Skip to content

temp changes to enable benchmarking on device via adb #18719

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
kimishpatel wants to merge 1 commit into
base: gh/kimishpatel/239/base
Choose a base branch
from
+928 −1
Open

temp changes to enable benchmarking on device via adb #18719

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: 6 additions & 0 deletions extension/llm/custom_ops/BUCK
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -134,6 +134,12 @@ fbcode_target(_kind = runtime.python_binary,
],
)

fbcode_target(_kind = runtime.python_binary,
name = "run_bench_on_device",
srcs = ["run_bench_on_device.py"],
main_function = "executorch.extension.llm.custom_ops.run_bench_on_device.main",
)

fbcode_target(_kind = cpp_benchmark,
name = "bench_sdpa",
srcs = ["bench_sdpa.cpp"],
Expand Down
6 changes: 5 additions & 1 deletion extension/llm/custom_ops/bench_sdpa.cpp
View file
Open in desktop
Original file line number Diff line number Diff line change
@@ -1,4 +1,4 @@
// (c) Meta Platforms, Inc. and affiliates. Confidential and proprietary.

Check failure on line 1 in extension/llm/custom_ops/bench_sdpa.cpp

View workflow job for this annotation

GitHub Actions / lintrunner

LICENSELINT Wrong license 

Code contributed to ExecuTorch open source repo should have BSD-license header

/*
* Benchmark for SDPA (scaled dot-product attention) implementations.
Expand Down Expand Up @@ -296,13 +296,14 @@
fill_random(k, gen);
fill_random(v, gen);

// Reference: ET custom_sdpa_out (10-param signature, standard layout)
// Reference: ET custom_sdpa_out (standard [B,S,H,D] layout)
Tensor out_ref = tf.zeros(
{(int32_t)batch, (int32_t)q_seq_len, (int32_t)Hq, (int32_t)D});
KernelRuntimeContext ctx{};
torch::executor::native::custom_sdpa_out(
ctx, q, k, v, start_pos,
std::nullopt, 0.0, true, std::nullopt,
false, false, false,
out_ref);

// Test: GEMM-based standard SDPA
Expand Down Expand Up @@ -473,6 +474,9 @@
0.0, // dropout_p
true, // is_causal
std::nullopt, // scale
false, // is_seq_dim_2
false, // is_k_seq_dim_2
false, // is_v_seq_dim_2
*output_);
}
}
Expand Down
Loading
Loading