Skip to content

Question: Why prompt is being run trough the network before generating new tokens? #719

Closed
@python273

Description

As I understand, the NN doesn't have a state, so you should be able to put whatever tokens into context and start generating new tokens immediately. But right now, it first runs the NN on the prompt it seems? So with a long prompt, it takes some time until it starts to generate new tokens.

I though I was missing something, but huggingface transformers starts to generate the tokens immediately.

Metadata

Assignees

No one assigned

    Labels

    No labels
    No labels

    Projects

    No projects

    Milestone

    No milestone

    Relationships

    None yet

    Development

    No branches or pull requests

    Issue actions