Skip to content

Question about float precision #20

@Seeker98

Description

@Seeker98

As I understand, the A100 GPU performs significantly better with half-precision (FP16) computations compared to single-precision (FP32). I would like to know:

In your training process, did you use single precision, mixed precision, or half precision?

Have you experimented with different precision settings and their impact on training effectiveness? If so, what differences did you observe?

Thank you again for your work. I look forward to your response.

Metadata

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