Commit cd089597 authored by Tri Dao's avatar Tri Dao
Browse files

[LayerNorm] Implement dropout in fused residual + LN/RMSNorm

parent 713bd3aa
# Copyright (c) 2023, Tri Dao. # Copyright (c) 2023, Tri Dao.
# Implement residual + layer_norm / rms_norm. # Implement dropout + residual + layer_norm / rms_norm.
# Based on the Triton LayerNorm tutorial: https://triton-lang.org/main/getting-started/tutorials/05-layer-norm.html # Based on the Triton LayerNorm tutorial: https://triton-lang.org/main/getting-started/tutorials/05-layer-norm.html
# For the backward pass, we keep weight_grad and bias_grad in registers and accumulate. # For the backward pass, we keep weight_grad and bias_grad in registers and accumulate.
...@@ -16,7 +16,17 @@ import triton ...@@ -16,7 +16,17 @@ import triton
import triton.language as tl import triton.language as tl
def layer_norm_ref(x, weight, bias, residual=None, eps=1e-6, prenorm=False, upcast=False): def layer_norm_ref(
x,
weight,
bias,
residual=None,
eps=1e-6,
dropout_p=0.0,
prenorm=False,
dropout_mask=None,
upcast=False,
):
dtype = x.dtype dtype = x.dtype
if upcast: if upcast:
weight = weight.float() weight = weight.float()
...@@ -24,6 +34,11 @@ def layer_norm_ref(x, weight, bias, residual=None, eps=1e-6, prenorm=False, upca ...@@ -24,6 +34,11 @@ def layer_norm_ref(x, weight, bias, residual=None, eps=1e-6, prenorm=False, upca
if upcast: if upcast:
x = x.float() x = x.float()
residual = residual.float() if residual is not None else residual residual = residual.float() if residual is not None else residual
if dropout_p > 0.0:
if dropout_mask is not None:
x = x.masked_fill(~dropout_mask, 0.0) / (1.0 - dropout_p)
else:
x = F.dropout(x, p=dropout_p)
if residual is not None: if residual is not None:
x = (x + residual).to(x.dtype) x = (x + residual).to(x.dtype)
out = F.layer_norm(x.to(weight.dtype), x.shape[-1:], weight=weight, bias=bias, eps=eps).to( out = F.layer_norm(x.to(weight.dtype), x.shape[-1:], weight=weight, bias=bias, eps=eps).to(
...@@ -32,7 +47,17 @@ def layer_norm_ref(x, weight, bias, residual=None, eps=1e-6, prenorm=False, upca ...@@ -32,7 +47,17 @@ def layer_norm_ref(x, weight, bias, residual=None, eps=1e-6, prenorm=False, upca
return out if not prenorm else (out, x) return out if not prenorm else (out, x)
def rms_norm_ref(x, weight, bias, residual=None, eps=1e-6, prenorm=False, upcast=False): def rms_norm_ref(
x,
weight,
bias,
residual=None,
eps=1e-6,
dropout_p=0.0,
prenorm=False,
dropout_mask=None,
upcast=False,
):
dtype = x.dtype dtype = x.dtype
if upcast: if upcast:
weight = weight.float() weight = weight.float()
...@@ -40,6 +65,11 @@ def rms_norm_ref(x, weight, bias, residual=None, eps=1e-6, prenorm=False, upcast ...@@ -40,6 +65,11 @@ def rms_norm_ref(x, weight, bias, residual=None, eps=1e-6, prenorm=False, upcast
if upcast: if upcast:
x = x.float() x = x.float()
residual = residual.float() if residual is not None else residual residual = residual.float() if residual is not None else residual
if dropout_p > 0.0:
if dropout_mask is not None:
x = x.masked_fill(~dropout_mask, 0.0) / (1.0 - dropout_p)
else:
x = F.dropout(x, p=dropout_p)
if residual is not None: if residual is not None:
