Add trtlllm to triton bench #379
Open
Conversation
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
|
This pull request was exported from Phabricator. Differential Revision: D81021980 |
Aya-ZIbra
temporarily deployed
to
docker-s3-upload
GitHub Actions
Inactive
Aya-ZIbra
had a problem deploying
to
docker-s3-upload
GitHub Actions
Failure
|
This pull request was exported from Phabricator. Differential Revision: D81021980 |
Aya-ZIbra
added a commit
to Aya-ZIbra/tritonbench
that referenced
this pull request
Aug 29, 2025
Summary: Pull Request resolved: meta-pytorch#379 Run C++ FLASHINFER_CUBIN_DIR=/data/users/$USER/fbsource/fbcode/deeplearning/flashinfer/fb/cubins/ buck2 run mode/opt mode/inplace -c fbcode.enable_gpu_sections=true -c fbcode.nvcc_arch=b200a -c fbcode.platform010_cuda_version=12.8 //deeplearning/flashinfer/trtllm_kernel_interfaces:run_example``` ------- Run Triton bench buck2 run mode/opt mode/inplace -c fbcode.enable_gpu_sections=true -c fbcode.nvcc_arch=b200a -c fbcode.platform010_cuda_version=12.8 //pytorch/tritonbench:run -- --op decoding_attention --only trtllm_decode_fmha --seq-len-q 1 --metrics gbps Todo: Support non-paged case Differential Revision: D81021980
Aya-ZIbra
force-pushed
the
export-D81021980
branch
from
c20554b to
97ade45
Compare
xuzhao9
approved these changes
Aug 30, 2025
Aya-ZIbra
had a problem deploying
to
docker-s3-upload
GitHub Actions
Failure
Aya-ZIbra
temporarily deployed
to
docker-s3-upload
GitHub Actions
Inactive
|
This pull request was exported from Phabricator. Differential Revision: D81021980 |
Aya-ZIbra
force-pushed
the
export-D81021980
branch
from
97ade45 to
e3b390c
Compare
Aya-ZIbra
added a commit
to Aya-ZIbra/tritonbench
that referenced
this pull request
Sep 4, 2025
Summary: Pull Request resolved: meta-pytorch#379 Run C++ FLASHINFER_CUBIN_DIR=/data/users/$USER/fbsource/fbcode/deeplearning/flashinfer/fb/cubins/ buck2 run mode/opt mode/inplace -c fbcode.enable_gpu_sections=true -c fbcode.nvcc_arch=b200a -c fbcode.platform010_cuda_version=12.8 //deeplearning/flashinfer/trtllm_kernel_interfaces:run_example``` ------- Run Triton bench buck2 run mode/opt mode/inplace -c fbcode.enable_gpu_sections=true -c fbcode.nvcc_arch=b200a -c fbcode.platform010_cuda_version=12.8 //pytorch/tritonbench:run -- --op decoding_attention --only trtllm_decode_fmha --seq-len-q 1 --metrics gbps Todo: Support non-paged case Differential Revision: D81021980
Aya-ZIbra
temporarily deployed
to
docker-s3-upload
GitHub Actions
Inactive
Aya-ZIbra
temporarily deployed
to
docker-s3-upload
GitHub Actions
Inactive
Summary: use of headq = 8 , is doing much better. Maybe because headq= 5 probably doesn't work with TMA_q used here. Differential Revision: D80830933
Aya-ZIbra
added a commit
to Aya-ZIbra/tritonbench
that referenced
this pull request
Sep 4, 2025
Summary:
Run C++
FLASHINFER_CUBIN_DIR=/data/users/$USER/fbsource/fbcode/deeplearning/flashinfer/fb/cubins/
buck2 run mode/opt mode/inplace -c fbcode.enable_gpu_sections=true -c fbcode.nvcc_arch=b200a -c fbcode.platform010_cuda_version=12.8 //deeplearning/flashinfer/trtllm_kernel_interfaces:run_example```
-------
Run Triton bench
buck2 run mode/opt mode/inplace -c fbcode.enable_gpu_sections=true -c fbcode.nvcc_arch=b200a -c fbcode.platform010_cuda_version=12.8 //pytorch/tritonbench:run -- --op decoding_attention --only trtllm_decode_fmha --seq-len-q 1 --metrics gbps
Todo: Support non-paged case
Differential Revision: D81021980
Aya-ZIbra
force-pushed
the
export-D81021980
branch
from
e3b390c to
a965274
Compare
Summary: Pull Request resolved: meta-pytorch#379 Run C++ FLASHINFER_CUBIN_DIR=/data/users/$USER/fbsource/fbcode/deeplearning/flashinfer/fb/cubins/ buck2 run mode/opt mode/inplace -c fbcode.enable_gpu_sections=true -c fbcode.nvcc_arch=b200a -c fbcode.platform010_cuda_version=12.8 //deeplearning/flashinfer/trtllm_kernel_interfaces:run_example``` ------- Run Triton bench buck2 run mode/opt mode/inplace -c fbcode.enable_gpu_sections=true -c fbcode.nvcc_arch=b200a -c fbcode.platform010_cuda_version=12.8 //pytorch/tritonbench:run -- --op decoding_attention --only trtllm_decode_fmha --seq-len-q 1 --metrics gbps Todo: Support non-paged case Differential Revision: D81021980
|
This pull request was exported from Phabricator. Differential Revision: D81021980 |
Aya-ZIbra
force-pushed
the
export-D81021980
branch
from
a965274 to
7a4063e
Compare
Aya-ZIbra
temporarily deployed
to
docker-s3-upload
GitHub Actions
Inactive
Aya-ZIbra
temporarily deployed
to
docker-s3-upload
GitHub Actions
Inactive
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.
This suggestion is invalid because no changes were made to the code.
Suggestions cannot be applied while the pull request is closed.
Suggestions cannot be applied while viewing a subset of changes.
Only one suggestion per line can be applied in a batch.
Add this suggestion to a batch that can be applied as a single commit.
Applying suggestions on deleted lines is not supported.
You must change the existing code in this line in order to create a valid suggestion.
Outdated suggestions cannot be applied.
This suggestion has been applied or marked resolved.
Suggestions cannot be applied from pending reviews.
Suggestions cannot be applied on multi-line comments.
Suggestions cannot be applied while the pull request is queued to merge.
Suggestion cannot be applied right now. Please check back later.
Summary:
Run C++
FLASHINFER_CUBIN_DIR=/data/users/$USER/fbsource/fbcode/deeplearning/flashinfer/fb/cubins/
buck2 run mode/opt mode/inplace -c fbcode.enable_gpu_sections=true -c fbcode.nvcc_arch=b200a -c fbcode.platform010_cuda_version=12.8 //deeplearning/flashinfer/trtllm_kernel_interfaces:run_example```
Run Triton bench
buck2 run mode/opt mode/inplace -c fbcode.enable_gpu_sections=true -c fbcode.nvcc_arch=b200a -c fbcode.platform010_cuda_version=12.8 //pytorch/tritonbench:run -- --op decoding_attention --only trtllm_decode_fmha --seq-len-q 1 --metrics gbps
Todo: Support non-paged case
Differential Revision: D81021980