ceos/unsloth
1
0
Fork 0
Code Issues Pull requests Projects Releases Packages Wiki Activity Actions

Update rms_layernorm.py

Browse source
This commit is contained in:
Daniel Han 2024-11-04 23:08:05 -08:00
commit 15fdce80b9
1 changed files with 21 additions and 21 deletions
Download patch file Download diff file Expand all files Collapse all files

42
unsloth/kernels/rms_layernorm.py
View file

@ -149,27 +149,27 @@ class Fast_RMS_Layernorm(torch.autograd.Function):
Y = torch.empty((n_rows, n_cols), dtype = X.dtype, device = "cuda:0")
r = torch.empty(n_rows, dtype = torch.float32, device = "cuda:0")
if gemma == False:
_rms_layernorm_forward[(n_rows,)](
Y, Y.stride(0),
X, X.stride(0),
W, W.stride(0),
r, r.stride(0),
n_cols = int(n_cols),
eps = float(eps),
BLOCK_SIZE = BLOCK_SIZE,
num_warps = num_warps,
)
else:
_gemma_rms_layernorm_forward[(n_rows,)](
Y, Y.stride(0),
X, X.stride(0),
W, W.stride(0),
r, r.stride(0),
n_cols, eps,
BLOCK_SIZE = BLOCK_SIZE,
num_warps = num_warps,
)
# if gemma == False:
_rms_layernorm_forward[(n_rows,)](
Y, Y.stride(0),
X, X.stride(0),
W, W.stride(0),
r, r.stride(0),
n_cols = int(n_cols),
eps = float(eps),
BLOCK_SIZE = BLOCK_SIZE,
num_warps = num_warps,
)
# else:
# _gemma_rms_layernorm_forward[(n_rows,)](
# Y, Y.stride(0),
# X, X.stride(0),
# W, W.stride(0),
# r, r.stride(0),
# n_cols, eps,
# BLOCK_SIZE = BLOCK_SIZE,
# num_warps = num_warps,
# )
ctx.eps = eps
ctx.BLOCK_SIZE = BLOCK_SIZE
ctx.num_warps = num_warps