x = (x + residual).to(x.dtype) x = (x + residual).to(x.dtype)
rstd = 1 / torch.sqrt((x.square()).mean(dim=-1, keepdim=True) + eps) rstd = 1 / torch.sqrt((x.square()).mean(dim=-1, keepdim=True) + eps)
...@@ -69,6 +99,8 @@ def _layer_norm_fwd_1pass_kernel( ...@@ -69,6 +99,8 @@ def _layer_norm_fwd_1pass_kernel(
B, # pointer to the biases B, # pointer to the biases
RESIDUAL, # pointer to the residual RESIDUAL, # pointer to the residual
RESIDUAL_OUT, # pointer to the residual RESIDUAL_OUT, # pointer to the residual
SEEDS, # Dropout seeds for each row
DROPOUT_MASK,
Mean, # pointer to the mean Mean, # pointer to the mean
Rstd, # pointer to the 1/std Rstd, # pointer to the 1/std
stride_x_row, # how much to increase the pointer when moving by 1 row stride_x_row, # how much to increase the pointer when moving by 1 row
...@@ -77,11 +109,14 @@ def _layer_norm_fwd_1pass_kernel( ...@@ -77,11 +109,14 @@ def _layer_norm_fwd_1pass_kernel(
stride_res_out_row, stride_res_out_row,
N, # number of columns in X N, # number of columns in X
eps, # epsilon to avoid division by zero eps, # epsilon to avoid division by zero
dropout_p, # Dropout probability
IS_RMS_NORM: tl.constexpr, IS_RMS_NORM: tl.constexpr,
BLOCK_N: tl.constexpr, BLOCK_N: tl.constexpr,
HAS_RESIDUAL: tl.constexpr, HAS_RESIDUAL: tl.constexpr,
STORE_RESIDUAL_OUT: tl.constexpr, STORE_RESIDUAL_OUT: tl.constexpr,
HAS_BIAS: tl.constexpr, HAS_BIAS: tl.constexpr,
HAS_DROPOUT: tl.constexpr,
STORE_DROPOUT_MASK: tl.constexpr,
): ):
# Map the program id to the row of X and Y it should compute. # Map the program id to the row of X and Y it should compute.
row = tl.program_id(0) row = tl.program_id(0)
...@@ -94,6 +129,13 @@ def _layer_norm_fwd_1pass_kernel( ...@@ -94,6 +129,13 @@ def _layer_norm_fwd_1pass_kernel(
# Compute mean and variance # Compute mean and variance
cols = tl.arange(0, BLOCK_N) cols = tl.arange(0, BLOCK_N)
x = tl.load(X + cols, mask=cols < N, other=0.0).to(tl.float32) x = tl.load(X + cols, mask=cols < N, other=0.0).to(tl.float32)
if HAS_DROPOUT:
# Compute dropout mask
# 7 rounds is good enough, and reduces register pressure
keep_mask = tl.rand(tl.load(SEEDS + row).to(tl.uint32), cols, n_rounds=7) > dropout_p
x = tl.where(keep_mask, x / (1.0 - dropout_p), 0.0)
if STORE_DROPOUT_MASK:
tl.store(DROPOUT_MASK + row * N + cols, keep_mask, mask=cols < N)
if HAS_RESIDUAL: if HAS_RESIDUAL:
residual = tl.load(RESIDUAL + cols, mask=cols < N, other=0.0).to(tl.float32) residual = tl.load(RESIDUAL + cols, mask=cols < N, other=0.0).to(tl.float32)
x += residual x += residual
...@@ -121,7 +163,16 @@ def _layer_norm_fwd_1pass_kernel( ...@@ -121,7 +163,16 @@ def _layer_norm_fwd_1pass_kernel(
def _layer_norm_fwd( def _layer_norm_fwd(
x, weight, bias, eps, residual=None, out_dtype=None, residual_dtype=None, is_rms_norm=False x,
weight,
bias,
eps,
residual=None,
dropout_p=0.0,
out_dtype=None,
residual_dtype=None,
is_rms_norm=False,
return_dropout_mask=False,
): ):
if residual is not None: if residual is not None:
residual_dtype = residual.dtype residual_dtype = residual.dtype
...@@ -138,13 +189,27 @@ def _layer_norm_fwd( ...@@ -138,13 +189,27 @@ def _layer_norm_fwd(
