Skip to content
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

Jump to bottom

[Misc] Upgrade to pytorch 2.5 #9588

Merged
merged 10 commits into from
Oct 27, 2024
+48 −25
Merged

[Misc] Upgrade to pytorch 2.5 #9588

Show file tree
Hide file tree
Changes from 8 commits
Commits
Show all changes
10 commits
Select commit Hold shift + click to select a range
e970933
upgrade to pytorch 2.5 (still needs newer xformers
bnellnm Oct 17, 2024
5a54b48
update xformers
bnellnm Oct 22, 2024
1ea632b
use better method for specifying cuda libs
bnellnm Oct 25, 2024
c00efe3
try to make pytorch 2.5 use CUDA 12.1
bnellnm Oct 25, 2024
c4ebdba
remove extra index 12.1 for now
bnellnm Oct 25, 2024
39f9c34
update flash attn git hash
bnellnm Oct 25, 2024
f1f3caa
downgrade cpu pytorch version
bnellnm Oct 25, 2024
568510a
Merge branch 'main' into pt25
youkaichao Oct 27, 2024
0068133
fix sdpa cudnn
youkaichao Oct 27, 2024
5250a65
fix tests
youkaichao Oct 27, 2024
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
4 changes: 2 additions & 2 deletions CMakeLists.txt
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -49,7 +49,7 @@ set(HIP_SUPPORTED_ARCHS "gfx906;gfx908;gfx90a;gfx940;gfx941;gfx942;gfx1030;gfx11
# requirements.txt files and should be kept consistent. The ROCm torch
# versions are derived from Dockerfile.rocm
#
set(TORCH_SUPPORTED_VERSION_CUDA "2.4.0")
set(TORCH_SUPPORTED_VERSION_CUDA "2.5.0")
set(TORCH_SUPPORTED_VERSION_ROCM "2.5.0")

#
Expand Down Expand Up @@ -507,7 +507,7 @@ else()
FetchContent_Declare(
vllm-flash-attn
GIT_REPOSITORY https://github.com/vllm-project/flash-attention.git
GIT_TAG 013f0c4fc47e6574060879d9734c1df8c5c273bd
GIT_TAG 5259c586c403a4e4d8bf69973c159b40cc346fb9
bnellnm marked this conversation as resolved.
Show resolved Hide resolved
GIT_PROGRESS TRUE
# Don't share the vllm-flash-attn build between build types
BINARY_DIR ${CMAKE_BINARY_DIR}/vllm-flash-attn
Expand Down
6 changes: 1 addition & 5 deletions cmake/utils.cmake
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -424,11 +424,7 @@ function (define_gpu_extension_target GPU_MOD_NAME)
# Don't use `TORCH_LIBRARIES` for CUDA since it pulls in a bunch of
# dependencies that are not necessary and may not be installed.
if (GPU_LANGUAGE STREQUAL "CUDA")
if ("${CUDA_CUDA_LIB}" STREQUAL "")
set(CUDA_CUDA_LIB "${CUDA_CUDA_LIBRARY}")
endif()
target_link_libraries(${GPU_MOD_NAME} PRIVATE ${CUDA_CUDA_LIB}
${CUDA_LIBRARIES})
target_link_libraries(${GPU_MOD_NAME} PRIVATE CUDA::cudart CUDA::cuda_driver)
Copy link
Collaborator

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Nice!

else()
target_link_libraries(${GPU_MOD_NAME} PRIVATE ${TORCH_LIBRARIES})
endif()
Expand Down
2 changes: 1 addition & 1 deletion pyproject.toml
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -6,7 +6,7 @@ requires = [
"packaging",
"setuptools>=61",
"setuptools-scm>=8.0",
"torch == 2.4.0",
"torch == 2.5.0",
"wheel",
"jinja2",
]
Expand Down
2 changes: 1 addition & 1 deletion requirements-build.txt
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -4,6 +4,6 @@ ninja
packaging
setuptools>=61
setuptools-scm>=8
torch==2.4.0
torch==2.5.0
wheel
jinja2
6 changes: 3 additions & 3 deletions requirements-cuda.txt
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -4,7 +4,7 @@
# Dependencies for NVIDIA GPUs
ray >= 2.9
nvidia-ml-py # for pynvml package
torch == 2.4.0
torch == 2.5.0
Copy link
Collaborator

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Only concern here is now torch==2.5.0 uses the 12.4 cuda bindings by default. We might want to update the installation docs (including on the readme) to alert users that they may want to pass --extra-index-url https://download.pytorch.org/whl/cu121 during installation depending on the machine they are using

Copy link
Contributor Author

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

@robertgshaw2-neuralmagic Does this mean that there would need to be multiple vllm packages (one for 12.1 and one for 12.4)? Or should I try to install pytorch 2.5 built with 12.1 (if such a thing exists)?

# These must be updated alongside torch
torchvision == 0.19 # Required for phi3v processor. See https://github.com/pytorch/vision?tab=readme-ov-file#installation for corresponding version
xformers == 0.0.27.post2; platform_system == 'Linux' and platform_machine == 'x86_64' # Requires PyTorch 2.4.0
torchvision == 0.20 # Required for phi3v processor. See https://github.com/pytorch/vision?tab=readme-ov-file#installation for corresponding version
xformers == 0.0.28.post2; platform_system == 'Linux' and platform_machine == 'x86_64' # Requires PyTorch 2.5.0
2 changes: 1 addition & 1 deletion requirements-openvino.txt
View file
Open in desktop
Original file line number Diff line number Diff line change
@@ -1,7 +1,7 @@
# Common dependencies
-r requirements-common.txt

torch == 2.4.0 # should be aligned with "common" vLLM torch version
torch == 2.5.0 # should be aligned with "common" vLLM torch version
openvino >= 2024.4.0 # since 2024.4.0 both CPU and GPU support Paged Attention

optimum @ git+https://github.com/huggingface/optimum.git@main # latest optimum is used to support latest transformers version
Expand Down