Skip to content

Commit a70b701

Browse files
junjieqianjunjieqian
authored
add extra datatypes in configs (#94)
1 parent 69661f9 commit a70b701

20 files changed

Lines changed: 84 additions & 1 deletion

‎Ironwood/configs/collectives/all_gather_1d.yaml‎

Lines changed: 4 additions & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -2,6 +2,10 @@ benchmarks:
22
- benchmark_name: all_gather
33
benchmark_sweep_params:
44
- {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}
5+
- {matrix_dim_range: {start: 2, end: 8192, multiplier: 2}, dtype: "float16", mesh_shape: "16x4x2", ici_size_range: 128, sharding_strategy: "1x4x1", op_dimension: 1, num_runs: 5}
6+
- {matrix_dim_range: {start: 2, end: 8192, multiplier: 2}, dtype: "float8", mesh_shape: "16x4x2", ici_size_range: 128, sharding_strategy: "1x4x1", op_dimension: 1, num_runs: 5}
7+
- {matrix_dim_range: {start: 2, end: 8192, multiplier: 2}, dtype: "float4", mesh_shape: "16x4x2", ici_size_range: 128, sharding_strategy: "1x4x1", op_dimension: 1, num_runs: 5}
8+
- {matrix_dim_range: {start: 2, end: 8192, multiplier: 2}, dtype: "bfloat16", mesh_shape: "16x4x2", ici_size_range: 128, sharding_strategy: "1x4x1", op_dimension: 1, num_runs: 5}
59
trace_dir: "../microbenchmarks/all_gather_1d"
610
csv_path: "../microbenchmarks/all_gather_1d"
711
xlml_metrics_dir: "../microbenchmarks/all_gather_1d"

‎Ironwood/configs/collectives/all_gather_tpu7x_2x2x1.yaml‎

Lines changed: 4 additions & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -2,6 +2,10 @@ benchmarks:
22
- benchmark_name: all_gather
33
benchmark_sweep_params:
44
- {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}
5+
- {matrix_dim_range: {start: 8, end: 16384, multiplier: 2}, dtype: "float16", mesh_shape: "2x2x2", ici_size_range: 8, sharding_strategy: "2x2x1", op_dimension: 1, num_runs: 5}
6+
- {matrix_dim_range: {start: 8, end: 16384, multiplier: 2}, dtype: "float8", mesh_shape: "2x2x2", ici_size_range: 8, sharding_strategy: "2x2x1", op_dimension: 1, num_runs: 5}
7+
- {matrix_dim_range: {start: 8, end: 16384, multiplier: 2}, dtype: "float4", mesh_shape: "2x2x2", ici_size_range: 8, sharding_strategy: "2x2x1", op_dimension: 1, num_runs: 5}
8+
- {matrix_dim_range: {start: 8, end: 16384, multiplier: 2}, dtype: "bfloat16", mesh_shape: "2x2x2", ici_size_range: 8, sharding_strategy: "2x2x1", op_dimension: 1, num_runs: 5}
59
trace_dir: "../microbenchmarks/all_gather_tpu7x_2x2x1"
610
csv_path: "../microbenchmarks/all_gather_tpu7x_2x2x1"
711
xlml_metrics_dir: "../microbenchmarks/all_gather_tpu7x_2x2x1"