# allocate output # allocate output
y = torch.empty_like(x, dtype=x.dtype if out_dtype is None else out_dtype) y = torch.empty_like(x, dtype=x.dtype if out_dtype is None else out_dtype)
assert y.stride(-1) == 1 assert y.stride(-1) == 1
if residual is not None or (residual_dtype is not None and residual_dtype != x.dtype): if (
residual_out = torch.empty(M, N, device=x.device, dtype=residual_dtype) residual is not None
or (residual_dtype is not None and residual_dtype != x.dtype)
or dropout_p > 0.0
):
residual_out = torch.empty(
M, N, device=x.device, dtype=residual_dtype if residual_dtype is not None else x.dtype
)
assert residual_out.stride(-1) == 1 assert residual_out.stride(-1) == 1
else: else:
residual_out = None residual_out = None
mean = torch.empty((M,), dtype=torch.float32, device="cuda") if not is_rms_norm else None mean = torch.empty((M,), dtype=torch.float32, device="cuda") if not is_rms_norm else None
rstd = torch.empty((M,), dtype=torch.float32, device="cuda") rstd = torch.empty((M,), dtype=torch.float32, device="cuda")
if dropout_p > 0.0:
seeds = torch.randint(2**32, (M,), device=x.device, dtype=torch.int64)
else:
seeds = None
if return_dropout_mask and dropout_p > 0.0:
dropout_mask = torch.empty_like(x, dtype=torch.bool)
else:
dropout_mask = None
# Less than 64KB per feature: enqueue fused kernel # Less than 64KB per feature: enqueue fused kernel
MAX_FUSED_SIZE = 65536 // x.element_size() MAX_FUSED_SIZE = 65536 // x.element_size()
BLOCK_N = min(MAX_FUSED_SIZE, triton.next_power_of_2(N)) BLOCK_N = min(MAX_FUSED_SIZE, triton.next_power_of_2(N))
...@@ -159,6 +224,8 @@ def _layer_norm_fwd( ...@@ -159,6 +224,8 @@ def _layer_norm_fwd(
bias, bias,
residual, residual,
residual_out, residual_out,
seeds,
dropout_mask,
mean, mean,
rstd, rstd,
x.stride(0), x.stride(0),
...@@ -167,14 +234,17 @@ def _layer_norm_fwd( ...@@ -167,14 +234,17 @@ def _layer_norm_fwd(
residual_out.stride(0) if residual_out is not None else 0, residual_out.stride(0) if residual_out is not None else 0,
N, N,
eps, eps,
dropout_p,
is_rms_norm, is_rms_norm,
BLOCK_N, BLOCK_N,
residual is not None, residual is not None,
residual_out is not None, residual_out is not None,
bias is not None, bias is not None,
dropout_p > 0.0,
dropout_mask is not None,
) )
# residual_out is None if residual is None and residual_dtype == input_dtype # residual_out is None if residual is None and residual_dtype == input_dtype and dropout_p == 0.0
return y, mean, rstd, residual_out if residual_out is not None else x return y, mean, rstd, residual_out if residual_out is not None else x, seeds, dropout_mask
@triton.autotune( @triton.autotune(
...@@ -186,7 +256,7 @@ def _layer_norm_fwd( ...@@ -186,7 +256,7 @@ def _layer_norm_fwd(
triton.Config({}, num_warps=16), triton.Config({}, num_warps=16),
triton.Config({}, num_warps=32), triton.Config({}, num_warps=32),
], ],
key=["N", "HAS_DRESIDUAL", "STORE_DRESIDUAL", "IS_RMS_NORM", "HAS_BIAS"], key=["N", "HAS_DRESIDUAL", "STORE_DRESIDUAL", "IS_RMS_NORM", "HAS_BIAS", "HAS_DROPOUT"],
) )
