Skip to content

Make lowest precision, parallelism, MBS, and config_filename mandatory for v6.1 #472

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
pavanky merged 2 commits into from
Aug 21, 2026
Merged

Make lowest precision, parallelism, MBS, and config_filename mandatory for v6.1 #472

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
2 commits
Select commit Hold shift + click to select a range
22cc11f
Make lowest precision and parallelism mandatory for v6.1
ShriyaRishab Aug 20, 2026
cb9a472
Also require micro_batch_size and config_filename in v6.1 logs
ShriyaRishab Aug 20, 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
31 changes: 16 additions & 15 deletions mlperf_logging/compliance_checker/training_6.1.0/common.yaml
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -144,48 +144,49 @@
REQ: EXACTLY_ONE
CHECK: " v['value'] != '' "

# Optional keys
# Mandatory precision, parallelism, and run-config disclosure (v6.1+)
- KEY:
NAME: lowest_numerical_precision_in_linear
REQ: OPTIONAL
REQ: EXACTLY_ONE
CHECK: " v['value'] in ['fp64', 'fp32', 'tf32', 'fp16', 'fp8', 'nvfp4', 'mxfp4', 'bfloat16', 'Graphcore FLOAT 16.16', 'int8', 'uint8', 'int4', 'uint4'] "

- KEY:
NAME: lowest_numerical_precision_in_attn
REQ: OPTIONAL
REQ: EXACTLY_ONE
CHECK: " v['value'] in ['fp64', 'fp32', 'tf32', 'fp16', 'fp8', 'nvfp4', 'mxfp4', 'bfloat16', 'Graphcore FLOAT 16.16', 'int8', 'uint8', 'int4', 'uint4'] "

- KEY:
NAME: lowest_numerical_precision_in_comm
REQ: OPTIONAL
REQ: EXACTLY_ONE
CHECK: " v['value'] in ['fp64', 'fp32', 'tf32', 'fp16', 'fp8', 'nvfp4', 'mxfp4', 'bfloat16', 'Graphcore FLOAT 16.16', 'int8', 'uint8', 'int4', 'uint4'] "

- KEY:
NAME: tensor_parallelism
REQ: OPTIONAL
CHECK: " is_integer(v['value']) "
REQ: EXACTLY_ONE
CHECK: " is_integer(v['value']) and v['value'] >= 1 "

- KEY:
NAME: pipeline_parallelism
REQ: OPTIONAL
CHECK: " is_integer(v['value']) "
REQ: EXACTLY_ONE
CHECK: " is_integer(v['value']) and v['value'] >= 1 "

- KEY:
NAME: context_parallelism
REQ: OPTIONAL
CHECK: " is_integer(v['value']) "
REQ: EXACTLY_ONE
CHECK: " is_integer(v['value']) and v['value'] >= 1 "

- KEY:
NAME: expert_parallelism
REQ: OPTIONAL
CHECK: " is_integer(v['value']) "
REQ: EXACTLY_ONE
CHECK: " is_integer(v['value']) and v['value'] >= 1 "

- KEY:
NAME: micro_batch_size
REQ: OPTIONAL
CHECK: " is_integer(v['value']) "
REQ: EXACTLY_ONE
CHECK: " is_integer(v['value']) and v['value'] >= 1 "

- KEY:
NAME: config_filename
REQ: OPTIONAL
REQ: EXACTLY_ONE
CHECK: " v['value'] != '' "

5 changes: 5 additions & 0 deletions mlperf_logging/mllog/constants.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -209,6 +209,11 @@
CONVERTION_EFF = "conversion_eff"
INTERCONNECT_POWER_EST = "interconnect_power_est"

# Precision constants
LOWEST_NUMERICAL_PRECISION_IN_ATTN = "lowest_numerical_precision_in_attn"
LOWEST_NUMERICAL_PRECISION_IN_COMM = "lowest_numerical_precision_in_comm"
LOWEST_NUMERICAL_PRECISION_IN_LINEAR = "lowest_numerical_precision_in_linear"

# Parallelism constants
TENSOR_PARALLELISM = "tensor_parallelism"
PIPELINE_PARALLELISM = "pipeline_parallelism"
Expand Down
11 changes: 8 additions & 3 deletions mlperf_logging/mllog/examples/parallelism.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -19,7 +19,7 @@


def parallelism_example():
"""Example usage of mllog with parallelism and config keys"""
"""Example usage of mllog with mandatory precision, parallelism, and config keys"""

mllogger = mllog.get_mllogger()

Expand All @@ -33,10 +33,15 @@ def parallelism_example():

mllogger.start(key=mllog.constants.RUN_START)

# Log the model config file used for this run
# Log the model config file used for this run. The name must match the submitted config.
mllogger.event(key=mllog.constants.CONFIG_FILENAME, value="llama31_405b_config.yaml")

# Log parallelism strategy
# Log lowest numerical precision used in linear, attention, and communication
mllogger.event(key=mllog.constants.LOWEST_NUMERICAL_PRECISION_IN_LINEAR, value="fp8")
mllogger.event(key=mllog.constants.LOWEST_NUMERICAL_PRECISION_IN_ATTN, value="bfloat16")
mllogger.event(key=mllog.constants.LOWEST_NUMERICAL_PRECISION_IN_COMM, value="fp8")

# Log parallelism strategy. Unused dimensions must still be logged as 1.
mllogger.event(key=mllog.constants.TENSOR_PARALLELISM, value=8)
mllogger.event(key=mllog.constants.PIPELINE_PARALLELISM, value=4)
mllogger.event(key=mllog.constants.CONTEXT_PARALLELISM, value=2)
Expand Down
Loading