‎Ironwood/configs/collectives/all_gather_tpu7x_2x2x2.yaml‎

Lines changed: 4 additions & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -2,6 +2,10 @@ benchmarks:
22
- benchmark_name: all_gather
33
benchmark_sweep_params:
44
- {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}
5+
- {matrix_dim_range: {start: 8, end: 16384, multiplier: 2}, dtype: "float16", mesh_shape: "4x2x2", ici_size_range: 16, sharding_strategy: "4x2x1", op_dimension: 1, num_runs: 5}
6+
- {matrix_dim_range: {start: 8, end: 16384, multiplier: 2}, dtype: "float8", mesh_shape: "4x2x2", ici_size_range: 16, sharding_strategy: "4x2x1", op_dimension: 1, num_runs: 5}
7+
- {matrix_dim_range: {start: 8, end: 16384, multiplier: 2}, dtype: "float4", mesh_shape: "4x2x2", ici_size_range: 16, sharding_strategy: "4x2x1", op_dimension: 1, num_runs: 5}
8+
- {matrix_dim_range: {start: 8, end: 16384, multiplier: 2}, dtype: "bfloat16", mesh_shape: "4x2x2", ici_size_range: 16, sharding_strategy: "4x2x1", op_dimension: 1, num_runs: 5}
59
trace_dir: "../microbenchmarks/all_gather_tpu7x_2x2x2"
610
csv_path: "../microbenchmarks/all_gather_tpu7x_2x2x2"
711
xlml_metrics_dir: "../microbenchmarks/all_gather_tpu7x_2x2x2"

‎Ironwood/configs/collectives/all_gather_tpu7x_2x2x4.yaml‎

Lines changed: 4 additions & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -2,6 +2,10 @@ benchmarks:
22
- benchmark_name: all_gather
33
benchmark_sweep_params:
44
- {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}
5+
- {matrix_dim_range: {start: 16, end: 16384, multiplier: 2}, dtype: "float16", mesh_shape: "4x4x2", ici_size_range: 32, sharding_strategy: "4x4x1", op_dimension: 1, num_runs: 5}
6+
- {matrix_dim_range: {start: 16, end: 16384, multiplier: 2}, dtype: "float8", mesh_shape: "4x4x2", ici_size_range: 32, sharding_strategy: "4x4x1", op_dimension: 1, num_runs: 5}
7+
- {matrix_dim_range: {start: 16, end: 16384, multiplier: 2}, dtype: "float4", mesh_shape: "4x4x2", ici_size_range: 32, sharding_strategy: "4x4x1", op_dimension: 1, num_runs: 5}
8+
- {matrix_dim_range: {start: 16, end: 16384, multiplier: 2}, dtype: "bfloat16", mesh_shape: "4x4x2", ici_size_range: 32, sharding_strategy: "4x4x1", op_dimension: 1, num_runs: 5}
59
trace_dir: "../microbenchmarks/all_gather_tpu7x_2x2x4"
610
csv_path: "../microbenchmarks/all_gather_tpu7x_2x2x4"
711
xlml_metrics_dir: "../microbenchmarks/all_gather_tpu7x_2x2x4"

‎Ironwood/configs/collectives/all_gather_tpu7x_2x4x4.yaml‎

Lines changed: 4 additions & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -2,6 +2,10 @@ benchmarks:
22
- benchmark_name: all_gather
33
benchmark_sweep_params:
44
- {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}
5+
- {matrix_dim_range: {start: 32, end: 16384, multiplier: 2}, dtype: "float16", mesh_shape: "8x4x2", ici_size_range: 64, sharding_strategy: "8x4x1", op_dimension: 1, num_runs: 5}
6+
- {matrix_dim_range: {start: 32, end: 16384, multiplier: 2}, dtype: "float8", mesh_shape: "8x4x2", ici_size_range: 64, sharding_strategy: "8x4x1", op_dimension: 1, num_runs: 5}
7+
- {matrix_dim_range: {start: 32, end: 16384, multiplier: 2}, dtype: "float4", mesh_shape: "8x4x2", ici_size_range: 64, sharding_strategy: "8x4x1", op_dimension: 1, num_runs: 5}
8+
- {matrix_dim_range: {start: 32, end: 16384, multiplier: 2}, dtype: "bfloat16", mesh_shape: "8x4x2", ici_size_range: 64, sharding_strategy: "8x4x1", op_dimension: 1, num_runs: 5}
59
trace_dir: "../microbenchmarks/all_gather_tpu7x_2x4x4"
610
csv_path: "../microbenchmarks/all_gather_tpu7x_2x4x4"
711
xlml_metrics_dir: "../microbenchmarks/all_gather_tpu7x_2x4x4"