# @triton.heuristics({"HAS_BIAS": lambda args: args["B"] is not None}) # @triton.heuristics({"HAS_BIAS": lambda args: args["B"] is not None})
# @triton.heuristics({"HAS_DRESIDUAL": lambda args: args["DRESIDUAL"] is not None}) # @triton.heuristics({"HAS_DRESIDUAL": lambda args: args["DRESIDUAL"] is not None})
...@@ -204,6 +274,7 @@ def _layer_norm_bwd_kernel( ...@@ -204,6 +274,7 @@ def _layer_norm_bwd_kernel(
DB, # pointer to the partial sum of biases gradient DB, # pointer to the partial sum of biases gradient
DRESIDUAL, DRESIDUAL,
DRESIDUAL_IN, DRESIDUAL_IN,
SEEDS,
Mean, # pointer to the mean Mean, # pointer to the mean
Rstd, # pointer to the 1/std Rstd, # pointer to the 1/std
stride_x_row, # how much to increase the pointer when moving by 1 row stride_x_row, # how much to increase the pointer when moving by 1 row
...@@ -215,12 +286,14 @@ def _layer_norm_bwd_kernel( ...@@ -215,12 +286,14 @@ def _layer_norm_bwd_kernel(
M, # number of rows in X M, # number of rows in X
N, # number of columns in X N, # number of columns in X
eps, # epsilon to avoid division by zero eps, # epsilon to avoid division by zero
dropout_p,
rows_per_program, rows_per_program,
IS_RMS_NORM: tl.constexpr, IS_RMS_NORM: tl.constexpr,
BLOCK_N: tl.constexpr, BLOCK_N: tl.constexpr,
HAS_DRESIDUAL: tl.constexpr, HAS_DRESIDUAL: tl.constexpr,
STORE_DRESIDUAL: tl.constexpr, STORE_DRESIDUAL: tl.constexpr,
HAS_BIAS: tl.constexpr, HAS_BIAS: tl.constexpr,
HAS_DROPOUT: tl.constexpr,
RECOMPUTE_OUTPUT: tl.constexpr, RECOMPUTE_OUTPUT: tl.constexpr,
): ):
# Map the program id to the elements of X, DX, and DY it should compute. # Map the program id to the elements of X, DX, and DY it should compute.
...@@ -274,6 +347,9 @@ def _layer_norm_bwd_kernel( ...@@ -274,6 +347,9 @@ def _layer_norm_bwd_kernel(
# Write dx # Write dx
if STORE_DRESIDUAL: if STORE_DRESIDUAL:
tl.store(DRESIDUAL_IN + cols, dx, mask=mask) tl.store(DRESIDUAL_IN + cols, dx, mask=mask)
if HAS_DROPOUT:
keep_mask = tl.rand(tl.load(SEEDS + row).to(tl.uint32), cols, n_rounds=7) > dropout_p
dx = tl.where(keep_mask, dx / (1.0 - dropout_p), 0.0)
tl.store(DX + cols, dx, mask=mask) tl.store(DX + cols, dx, mask=mask)
X += stride_x_row X += stride_x_row
...@@ -299,6 +375,8 @@ def _layer_norm_bwd( ...@@ -299,6 +375,8 @@ def _layer_norm_bwd(
mean, mean,
rstd, rstd,
dresidual=None, dresidual=None,
seeds=None,
dropout_p=0.0,
has_residual=False, has_residual=False,
is_rms_norm=False, is_rms_norm=False,
x_dtype=None, x_dtype=None,
...@@ -316,13 +394,18 @@ def _layer_norm_bwd( ...@@ -316,13 +394,18 @@ def _layer_norm_bwd(
if bias is not None: if bias is not None:
assert bias.stride(-1) == 1 assert bias.stride(-1) == 1
assert bias.shape == (N,) assert bias.shape == (N,)
if seeds is not None:
assert seeds.is_contiguous()
assert seeds.shape == (M,)
# allocate output # allocate output
dx = ( dx = (
torch.empty_like(x) torch.empty_like(x)
if x_dtype is None if x_dtype is None
else torch.empty(M, N, dtype=x_dtype, device=x.device) else torch.empty(M, N, dtype=x_dtype, device=x.device)
) )
dresidual_in = torch.empty_like(x) if has_residual and dx.dtype != x.dtype else None dresidual_in = (
torch.empty_like(x) if has_residual and (dx.dtype != x.dtype or dropout_p > 0.0) else None
)
y = torch.empty(M, N, dtype=dy.dtype, device=dy.device) if recompute_output else None y = torch.empty(M, N, dtype=dy.dtype, device=dy.device) if recompute_output else None
