Skip to content

Compile CP softmax LSE corrections with dynamic shapes #695

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Open
JessicaJiang-123 wants to merge 1 commit into
base: dev
Choose a base branch
from
+24 −3
Open

Compile CP softmax LSE corrections with dynamic shapes #695

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
1 commit
Select commit
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
6 changes: 3 additions & 3 deletions transformer_engine/pytorch/attention/dot_product_attention/context_parallel.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -27,7 +27,7 @@
from transformer_engine.pytorch.tensor.float8_tensor import Float8Tensor
from transformer_engine.pytorch.tensor.storage.float8_tensor_storage import Float8TensorStorage
from transformer_engine.pytorch.quantized_tensor import QuantizedTensorStorage
from transformer_engine.pytorch.jit import jit_fuser
from transformer_engine.pytorch.jit import jit_fuser, jit_fuser_dynamic
from transformer_engine.pytorch.graph import is_graph_capturing
from transformer_engine.pytorch.constants import dist_group_type
from transformer_engine.pytorch.distributed import (
Expand Down Expand Up @@ -184,7 +184,7 @@ def flash_attn_fwd_second_half_out_correction(
out_.add_(out_corrected)


@jit_fuser
@jit_fuser_dynamic
def flash_attn_fwd_softmax_lse_correction(
softmax_lse: torch.Tensor,
softmax_lse_per_step: torch.Tensor,
Expand All @@ -196,7 +196,7 @@ def flash_attn_fwd_softmax_lse_correction(
softmax_lse.copy_(new_scale)


@jit_fuser
@jit_fuser_dynamic
def flash_attn_fwd_second_half_softmax_lse_correction(
softmax_lse: torch.Tensor,
softmax_lse_per_step: torch.Tensor,
Expand Down
21 changes: 21 additions & 0 deletions transformer_engine/pytorch/jit.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -36,9 +36,30 @@ def wrapper(*args, **kwargs):
return wrapper


# See: https://github.com/ROCm/TransformerEngine/issues/693
def lazy_compile_dynamic(func):
"""Lazy compile a function with torch.compile, using dynamic shapes

Compiling dynamic up front skips the static specialization dynamo would otherwise build on
the first shape it sees, leaving one compiled artifact for the process instead of two.
"""
compiled_func = None

@wraps(func)
def wrapper(*args, **kwargs):
nonlocal compiled_func
if compiled_func is None:
compiled_func = torch.compile(func, dynamic=True)
return compiled_func(*args, **kwargs)

return wrapper


jit_fuser = lambda func: func
jit_fuser_dynamic = lambda func: func
if torch_version() >= (2, 0, 0) and bool(int(os.getenv("NVTE_TORCH_COMPILE", "1"))):
jit_fuser = lazy_compile
jit_fuser_dynamic = lazy_compile_dynamic


# See: https://github.com/NVIDIA/TransformerEngine/issues/597
Expand Down
Loading