Skip to content

Revert https://github.com/AI-Hypercomputer/accelerator-microbenchmarks/pull/90 #103

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

Closed
leonardchan wants to merge 8 commits into from
Closed

Revert https://github.com/AI-Hypercomputer/accelerator-microbenchmarks/pull/90 #103

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
8 commits
Select commit Hold shift + click to select a range
90a53bb
Merge pull request #1 from AI-Hypercomputer/tpu7x-auto
leonardchan Feb 5, 2026
b25f34e
Add baseline pipelined flow to H2D benchmark
leonardchan Feb 5, 2026
bbb3163
Add --numactl_binding flag to host_device YAMLs
leonardchan Feb 5, 2026
bf5c79c
Add h2d_type column to H2D/D2H output
leonardchan Feb 5, 2026
1fd1810
Revert "Add baseline pipelined flow to H2D benchmark"
leonardchan Feb 5, 2026
f0d7535
Revert "Add --numactl_binding flag to host_device YAMLs"
leonardchan Feb 5, 2026
1592c08
Revert "Add h2d_type column to H2D/D2H output"
leonardchan Feb 5, 2026
0d6597f
Revert "Revert the changes that were made for an urgent demo (#90)"
leonardchan Feb 5, 2026
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
3 changes: 1 addition & 2 deletions Ironwood/configs/collectives/all_gather_1d.yaml
View file
Open in desktop
Original file line number Diff line number Diff line change
@@ -1,8 +1,7 @@
benchmarks:
- benchmark_name: all_gather
benchmark_sweep_params:
- {matrix_dim_range: {start: 2, end: 8192, multiplier: 2}, dtype: "float32", mesh_shape: "16x4x2", ici_size_range: 128, sharding_strategy: "1x4x1", op_dimension: 1, num_runs: 5} # Parallel Replica
- {matrix_dim_range: {start: 2, end: 8192, multiplier: 2}, dtype: "float32", mesh_shape: "4x4x8", ici_size_range: 128, sharding_strategy: "1x1x4", op_dimension: 1, num_runs: 5} # Non-Parallel Replica
- {matrix_dim_range: {start: 2, end: 8192, multiplier: 2}, dtype: "float32", mesh_shape: "16x4x2", ici_size_range: 128, sharding_strategy: "1x4x1", op_dimension: 1, num_runs: 5}
trace_dir: "../microbenchmarks/all_gather_1d"
csv_path: "../microbenchmarks/all_gather_1d"
xlml_metrics_dir: "../microbenchmarks/all_gather_1d"
Expand Down
3 changes: 1 addition & 2 deletions Ironwood/configs/collectives/all_gather_2d.yaml
View file
Open in desktop
Original file line number Diff line number Diff line change
@@ -1,8 +1,7 @@
benchmarks:
- benchmark_name: all_gather
benchmark_sweep_params:
- {matrix_dim_range: {start: 2, end: 8192, multiplier: 2}, dtype: "float32", mesh_shape: "4x16x2", ici_size_range: 128, sharding_strategy: "1x16x1", op_dimension: 2, num_runs: 5} # Parallel Replica Groups
- {matrix_dim_range: {start: 2, end: 8192, multiplier: 2}, dtype: "float32", mesh_shape: "4x32", ici_size_range: 128, sharding_strategy: "1x32", op_dimension: 2, num_runs: 5} # Non Parallel Replica Groups
- {matrix_dim_range: {start: 2, end: 8192, multiplier: 2}, dtype: "float32", mesh_shape: "4x16x2", ici_size_range: 128, sharding_strategy: "1x16x1", op_dimension: 2, num_runs: 5}
trace_dir: "../microbenchmarks/all_gather_2d"
csv_path: "../microbenchmarks/all_gather_2d"