‎Ironwood/configs/collectives/all_gather_tpu7x_4x4x4.yaml‎

Lines changed: 4 additions & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -2,6 +2,10 @@ benchmarks:
22
- benchmark_name: all_gather
33
benchmark_sweep_params:
44
- {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}
5+
- {matrix_dim_range: {start: 64, end: 16384, multiplier: 2}, dtype: "float16", mesh_shape: "16x4x2", ici_size_range: 128, sharding_strategy: "16x4x1", op_dimension: 1, num_runs: 10}
6+
- {matrix_dim_range: {start: 64, end: 16384, multiplier: 2}, dtype: "float8", mesh_shape: "16x4x2", ici_size_range: 128, sharding_strategy: "16x4x1", op_dimension: 1, num_runs: 10}
7+
- {matrix_dim_range: {start: 64, end: 16384, multiplier: 2}, dtype: "float4", mesh_shape: "16x4x2", ici_size_range: 128, sharding_strategy: "16x4x1", op_dimension: 1, num_runs: 10}
8+
- {matrix_dim_range: {start: 64, end: 16384, multiplier: 2}, dtype: "bfloat16", mesh_shape: "16x4x2", ici_size_range: 128, sharding_strategy: "16x4x1", op_dimension: 1, num_runs: 10}
59
trace_dir: "../microbenchmarks/all_gather_tpu7x_4x4x4"
610
csv_path: "../microbenchmarks/all_gather_tpu7x_4x4x4"
711
xlml_metrics_dir: "../microbenchmarks/all_gather_tpu7x_4x4x4"

‎Ironwood/configs/collectives/all_gather_tpu7x_4x4x8.yaml‎

Lines changed: 4 additions & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -2,6 +2,10 @@ benchmarks:
22
- benchmark_name: all_gather
33
benchmark_sweep_params:
44
- {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}
5+
- {matrix_dim_range: {start: 128, end: 16384, multiplier: 2}, dtype: "float16", mesh_shape: "32x4x2", ici_size_range: 256, sharding_strategy: "32x4x1", op_dimension: 1, num_runs: 5}
6+
- {matrix_dim_range: {start: 128, end: 16384, multiplier: 2}, dtype: "float8", mesh_shape: "32x4x2", ici_size_range: 256, sharding_strategy: "32x4x1", op_dimension: 1, num_runs: 5}
7+
- {matrix_dim_range: {start: 128, end: 16384, multiplier: 2}, dtype: "float4", mesh_shape: "32x4x2", ici_size_range: 256, sharding_strategy: "32x4x1", op_dimension: 1, num_runs: 5}
8+
- {matrix_dim_range: {start: 128, end: 16384, multiplier: 2}, dtype: "bfloat16", mesh_shape: "32x4x2", ici_size_range: 256, sharding_strategy: "32x4x1", op_dimension: 1, num_runs: 5}
59
trace_dir: "../microbenchmarks/all_gather_tpu7x_4x4x8"
610
csv_path: "../microbenchmarks/all_gather_tpu7x_4x4x8"
711
xlml_metrics_dir: "../microbenchmarks/all_gather_tpu7x_4x4x8"

