scgpt
Datasets
All datasets matching “scgpt”scgpt-replogle-activations
scGPT activation captures on Replogle (base vs ESM)
Per-cell layer-{5,7,10} hidden states from the two scGPT checkpoints in
matthewshu/scgpt-replogle-base-ft
and matthewshu/scgpt-replogle-esm-ft,
captured over the same 72,100-cell balanced sample of the
State-Replogle-Filtered
dataset. Sample is balanced over (cell_line, gene) buckets at 25 cells per
bucket. Intended for crosscoder diffing and other mechanistic-interpretability
analyses comparing the two models on identical inputs.… See the full description on the dataset page: https://huggingface.co/datasets/matthewshu/scgpt-replogle-activations.scgpt-norman-activationsscGPT_hidden_layer_embeddingsscgpt-activations-norman-calib-1k-f32scGPT_hidden_layer_embeddings_tokenizedscgpt-base-norman-activations