xlml_metrics_dir: "../microbenchmarks/all_gather_2d"
Expand Down
3 changes: 1 addition & 2 deletions Ironwood/configs/collectives/all_gather_3d.yaml
View file
Open in desktop
Original file line number Diff line number Diff line change
@@ -1,8 +1,7 @@
benchmarks:
- benchmark_name: all_gather
benchmark_sweep_params:
- {matrix_dim_range: {start: 2, end: 8192, multiplier: 2}, dtype: "float32", mesh_shape: "64x2", ici_size_range: 128, sharding_strategy: "64x1", op_dimension: 3, num_runs: 5} # Parallel Replica Groups
- {matrix_dim_range: {start: 2, end: 8192, multiplier: 2}, dtype: "float32", mesh_shape: "4x4x8", ici_size_range: 128, sharding_strategy: "4x4x8", op_dimension: 3, num_runs: 5} # Non Parallel Replica Groups
- {matrix_dim_range: {start: 2, end: 8192, multiplier: 2}, dtype: "float32", mesh_shape: "64x2", ici_size_range: 128, sharding_strategy: "64x1", op_dimension: 3, num_runs: 5}
trace_dir: "../microbenchmarks/all_gather_3d"
csv_path: "../microbenchmarks/all_gather_3d"
xlml_metrics_dir: "../microbenchmarks/all_gather_3d"
Expand Down
10 changes: 3 additions & 7 deletions Ironwood/configs/collectives/all_gather_demo.yaml
View file
Open in desktop
Original file line number Diff line number Diff line change
@@ -1,13 +1,9 @@
benchmarks:
- benchmark_name: all_gather
benchmark_sweep_params:
- {matrix_dim_range: {start: 2, end: 512, multiplier: 2}, dtype: "float32", mesh_shape: "16x4x2", ici_size_range: 128, sharding_strategy: "1x4x1", op_dimension: 1} # Parallel Replica
- {matrix_dim_range: {start: 2, end: 512, multiplier: 2}, dtype: "float32", mesh_shape: "4x4x8", ici_size_range: 128, sharding_strategy: "1x1x4", op_dimension: 1} # Non-Parallel Replica
- {matrix_dim_range: {start: 2, end: 512, multiplier: 2}, dtype: "float32", mesh_shape: "4x32", ici_size_range: 128, sharding_strategy: "1x32", op_dimension: 2} # Non Parallel Replica Groups
- {matrix_dim_range: {start: 2, end: 512, multiplier: 2}, dtype: "float32", mesh_shape: "4x16x2", ici_size_range: 128, sharding_strategy: "1x16x1", op_dimension: 2} # Parallel Replica Groups
- {matrix_dim_range: {start: 2, end: 512, multiplier: 2}, dtype: "float32", mesh_shape: "64x2", ici_size_range: 128, sharding_strategy: "64x1", op_dimension: 3} # Parallel Replica Groups
- {matrix_dim_range: {start: 2, end: 512, multiplier: 2}, dtype: "float32", mesh_shape: "4x4x8", ici_size_range: 128, sharding_strategy: "4x4x8", op_dimension: 3} # Non Parallel Replica Groups

- {matrix_dim_range: {start: 2, end: 512, multiplier: 2}, dtype: "float32", mesh_shape: "16x4x2", ici_size_range: 128, sharding_strategy: "1x4x1", op_dimension: 1}
- {matrix_dim_range: {start: 2, end: 512, multiplier: 2}, dtype: "float32", mesh_shape: "4x16x2", ici_size_range: 128, sharding_strategy: "1x16x1", op_dimension: 2}
- {matrix_dim_range: {start: 2, end: 512, multiplier: 2}, dtype: "float32", mesh_shape: "64x2", ici_size_range: 128, sharding_strategy: "64x1", op_dimension: 3}

warmup_tries: 10
trace_dir: "../microbenchmarks/all_gather_demo"
Expand Down
3 changes: 1 addition & 2 deletions Ironwood/configs/collectives/all_gather_tpu7x_2x2x1.yaml