‎Ironwood/configs/collectives/all_reduce_tpu7x_2x2x1.yaml‎

Lines changed: 4 additions & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -2,6 +2,10 @@ benchmarks:
22
- benchmark_name: psum
33
benchmark_sweep_params:
44
- {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}
5+
- {matrix_dim_range: {start: 8, end: 16384, multiplier: 2}, dtype: "float16", mesh_shape: "2x2x2", ici_size_range: 8, sharding_strategy: "2x2x1", op_dimension: 1, num_runs: 5}
6+
- {matrix_dim_range: {start: 8, end: 16384, multiplier: 2}, dtype: "float8", mesh_shape: "2x2x2", ici_size_range: 8, sharding_strategy: "2x2x1", op_dimension: 1, num_runs: 5}
7+
- {matrix_dim_range: {start: 8, end: 16384, multiplier: 2}, dtype: "float4", mesh_shape: "2x2x2", ici_size_range: 8, sharding_strategy: "2x2x1", op_dimension: 1, num_runs: 5}
8+
- {matrix_dim_range: {start: 8, end: 16384, multiplier: 2}, dtype: "bfloat16", mesh_shape: "2x2x2", ici_size_range: 8, sharding_strategy: "2x2x1", op_dimension: 1, num_runs: 5}
59
trace_dir: "../microbenchmarks/psum_tpu7x_2x2x1"
610
csv_path: "../microbenchmarks/psum_tpu7x_2x2x1"
711
xlml_metrics_dir: "../microbenchmarks/psum_tpu7x_2x2x1"

‎Ironwood/configs/collectives/all_reduce_tpu7x_2x2x2.yaml‎

Lines changed: 4 additions & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -2,6 +2,10 @@ benchmarks:
22
- benchmark_name: psum
33
benchmark_sweep_params:
44
- {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}
5+
- {matrix_dim_range: {start: 8, end: 16384, multiplier: 2}, dtype: "float16", mesh_shape: "4x2x2", ici_size_range: 16, sharding_strategy: "4x2x1", op_dimension: 1, num_runs: 5}
6+
- {matrix_dim_range: {start: 8, end: 16384, multiplier: 2}, dtype: "float8", mesh_shape: "4x2x2", ici_size_range: 16, sharding_strategy: "4x2x1", op_dimension: 1, num_runs: 5}
7+
- {matrix_dim_range: {start: 8, end: 16384, multiplier: 2}, dtype: "float4", mesh_shape: "4x2x2", ici_size_range: 16, sharding_strategy: "4x2x1", op_dimension: 1, num_runs: 5}
8+
- {matrix_dim_range: {start: 8, end: 16384, multiplier: 2}, dtype: "bfloat16", mesh_shape: "4x2x2", ici_size_range: 16, sharding_strategy: "4x2x1", op_dimension: 1, num_runs: 5}
59
trace_dir: "../microbenchmarks/psum_tpu7x_2x2x2"
610
csv_path: "../microbenchmarks/psum_tpu7x_2x2x2"
711
xlml_metrics_dir: "../microbenchmarks/psum_tpu7x_2x2x2"

‎Ironwood/configs/collectives/all_reduce_tpu7x_2x2x4.yaml‎

Lines changed: 4 additions & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -2,6 +2,10 @@ benchmarks:
22
- benchmark_name: psum
33
benchmark_sweep_params:
44
- {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}
5+
- {matrix_dim_range: {start: 16, end: 16384, multiplier: 2}, dtype: "float16", mesh_shape: "4x4x2", ici_size_range: 32, sharding_strategy: "4x4x1", op_dimension: 1, num_runs: 5}
6+
- {matrix_dim_range: {start: 16, end: 16384, multiplier: 2}, dtype: "float8", mesh_shape: "4x4x2", ici_size_range: 32, sharding_strategy: "4x4x1", op_dimension: 1, num_runs: 5}
7+
- {matrix_dim_range: {start: 16, end: 16384, multiplier: 2}, dtype: "float4", mesh_shape: "4x4x2", ici_size_range: 32, sharding_strategy: "4x4x1", op_dimension: 1, num_runs: 5}
8+
- {matrix_dim_range: {start: 16, end: 16384, multiplier: 2}, dtype: "bfloat16", mesh_shape: "4x4x2", ici_size_range: 32, sharding_strategy: "4x4x1", op_dimension: 1, num_runs: 5}
59
trace_dir: "../microbenchmarks/psum_tpu7x_2x2x4"
610
csv_path: "../microbenchmarks/psum_tpu7x_2x2x4"
711
xlml_metrics_dir: "../microbenchmarks/psum_tpu7x_2x2x4"

0 commit comments

Comments
 (0)