# Less than 64KB per feature: enqueue fused kernel # Less than 64KB per feature: enqueue fused kernel
...@@ -351,6 +434,7 @@ def _layer_norm_bwd( ...@@ -351,6 +434,7 @@ def _layer_norm_bwd(
_db, _db,
dresidual, dresidual,
dresidual_in, dresidual_in,
seeds,
mean, mean,
rstd, rstd,
x.stride(0), x.stride(0),
...@@ -362,17 +446,19 @@ def _layer_norm_bwd( ...@@ -362,17 +446,19 @@ def _layer_norm_bwd(
M, M,
N, N,
eps, eps,
dropout_p,
rows_per_program, rows_per_program,
is_rms_norm, is_rms_norm,
BLOCK_N, BLOCK_N,
dresidual is not None, dresidual is not None,
dresidual_in is not None, dresidual_in is not None,
bias is not None, bias is not None,
dropout_p > 0.0,
) )
dw = _dw.sum(0).to(weight.dtype) dw = _dw.sum(0).to(weight.dtype)
db = _db.sum(0).to(bias.dtype) if bias is not None else None db = _db.sum(0).to(bias.dtype) if bias is not None else None
# Don't need to compute dresidual_in separately in this case # Don't need to compute dresidual_in separately in this case
if has_residual and dx.dtype == x.dtype: if has_residual and dx.dtype == x.dtype and dropout_p == 0.0:
dresidual_in = dx dresidual_in = dx
return (dx, dw, db, dresidual_in) if not recompute_output else (dx, dw, db, dresidual_in, y) return (dx, dw, db, dresidual_in) if not recompute_output else (dx, dw, db, dresidual_in, y)
...@@ -386,9 +472,11 @@ class LayerNormFn(torch.autograd.Function): ...@@ -386,9 +472,11 @@ class LayerNormFn(torch.autograd.Function):
bias, bias,
residual=None, residual=None,
eps=1e-6, eps=1e-6,
dropout_p=0.0,
prenorm=False, prenorm=False,
residual_in_fp32=False, residual_in_fp32=False,
is_rms_norm=False, is_rms_norm=False,
return_dropout_mask=False,
): ):
x_shape_og = x.shape x_shape_og = x.shape
# reshape input data into 2D tensor # reshape input data into 2D tensor
...@@ -408,22 +496,36 @@ class LayerNormFn(torch.autograd.Function): ...@@ -408,22 +496,36 @@ class LayerNormFn(torch.autograd.Function):
if residual is not None if residual is not None
else (torch.float32 if residual_in_fp32 else None) else (torch.float32 if residual_in_fp32 else None)
) )
y, mean, rstd, residual_out = _layer_norm_fwd( y, mean, rstd, residual_out, seeds, dropout_mask = _layer_norm_fwd(
x, weight, bias, eps, residual, residual_dtype=residual_dtype, is_rms_norm=is_rms_norm x,
weight,
bias,
eps,
residual,
dropout_p=dropout_p,
residual_dtype=residual_dtype,
is_rms_norm=is_rms_norm,
return_dropout_mask=return_dropout_mask,
) )
ctx.save_for_backward(residual_out, weight, bias, mean, rstd) ctx.save_for_backward(residual_out, weight, bias, seeds, mean, rstd)
ctx.x_shape_og = x_shape_og ctx.x_shape_og = x_shape_og
ctx.eps = eps ctx.eps = eps
ctx.dropout_p = dropout_p
ctx.is_rms_norm = is_rms_norm ctx.is_rms_norm = is_rms_norm
ctx.has_residual = residual is not None ctx.has_residual = residual is not None
ctx.prenorm = prenorm ctx.prenorm = prenorm
ctx.x_dtype = x.dtype ctx.x_dtype = x.dtype
y = y.reshape(x_shape_og) y = y.reshape(x_shape_og)
return y if not prenorm else (y, residual_out.reshape(x_shape_og)) residual_out = residual_out.reshape(x_shape_og) if residual_out is not None else None