View file
Open in desktop
Original file line number Diff line number Diff line change
@@ -1,8 +1,7 @@
benchmarks:
- benchmark_name: all_gather
benchmark_sweep_params:
- {matrix_dim_range: {start: 8, end: 16384, multiplier: 2}, dtype: "float32", mesh_shape: "2x2x2", ici_size_range: 8, sharding_strategy: "2x2x1", op_dimension: 1, num_runs: 5} # Parallel Replica
- {matrix_dim_range: {start: 8, end: 16384, multiplier: 2}, dtype: "float32", mesh_shape: "2x2x2", ici_size_range: 8, sharding_strategy: "2x2x2", op_dimension: 1, num_runs: 5} # Non-Parallel Replica
- {matrix_dim_range: {start: 8, end: 16384, multiplier: 2}, dtype: "float32", mesh_shape: "2x2x2", ici_size_range: 8, sharding_strategy: "2x2x1", op_dimension: 1, num_runs: 5}
trace_dir: "../microbenchmarks/all_gather_tpu7x_2x2x1"
csv_path: "../microbenchmarks/all_gather_tpu7x_2x2x1"
xlml_metrics_dir: "../microbenchmarks/all_gather_tpu7x_2x2x1"
Expand Down
3 changes: 1 addition & 2 deletions Ironwood/configs/collectives/all_gather_tpu7x_2x2x2.yaml
View file
Open in desktop
Original file line number Diff line number Diff line change
@@ -1,8 +1,7 @@
benchmarks:
- benchmark_name: all_gather
benchmark_sweep_params:
- {matrix_dim_range: {start: 8, end: 16384, multiplier: 2}, dtype: "float32", mesh_shape: "4x2x2", ici_size_range: 16, sharding_strategy: "4x2x1", op_dimension: 1, num_runs: 5} # Parallel Replica
- {matrix_dim_range: {start: 8, end: 16384, multiplier: 2}, dtype: "float32", mesh_shape: "4x2x2", ici_size_range: 16, sharding_strategy: "4x2x2", op_dimension: 1, num_runs: 5} # Non-Parallel Replica
- {matrix_dim_range: {start: 8, end: 16384, multiplier: 2}, dtype: "float32", mesh_shape: "4x2x2", ici_size_range: 16, sharding_strategy: "4x2x1", op_dimension: 1, num_runs: 5}
trace_dir: "../microbenchmarks/all_gather_tpu7x_2x2x2"
csv_path: "../microbenchmarks/all_gather_tpu7x_2x2x2"
xlml_metrics_dir: "../microbenchmarks/all_gather_tpu7x_2x2x2"
Expand Down
3 changes: 1 addition & 2 deletions Ironwood/configs/collectives/all_gather_tpu7x_2x2x4.yaml
View file
Open in desktop
Original file line number Diff line number Diff line change
@@ -1,8 +1,7 @@
benchmarks:
- benchmark_name: all_gather
benchmark_sweep_params:
- {matrix_dim_range: {start: 16, end: 16384, multiplier: 2}, dtype: "float32", mesh_shape: "4x4x2", ici_size_range: 32, sharding_strategy: "4x4x1", op_dimension: 1, num_runs: 5} # Parallel Replica
- {matrix_dim_range: {start: 16, end: 16384, multiplier: 2}, dtype: "float32", mesh_shape: "4x4x2", ici_size_range: 32, sharding_strategy: "4x4x2", op_dimension: 1, num_runs: 5} # Non-Parallel Replica
- {matrix_dim_range: {start: 16, end: 16384, multiplier: 2}, dtype: "float32", mesh_shape: "4x4x2", ici_size_range: 32, sharding_strategy: "4x4x1", op_dimension: 1, num_runs: 5}
trace_dir: "../microbenchmarks/all_gather_tpu7x_2x2x4"
csv_path: "../microbenchmarks/all_gather_tpu7x_2x2x4"
xlml_metrics_dir: "../microbenchmarks/all_gather_tpu7x_2x2x4"
Expand Down
3 changes: 1 addition & 2 deletions Ironwood/configs/collectives/all_gather_tpu7x_2x4x4.yaml
View file
Open in desktop
