbeddingPseudo-likelihoodeukaryote-1.2b-cds-annotator-previewEmbeddingAutoregressive gene-region model family with eukaryotic and prokaryotic checkpoints.
eukaryote-1.2b-baseEmbeddingCausal likelihoodv2-eukaryote-1.2b-baseEmbeddingCausal likelihoodv2-prokaryote-1.2b-baseEmbeddingCausal likelihoodeukaryote-3b-baseEmbeddingCausal likelihoodv2-eukaryote-3b-basedefaultEmbeddingCausal likelihoodv2-prokaryote-3b-baseEmbeddingCausal likelihoodLong-context genomic language model with checkpoints spanning 16 thousand to one million nucleotides.
hyenadna-large-1m-seqlen-hfEmbeddingCausal likelihoodhyenadna-medium-450k-seqlen-hfdefaultEmbeddingCausal likelihoodhyenadna-medium-160k-seqlen-hfEmbeddingCausal likelihoodhyenadna-small-32k-seqlen-hfEmbeddingCausal likelihoodhyenadna-tiny-16k-seqlen-d128-hfEmbeddingCausal likelihoodMasked DNA Transformer family spanning human-reference and multi-species pre-training sets.
2.5b-multi-speciesdefaultEmbeddingPseudo-likelihood2.5b-1000gEmbeddingPseudo-likelihood500m-human-refEmbeddingPseudo-likelihood500m-1000gEmbeddingPseudo-likelihoodv2-50m-multi-speciesEmbeddingPseudo-likelihoodv2-100m-multi-speciesEmbeddingPseudo-likelihoodv2-250m-multi-speciesEmbeddingPseudo-likelihoodv2-500m-multi-speciesEmbeddingPseudo-likelihoodSingle-nucleotide, long-sequence model with checkpoints from sequence pretraining and genomic-track post-training.
v3_8M_preEmbeddingPseudo-likelihoodv3_100M_preEmbeddingPseudo-likelihoodv3_650M_preEmbeddingPseudo-likelihoodv3_100M_postEmbeddingTrack predictionv3_650M_postdefaultEmbeddingTrack predictionFixed k-mer count, GC-content, and sequence-statistic features.
naive-4-trackdefaultEmbeddingNaive sequence features augmented with coding-region length and exon count.
naive-6-trackdefaultEmbeddingRandomly initialized six-track Mamba model used as an untrained reference.
naive-mambadefaultEmbedding