Skip to content

fix: support run_on_local_node in gemm_throttling #130

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

Merged
linamy85 merged 1 commit into from
Aug 20, 2026
Merged

fix: support run_on_local_node in gemm_throttling #130

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
9 changes: 7 additions & 2 deletions Ironwood/src/benchmark_gemm_throttling.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -46,6 +46,7 @@ def gemm_throttling(
dtype: jnp.dtype = jax.numpy.float8_e4m3fn,
gap_strategy: str = "data_gen_every_iter_block_every_iter",
trace_dir: str = None,
run_on_local_node: bool = False,
Comment thread
simrankaurb marked this conversation as resolved.
) -> Dict[str, Any]:
"""Benchmarks the OUT<M, N>:BF16 = IN0<M, K>:FP8 x IN1<N, K>:FP8.

Expand All @@ -59,7 +60,7 @@ def f(x, y):
)
return acc.astype(jnp.bfloat16)

mesh = create_mesh(SHARDING_STRATEGY)
mesh = create_mesh(SHARDING_STRATEGY, local_mesh=run_on_local_node)
lhs_sharding = get_lhs_named_shading(mesh, SHARDING_STRATEGY)
rhs_sharding = get_rhs_named_shading(mesh, SHARDING_STRATEGY)
out_sharding = get_out_sharding(SHARDING_STRATEGY)
Expand Down Expand Up @@ -120,12 +121,16 @@ def gemm_throttling_calculate_metrics(
gap_strategy: str,
dtype: jnp.dtype,
time_ms_list: list[float],
run_on_local_node: bool = False,
) -> Dict[str, Any]:
# pylint: disable=unused-argument
# Calculate FLOPs
total_flops = 2 * m * k * n # Total floating-point operations
device_count = (
jax.local_device_count() if run_on_local_node else jax.device_count()
)
total_flops, total_flops_all_devices = handle_based_on_sharding(
total_flops, SHARDING_STRATEGY
total_flops, SHARDING_STRATEGY, device_count=device_count
)
return unified_flops_metrics(
m,
Expand Down
Loading