Original file line number Diff line number Diff line change
@@ -1,8 +1,7 @@
benchmarks:
- benchmark_name: all_gather
benchmark_sweep_params:
- {matrix_dim_range: {start: 32, end: 16384, multiplier: 2}, dtype: "float32", mesh_shape: "8x4x2", ici_size_range: 64, sharding_strategy: "8x4x1", op_dimension: 1, num_runs: 5} # Parallel Replica
- {matrix_dim_range: {start: 32, end: 16384, multiplier: 2}, dtype: "float32", mesh_shape: "8x4x2", ici_size_range: 64, sharding_strategy: "8x4x2", op_dimension: 1, num_runs: 5} # Non-Parallel Replica
- {matrix_dim_range: {start: 32, end: 16384, multiplier: 2}, dtype: "float32", mesh_shape: "8x4x2", ici_size_range: 64, sharding_strategy: "8x4x1", op_dimension: 1, num_runs: 5}
trace_dir: "../microbenchmarks/all_gather_tpu7x_2x4x4"
csv_path: "../microbenchmarks/all_gather_tpu7x_2x4x4"
xlml_metrics_dir: "../microbenchmarks/all_gather_tpu7x_2x4x4"
Expand Down
5 changes: 2 additions & 3 deletions Ironwood/configs/collectives/all_gather_tpu7x_4x4x4.yaml
View file
Open in desktop
Original file line number Diff line number Diff line change
@@ -1,9 +1,8 @@
benchmarks:
- benchmark_name: all_gather
benchmark_sweep_params:
- {matrix_dim_range: {start: 64, end: 16384, multiplier: 2}, dtype: "float32", mesh_shape: "16x4x2", ici_size_range: 128, sharding_strategy: "16x4x1", op_dimension: 1, num_runs: 5} # Parallel Replica
- {matrix_dim_range: {start: 64, end: 16384, multiplier: 2}, dtype: "float32", mesh_shape: "16x4x2", ici_size_range: 128, sharding_strategy: "16x4x2", op_dimension: 1, num_runs: 5} # Non-Parallel Replica
- {matrix_dim_range: {start: 64, end: 16384, multiplier: 2}, dtype: "float32", mesh_shape: "16x4x2", ici_size_range: 128, sharding_strategy: "16x4x1", op_dimension: 1, num_runs: 10}
trace_dir: "../microbenchmarks/all_gather_tpu7x_4x4x4"
csv_path: "../microbenchmarks/all_gather_tpu7x_4x4x4"
xlml_metrics_dir: "../microbenchmarks/all_gather_tpu7x_4x4x4"
xla_dump_dir: "../microbenchmarks/all_gather_tpu7x_4x4x4/hlo_graphs"
xla_dump_dir: "../microbenchmarks/all_gather_tpu7x_4x4x4/hlo_graphs"
3 changes: 1 addition & 2 deletions Ironwood/configs/collectives/all_gather_tpu7x_4x4x8.yaml
View file
Open in desktop
Original file line number Diff line number Diff line change
@@ -1,8 +1,7 @@
benchmarks:
- benchmark_name: all_gather
benchmark_sweep_params:
- {matrix_dim_range: {start: 128, end: 16384, multiplier: 2}, dtype: "float32", mesh_shape: "32x4x2", ici_size_range: 256, sharding_strategy: "32x4x1", op_dimension: 1, num_runs: 5} # Parallel Replica
- {matrix_dim_range: {start: 128, end: 16384, multiplier: 2}, dtype: "float32", mesh_shape: "32x4x2", ici_size_range: 256, sharding_strategy: "32x4x2", op_dimension: 1, num_runs: 5} # Non-Parallel Replica
- {matrix_dim_range: {start: 128, end: 16384, multiplier: 2}, dtype: "float32", mesh_shape: "32x4x2", ici_size_range: 256, sharding_strategy: "32x4x1", op_dimension: 1, num_runs: 5}
trace_dir: "../microbenchmarks/all_gather_tpu7x_4x4x8"
csv_path: "../microbenchmarks/all_gather_tpu7x_4x4x8"
xlml_metrics_dir: "../microbenchmarks/all_gather_tpu7x_4x4x8"