dropout_mask = dropout_mask.reshape(x_shape_og) if dropout_mask is not None else None
if not return_dropout_mask:
return y if not prenorm else (y, residual_out)
else:
return (y, dropout_mask) if not prenorm else (y, residual_out, dropout_mask)
@staticmethod @staticmethod
def backward(ctx, dy, *args): def backward(ctx, dy, *args):
x, weight, bias, mean, rstd = ctx.saved_tensors x, weight, bias, seeds, mean, rstd = ctx.saved_tensors
dy = dy.reshape(-1, dy.shape[-1]) dy = dy.reshape(-1, dy.shape[-1])
if dy.stride(-1) != 1: if dy.stride(-1) != 1:
dy = dy.contiguous() dy = dy.contiguous()
...@@ -445,6 +547,8 @@ class LayerNormFn(torch.autograd.Function): ...@@ -445,6 +547,8 @@ class LayerNormFn(torch.autograd.Function):
mean, mean,
rstd, rstd,
dresidual, dresidual,
seeds,
ctx.dropout_p,
ctx.has_residual, ctx.has_residual,
ctx.is_rms_norm, ctx.is_rms_norm,
x_dtype=ctx.x_dtype, x_dtype=ctx.x_dtype,
...@@ -458,6 +562,8 @@ class LayerNormFn(torch.autograd.Function): ...@@ -458,6 +562,8 @@ class LayerNormFn(torch.autograd.Function):
None, None,
None, None,
None, None,
None,
None,
) )
...@@ -467,22 +573,57 @@ def layer_norm_fn( ...@@ -467,22 +573,57 @@ def layer_norm_fn(
bias, bias,
residual=None, residual=None,
eps=1e-6, eps=1e-6,
dropout_p=0.0,
prenorm=False, prenorm=False,
residual_in_fp32=False, residual_in_fp32=False,
is_rms_norm=False, is_rms_norm=False,
return_dropout_mask=False,
): ):
return LayerNormFn.apply(x, weight, bias, residual, eps, prenorm, residual_in_fp32, is_rms_norm) return LayerNormFn.apply(
x,
weight,
bias,
residual,
eps,
dropout_p,
prenorm,
residual_in_fp32,
is_rms_norm,
return_dropout_mask,
)
def rms_norm_fn(x, weight, bias, residual=None, prenorm=False, residual_in_fp32=False, eps=1e-6): def rms_norm_fn(
return LayerNormFn.apply(x, weight, bias, residual, eps, prenorm, residual_in_fp32, True) x,
weight,
bias,
residual=None,
eps=1e-6,
dropout_p=0.0,
prenorm=False,
residual_in_fp32=False,
return_dropout_mask=False,
):
return LayerNormFn.apply(
x,
weight,
bias,
residual,
eps,
dropout_p,
prenorm,
residual_in_fp32,
True,
return_dropout_mask,
)
class RMSNorm(torch.nn.Module): class RMSNorm(torch.nn.Module):
def __init__(self, hidden_size, eps=1e-5, device=None, dtype=None): def __init__(self, hidden_size, eps=1e-5, dropout_p=0.0, device=None, dtype=None):
factory_kwargs = {"device": device, "dtype": dtype} factory_kwargs = {"device": device, "dtype": dtype}
super().__init__() super().__init__()
self.eps = eps self.eps = eps
self.dropout_p = dropout_p
self.weight = torch.nn.Parameter(torch.empty(hidden_size, **factory_kwargs)) self.weight = torch.nn.Parameter(torch.empty(hidden_size, **factory_kwargs))
self.register_parameter("bias", None) self.register_parameter("bias", None)
self.reset_parameters() self.reset_parameters()
...@@ -497,9 +638,9 @@ class RMSNorm(torch.nn.Module): ...@@ -497,9 +638,9 @@ class RMSNorm(torch.nn.Module):
self.bias, self.bias,
residual=residual, residual=residual,
eps=self.eps, eps=self.eps,
dropout_p=self.dropout_p if self.training else 0.0,
prenorm=prenorm, prenorm=prenorm,
residual_in_fp32=residual_in_fp32, residual_in_fp32=residual_in_fp32,
is_rms_norm=True,
) )
......
