Update rms_layernorm.py

This commit is contained in:
Daniel Han 2024-11-04 23:20:32 -08:00
commit 40858e4f76

View file

@ -171,8 +171,8 @@ class Fast_RMS_Layernorm(torch.autograd.Function):
num_warps = num_warps,
)
ctx.eps = eps
ctx.BLOCK_SIZE = BLOCK_SIZE
ctx.num_warps = num_warps
ctx.BLOCK_SIZE = 4096
ctx.num_warps = 16
ctx.GEMMA = gemma
ctx.save_for_backward(X, W, r)
return Y.view(*shape)