Expand Down
3 changes: 1 addition & 2 deletions Ironwood/configs/collectives/all_reduce_1d.yaml
View file
Open in desktop
Original file line number Diff line number Diff line change
@@ -1,8 +1,7 @@
benchmarks:
- benchmark_name: psum
benchmark_sweep_params:
- {matrix_dim_range: {start: 2, end: 8192, multiplier: 2}, dtype: "float32", mesh_shape: "16x4x2", ici_size_range: 128, sharding_strategy: "1x4x1" , op_dimension: 1, num_runs: 5} # Parallel Replica
- {matrix_dim_range: {start: 2, end: 8192, multiplier: 2}, dtype: "float32", mesh_shape: "4x4x8", ici_size_range: 128, sharding_strategy: "1x1x4", op_dimension: 1, num_runs: 5} # Non Parallel Replica
- {matrix_dim_range: {start: 2, end: 8192, multiplier: 2}, dtype: "float32", mesh_shape: "16x4x2", ici_size_range: 128, sharding_strategy: "1x4x1" , op_dimension: 1, num_runs: 5}
trace_dir: "../microbenchmarks/all_reduce_1d"
csv_path: "../microbenchmarks/all_reduce_1d"
xlml_metrics_dir: "../microbenchmarks/all_reduce_1d"
Expand Down
3 changes: 1 addition & 2 deletions Ironwood/configs/collectives/all_reduce_2d.yaml
View file
Open in desktop
Original file line number Diff line number Diff line change
@@ -1,8 +1,7 @@
benchmarks:
- benchmark_name: psum
benchmark_sweep_params:
- {matrix_dim_range: {start: 2, end: 8192, multiplier: 2}, dtype: "float32", mesh_shape: "4x16x2", ici_size_range: 128, sharding_strategy: "1x16x1", op_dimension: 2, num_runs: 5} # Parallel Replica Groups
- {matrix_dim_range: {start: 2, end: 8192, multiplier: 2}, dtype: "float32", mesh_shape: "4x32", ici_size_range: 128, sharding_strategy: "1x32", op_dimension: 2, num_runs: 5} # Non Parallel Replica Groups
- {matrix_dim_range: {start: 2, end: 8192, multiplier: 2}, dtype: "float32", mesh_shape: "4x16x2", ici_size_range: 128, sharding_strategy: "1x16x1", op_dimension: 2, num_runs: 5}
trace_dir: "../microbenchmarks/all_reduce_2d"
csv_path: "../microbenchmarks/all_reduce_2d"
xlml_metrics_dir: "../microbenchmarks/all_reduce_2d"
Expand Down
3 changes: 1 addition & 2 deletions Ironwood/configs/collectives/all_reduce_3d.yaml
View file
Open in desktop
Original file line number Diff line number Diff line change
@@ -1,8 +1,7 @@
benchmarks:
- benchmark_name: psum
benchmark_sweep_params:
- {matrix_dim_range: {start: 2, end: 8192, multiplier: 2}, dtype: "float32", mesh_shape: "64x2", ici_size_range: 128, sharding_strategy: "64x1", op_dimension: 3, num_runs: 5} # Parallel Replica Groups
- {matrix_dim_range: {start: 2, end: 8192, multiplier: 2}, dtype: "float32", mesh_shape: "4x4x8", ici_size_range: 128, sharding_strategy: "4x4x8", op_dimension: 3, num_runs: 5} # Non Parallel Replica Groups
- {matrix_dim_range: {start: 2, end: 8192, multiplier: 2}, dtype: "float32", mesh_shape: "64x2", ici_size_range: 128, sharding_strategy: "64x1", op_dimension: 3, num_runs: 5}
trace_dir: "../microbenchmarks/all_reduce_3d"
csv_path: "../microbenchmarks/all_reduce_3d"
xlml_metrics_dir: "../microbenchmarks/all_reduce_3d"
Expand Down
3 changes: 1 addition & 2 deletions Ironwood/configs/collectives/all_reduce_tpu7x_2x2x1.yaml
View file
Open in desktop
