Skip to content

Make the context_size configurable when running encode_tokenized_cells #10

Description

@nleroy917

To change the models contex size, it needs to be set "globally":

model = AtacformerForCellClustering.from_pretrained("databio/atacformer-base-hg38")
model.max_position_embeddings = 1024 # set globally for the model here
model.encode_tokenized_cells(...)

This can be confusing, so maybe we introduce a new parameter to that function that will override whatever the model is set to:

model = AtacformerForCellClustering.from_pretrained("databio/atacformer-base-hg38")
model.encode_tokenized_cells(..., max_tokens_per_cell=1024)

The function signature for encode_tokenized_cells would now become:

def encode_tokenized_cells(
    self,
    input_ids: List[List[int]],
    batch_size: int = 16,
    max_tokens_per_cell: int = None
) -> torch.Tensor:
    max_ctx = max_tokens_per_cell or self.config.max_position_embeddings

Activity

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Metadata

Metadata

Assignees

No one assigned

    Type

    No type

    Projects

    No projects

      Milestone

      No milestone

      Relationships

      None yet

      Development

      No branches or pull requests

      Issue actions