...@@ -16,12 +16,14 @@ from flash_attn.ops.triton.layernorm import ( ...@@ -16,12 +16,14 @@ from flash_attn.ops.triton.layernorm import (
is_sm8x = torch.cuda.get_device_capability("cuda")[0] >= 8 is_sm8x = torch.cuda.get_device_capability("cuda")[0] >= 8
@pytest.mark.parametrize("dropout_p", [0.0, 0.27])
# @pytest.mark.parametrize("dropout_p", [0.27])
@pytest.mark.parametrize("prenorm", [True, False]) @pytest.mark.parametrize("prenorm", [True, False])
# @pytest.mark.parametrize("prenorm", [True]) # @pytest.mark.parametrize("prenorm", [False])
@pytest.mark.parametrize("is_rms_norm", [False, True]) @pytest.mark.parametrize("is_rms_norm", [False, True])
# @pytest.mark.parametrize("is_rms_norm", [True]) # @pytest.mark.parametrize("is_rms_norm", [True])
@pytest.mark.parametrize("has_residual", [True, False]) @pytest.mark.parametrize("has_residual", [True, False])
# @pytest.mark.parametrize("has_residual", [False]) # @pytest.mark.parametrize("has_residual", [True])
@pytest.mark.parametrize( @pytest.mark.parametrize(
"weight_dtype", [torch.float32, torch.float16] + ([torch.bfloat16] if is_sm8x else []) "weight_dtype", [torch.float32, torch.float16] + ([torch.bfloat16] if is_sm8x else [])
) )
...@@ -31,11 +33,18 @@ is_sm8x = torch.cuda.get_device_capability("cuda")[0] >= 8 ...@@ -31,11 +33,18 @@ is_sm8x = torch.cuda.get_device_capability("cuda")[0] >= 8
[(torch.float16, torch.float16), (torch.float16, torch.float32), (torch.float32, torch.float32)] [(torch.float16, torch.float16), (torch.float16, torch.float32), (torch.float32, torch.float32)]
+ ([(torch.bfloat16, torch.bfloat16), (torch.bfloat16, torch.float32)] if is_sm8x else []), + ([(torch.bfloat16, torch.bfloat16), (torch.bfloat16, torch.float32)] if is_sm8x else []),
) )
# @pytest.mark.parametrize("input_dtype,residual_dtype", [(torch.bfloat16, torch.float32)]) # @pytest.mark.parametrize("input_dtype,residual_dtype", [(torch.float16, torch.float16)])
@pytest.mark.parametrize("hidden_size", [192, 2048, 2560, 3000, 8192]) @pytest.mark.parametrize("hidden_size", [192, 2048, 2560, 3000, 4096])
# @pytest.mark.parametrize("hidden_size", [256]) # @pytest.mark.parametrize("hidden_size", [256])
def test_layer_norm( def test_layer_norm(
hidden_size, input_dtype, residual_dtype, weight_dtype, has_residual, is_rms_norm, prenorm hidden_size,
input_dtype,
residual_dtype,
weight_dtype,
has_residual,
is_rms_norm,
prenorm,
dropout_p,
): ):
device = "cuda" device = "cuda"
if any(x == torch.bfloat16 for x in [input_dtype, residual_dtype, weight_dtype]): if any(x == torch.bfloat16 for x in [input_dtype, residual_dtype, weight_dtype]):
...@@ -48,8 +57,6 @@ def test_layer_norm( ...@@ -48,8 +57,6 @@ def test_layer_norm(
torch.random.manual_seed(0) torch.random.manual_seed(0)
batch_size = 8 batch_size = 8
seqlen = 512 seqlen = 512
# batch_size = 1
# seqlen = 1
layer_norm_ref_fn = layer_norm_ref if not is_rms_norm else rms_norm_ref layer_norm_ref_fn = layer_norm_ref if not is_rms_norm else rms_norm_ref
allclose = ( allclose = (