Original file line number Diff line number Diff line change
@@ -1,8 +1,7 @@
benchmarks:
- benchmark_name: psum
benchmark_sweep_params:
- {matrix_dim_range: {start: 8, end: 16384, multiplier: 2}, dtype: "float32", mesh_shape: "2x2x2", ici_size_range: 8, sharding_strategy: "2x2x1", op_dimension: 1, num_runs: 5} # Parallel Replica
- {matrix_dim_range: {start: 8, end: 16384, multiplier: 2}, dtype: "float32", mesh_shape: "2x2x2", ici_size_range: 8, sharding_strategy: "2x2x2", op_dimension: 1, num_runs: 5} # Non-Parallel Replica
- {matrix_dim_range: {start: 8, end: 16384, multiplier: 2}, dtype: "float32", mesh_shape: "2x2x2", ici_size_range: 8, sharding_strategy: "2x2x1", op_dimension: 1, num_runs: 5}
trace_dir: "../microbenchmarks/psum_tpu7x_2x2x1"
csv_path: "../microbenchmarks/psum_tpu7x_2x2x1"
xlml_metrics_dir: "../microbenchmarks/psum_tpu7x_2x2x1"
Expand Down
3 changes: 1 addition & 2 deletions Ironwood/configs/collectives/all_reduce_tpu7x_2x2x2.yaml
View file
Open in desktop
Original file line number Diff line number Diff line change
@@ -1,8 +1,7 @@
benchmarks:
- benchmark_name: psum
benchmark_sweep_params:
- {matrix_dim_range: {start: 8, end: 16384, multiplier: 2}, dtype: "float32", mesh_shape: "4x2x2", ici_size_range: 16, sharding_strategy: "4x2x1", op_dimension: 1, num_runs: 5} # Parallel Replica
- {matrix_dim_range: {start: 8, end: 16384, multiplier: 2}, dtype: "float32", mesh_shape: "4x2x2", ici_size_range: 16, sharding_strategy: "4x2x2", op_dimension: 1, num_runs: 5} # Non-Parallel Replica
- {matrix_dim_range: {start: 8, end: 16384, multiplier: 2}, dtype: "float32", mesh_shape: "4x2x2", ici_size_range: 16, sharding_strategy: "4x2x1", op_dimension: 1, num_runs: 5}
trace_dir: "../microbenchmarks/psum_tpu7x_2x2x2"
csv_path: "../microbenchmarks/psum_tpu7x_2x2x2"
xlml_metrics_dir: "../microbenchmarks/psum_tpu7x_2x2x2"
Expand Down
3 changes: 1 addition & 2 deletions Ironwood/configs/collectives/all_reduce_tpu7x_2x2x4.yaml
View file
Open in desktop
Original file line number Diff line number Diff line change
@@ -1,8 +1,7 @@
benchmarks:
- benchmark_name: psum
benchmark_sweep_params:
- {matrix_dim_range: {start: 16, end: 16384, multiplier: 2}, dtype: "float32", mesh_shape: "4x4x2", ici_size_range: 32, sharding_strategy: "4x4x1", op_dimension: 1, num_runs: 5} # Parallel Replica
- {matrix_dim_range: {start: 16, end: 16384, multiplier: 2}, dtype: "float32", mesh_shape: "4x4x2", ici_size_range: 32, sharding_strategy: "4x4x2", op_dimension: 1, num_runs: 5} # Non-Parallel Replica
- {matrix_dim_range: {start: 16, end: 16384, multiplier: 2}, dtype: "float32", mesh_shape: "4x4x2", ici_size_range: 32, sharding_strategy: "4x4x1", op_dimension: 1, num_runs: 5}
trace_dir: "../microbenchmarks/psum_tpu7x_2x2x4"
csv_path: "../microbenchmarks/psum_tpu7x_2x2x4"
xlml_metrics_dir: "../microbenchmarks/psum_tpu7x_2x2x4"
Expand Down
3 changes: 1 addition & 2 deletions Ironwood/configs/collectives/all_reduce_tpu7x_2x4x4.yaml
View file
Open in desktop
Original file line number Diff line number Diff line change
@@ -1,8 +1,7 @@
benchmarks:
- benchmark_name: psum
