Hello,
Following the question I had on running a GEMM only model, I have settled for a model that has Id Conv → Flatten → Gemm . I was wondering if the Flatten operation is just rearranging some pointer values or if it is doing a memory operation (like copying the tensor or something like that)?
Thank you very much,
+2Hi
Good question. Flatten is generally a "view" or "reshape" operation. to my knowledge. It changes how the tensor dimensions are interpreted. So the underlying data stays in the same order; only the logical shape changes.
Your pattern (Conv → Flatten → Gemm) is a supported configuration, so I don’t think you should see significant overhead from the Flatten itself, although I’d be very interested in seeing your results when you test this out.
Is there a specific performance concern you're tackling here?
Already have an account? Login
Enter your E-mail address. We'll send you an e-mail with instructions to reset your password.