lambda x, x_pt, x_ref, atol=atol: (x - x_ref).abs().max() lambda x, x_pt, x_ref, atol=atol: (x - x_ref).abs().max()
...@@ -83,25 +90,46 @@ def test_layer_norm( ...@@ -83,25 +90,46 @@ def test_layer_norm(
bias, bias,
residual=res, residual=res,
eps=1e-6, eps=1e-6,
dropout_p=dropout_p,
prenorm=prenorm, prenorm=prenorm,
residual_in_fp32=residual_in_fp32, residual_in_fp32=residual_in_fp32,
is_rms_norm=is_rms_norm, is_rms_norm=is_rms_norm,
return_dropout_mask=True,
) )
out_pt, *rest_pt = layer_norm_ref_fn( dropout_mask = rest[-1] if dropout_p > 0.0 else None
x0_pt, weight_pt, bias_pt, residual=res_pt, eps=1e-6, prenorm=prenorm out_pt = layer_norm_ref_fn(
x0_pt,
weight_pt,
bias_pt,
residual=res_pt,
eps=1e-6,
dropout_p=dropout_p,
prenorm=prenorm,
dropout_mask=dropout_mask,
) )
out_ref, *rest_ref = layer_norm_ref_fn( out_ref = layer_norm_ref_fn(
x0_ref, weight_ref, bias_ref, residual=res_ref, eps=1e-6, prenorm=prenorm, upcast=True x0_ref,
weight_ref,
bias_ref,
residual=res_ref,
eps=1e-6,
dropout_p=dropout_p,
prenorm=prenorm,
dropout_mask=dropout_mask,
upcast=True,
) )
if prenorm: if prenorm:
residual = rest[0] residual = rest[0]
residual_pt = rest_pt[0] out_pt, residual_pt = out_pt
residual_ref = rest_ref[0] out_ref, residual_ref = out_ref
assert out.dtype == input_dtype assert out.dtype == input_dtype
if prenorm: if prenorm:
assert residual.dtype == residual_dtype assert residual.dtype == residual_dtype
assert allclose(residual, residual_pt, residual_ref) assert allclose(residual, residual_pt, residual_ref)
assert allclose(out, out_pt, out_ref) assert allclose(out, out_pt, out_ref)
if dropout_mask is not None:
dropout_fraction = 1.0 - dropout_mask.float().mean()
assert abs(dropout_fraction - dropout_p) < 0.01
g = torch.randn_like(out) / batch_size g = torch.randn_like(out) / batch_size
if not prenorm: if not prenorm:
...@@ -128,9 +156,9 @@ def test_layer_norm( ...@@ -128,9 +156,9 @@ def test_layer_norm(
# @pytest.mark.parametrize("has_residual", [False]) # @pytest.mark.parametrize("has_residual", [False])
@pytest.mark.parametrize("weight_dtype", [torch.float32]) @pytest.mark.parametrize("weight_dtype", [torch.float32])
@pytest.mark.parametrize( @pytest.mark.parametrize(
"input_dtype,residual_dtype", "input_dtype,residual_dtype",
[(torch.float16, torch.float16), (torch.float16, torch.float32)] [(torch.float16, torch.float16), (torch.float16, torch.float32)]
+ ([(torch.bfloat16, torch.bfloat16), (torch.bfloat16, torch.float32)] if is_sm8x else []), + ([(torch.bfloat16, torch.bfloat16), (torch.bfloat16, torch.float32)] if is_sm8x else []),
) )
# @pytest.mark.parametrize("input_dtype,residual_dtype", [(torch.bfloat16, torch.float32)]) # @pytest.mark.parametrize("input_dtype,residual_dtype", [(torch.bfloat16, torch.float32)])
@pytest.mark.parametrize("hidden_size", [192, 2048, 2560, 3000]) @pytest.mark.parametrize("hidden_size", [192, 2048, 2560, 3000])
......
Markdown is supported
0% or .
You are about to add 0 people to the discussion. Proceed with caution.
Finish editing this message first!
Please register or to comment