benchmark_sweep_params:
- {matrix_dim_range: {start: 32, end: 16384, multiplier: 2}, dtype: "float32", mesh_shape: "8x4x2", ici_size_range: 64, sharding_strategy: "8x4x1", op_dimension: 1, num_runs: 5} # Parallel Replica
- {matrix_dim_range: {start: 32, end: 16384, multiplier: 2}, dtype: "float32", mesh_shape: "8x4x2", ici_size_range: 64, sharding_strategy: "8x4x2", op_dimension: 1, num_runs: 5} # Non-Parallel Replica
- {matrix_dim_range: {start: 32, end: 16384, multiplier: 2}, dtype: "float32", mesh_shape: "8x4x2", ici_size_range: 64, sharding_strategy: "8x4x1", op_dimension: 1, num_runs: 5}
trace_dir: "../microbenchmarks/psum_tpu7x_2x4x4"
csv_path: "../microbenchmarks/psum_tpu7x_2x4x4"
xlml_metrics_dir: "../microbenchmarks/psum_tpu7x_2x4x4"
Expand Down
3 changes: 1 addition & 2 deletions Ironwood/configs/collectives/all_reduce_tpu7x_4x4x4.yaml
View file
Open in desktop
Original file line number Diff line number Diff line change
@@ -1,8 +1,7 @@
benchmarks:
- benchmark_name: psum
benchmark_sweep_params:
- {matrix_dim_range: {start: 64, end: 16384, multiplier: 2}, dtype: "float32", mesh_shape: "16x4x2", ici_size_range: 128, sharding_strategy: "16x4x1", op_dimension: 1, num_runs: 5} # Parallel Replica
- {matrix_dim_range: {start: 64, end: 16384, multiplier: 2}, dtype: "float32", mesh_shape: "16x4x2", ici_size_range: 128, sharding_strategy: "16x4x2", op_dimension: 1, num_runs: 5} # Non-Parallel Replica
- {matrix_dim_range: {start: 64, end: 16384, multiplier: 2}, dtype: "float32", mesh_shape: "16x4x2", ici_size_range: 128, sharding_strategy: "16x4x1", op_dimension: 1, num_runs: 5}
trace_dir: "../microbenchmarks/psum_tpu7x_4x4x4"
csv_path: "../microbenchmarks/psum_tpu7x_4x4x4"
xlml_metrics_dir: "../microbenchmarks/psum_tpu7x_4x4x4"
Expand Down
3 changes: 1 addition & 2 deletions Ironwood/configs/collectives/all_reduce_tpu7x_4x4x8.yaml
View file
Open in desktop
Original file line number Diff line number Diff line change
@@ -1,8 +1,7 @@
benchmarks:
- benchmark_name: psum
benchmark_sweep_params:
- {matrix_dim_range: {start: 128, end: 16384, multiplier: 2}, dtype: "float32", mesh_shape: "32x4x2", ici_size_range: 256, sharding_strategy: "32x4x1", op_dimension: 1, num_runs: 5} # Parallel Replica
- {matrix_dim_range: {start: 128, end: 16384, multiplier: 2}, dtype: "float32", mesh_shape: "32x4x2", ici_size_range: 256, sharding_strategy: "32x4x2", op_dimension: 1, num_runs: 5} # Non-Parallel Replica
- {matrix_dim_range: {start: 128, end: 16384, multiplier: 2}, dtype: "float32", mesh_shape: "32x4x2", ici_size_range: 256, sharding_strategy: "32x4x1", op_dimension: 1, num_runs: 5}
trace_dir: "../microbenchmarks/psum_tpu7x_4x4x8"
csv_path: "../microbenchmarks/psum_tpu7x_4x4x8"
xlml_metrics_dir: "../microbenchmarks/psum_tpu7x_4x4x8"
Expand Down
2 changes: 1 addition & 1 deletion Ironwood/configs/collectives/all_to_all_1d.yaml
View file
Open in desktop
Original file line number Diff line number Diff line change
@@ -1,7 +1,7 @@
benchmarks:
- benchmark_name: all_to_all
benchmark_sweep_params:
