Skip to content

CLI improvement to TensorParallelism/02_matmul_tp/matmul_tp_big.cu #17

Description

@dkennetzoracle

The code could benefit from the following changes:

A CLI with the following abstractions:

#define M 4096  // Rows of A, Rows of C
#define N 8192  // Columns of B, Columns of C
#define K 1024  // Columns of A, Rows of B
#define NGPUS 2 // Number of GPUs to use for computation
  • variable sizes of M, N, K
  • number of GPUs to use
  • error checking for matrix dimensions from CLI values
  • device detection to determine if number of GPUs selected is supported (IE 4 is picked, but you only have 2)
  • boolean of whether to run host matmul or not, and if not - don't assert correctness or measure cpu timing. Should be pulled into a function.

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

    Labels

    No labels
    No labels

    Projects

    No projects

      Milestone

      No milestone

      Relationships

      None yet

      Development

      No branches or pull requests

      Issue actions