- {matrix_dim_range: {start: 2, end: 8192, multiplier: 2}, dtype: "float32", mesh_shape: "4x4x8", ici_size_range: 128, sharding_strategy: "1x1x4", op_dimension: 1, num_runs: 5} # Non Parallel Replica
- {matrix_dim_range: {start: 2, end: 8192, multiplier: 2}, dtype: "float32", mesh_shape: "4x4x8", ici_size_range: 128, sharding_strategy: "1x1x4", op_dimension: 1, num_runs: 5}
trace_dir: "../microbenchmarks/all_to_all_1d"
csv_path: "../microbenchmarks/all_to_all_1d"
xlml_metrics_dir: "../microbenchmarks/all_to_all_1d"
Expand Down
2 changes: 1 addition & 1 deletion Ironwood/configs/collectives/all_to_all_2d.yaml
View file
Open in desktop
Original file line number Diff line number Diff line change
@@ -1,7 +1,7 @@
benchmarks:
- benchmark_name: all_to_all
benchmark_sweep_params:
- {matrix_dim_range: {start: 2, end: 8192, multiplier: 2}, dtype: "float32", mesh_shape: "4x32", ici_size_range: 128, sharding_strategy: "1x32", op_dimension: 2, num_runs: 5} # Non Parallel Replica
- {matrix_dim_range: {start: 2, end: 8192, multiplier: 2}, dtype: "float32", mesh_shape: "4x32", ici_size_range: 128, sharding_strategy: "1x32", op_dimension: 2, num_runs: 5}
trace_dir: "../microbenchmarks/all_to_all_2d"
csv_path: "../microbenchmarks/all_to_all_2d"
xlml_metrics_dir: "../microbenchmarks/all_to_all_2d"
Expand Down
2 changes: 1 addition & 1 deletion Ironwood/configs/collectives/all_to_all_3d.yaml
View file
Open in desktop
Original file line number Diff line number Diff line change
@@ -1,7 +1,7 @@
benchmarks:
- benchmark_name: all_to_all
benchmark_sweep_params:
- {matrix_dim_range: {start: 2, end: 8192, multiplier: 2}, dtype: "float32", mesh_shape: "4x4x8", ici_size_range: 128, sharding_strategy: "4x4x8", op_dimension: 3, num_runs: 5} # Non Parallel Replica Groups
- {matrix_dim_range: {start: 2, end: 8192, multiplier: 2}, dtype: "float32", mesh_shape: "4x4x8", ici_size_range: 128, sharding_strategy: "4x4x8", op_dimension: 3, num_runs: 5}
trace_dir: "../microbenchmarks/all_to_all_3d"
csv_path: "../microbenchmarks/all_to_all_3d"
xlml_metrics_dir: "../microbenchmarks/all_to_all_3d"
Expand Down
3 changes: 1 addition & 2 deletions Ironwood/configs/collectives/all_to_all_tpu7x_2x2x1.yaml
View file
Open in desktop
Original file line number Diff line number Diff line change
@@ -1,8 +1,7 @@
benchmarks:
- benchmark_name: all_to_all
benchmark_sweep_params:
- {matrix_dim_range: {start: 8, end: 16384, multiplier: 2}, dtype: "float32", mesh_shape: "2x2x2", ici_size_range: 8, sharding_strategy: "2x2x1", op_dimension: 1, num_runs: 5} # Parallel Replica
- {matrix_dim_range: {start: 8, end: 16384, multiplier: 2}, dtype: "float32", mesh_shape: "2x2x2", ici_size_range: 8, sharding_strategy: "2x2x2", op_dimension: 1, num_runs: 5} # Non-Parallel Replica
- {matrix_dim_range: {start: 8, end: 16384, multiplier: 2}, dtype: "float32", mesh_shape: "2x2x2", ici_size_range: 8, sharding_strategy: "2x2x1", op_dimension: 1, num_runs: 5}
trace_dir: "../microbenchmarks/all_to_all_tpu7x_2x2x1"
csv_path: "../microbenchmarks/all_to_all_tpu7x_2x2x1"
xlml_metrics_dir: "../microbenchmarks/all_to_all_tpu7x_2x2x1"
Expand Down
Loading