Repository navigation
Refactor Dynamo benchmark parsing and phase resolution - #1124
podkidyshev wants to merge 1 commit into
Conversation
|
No actionable comments were generated in the recent review.
|
| Layer / File(s) | Summary |
|---|---|
Resolve AIPerf phase configuration src/cloudai/workloads/ai_dynamo/ai_dynamo.py, src/cloudai/workloads/ai_dynamo/slurm_command_gen_strategy.py |
AIDynamoCmdArgs.resolve_aiperf_phase merges phase overrides into the base configuration. For multiple phases, missing artifact directories and report names receive phase-specific defaults. The Slurm strategy delegates phase resolution to this method. |
Parse and normalize benchmark results src/cloudai/workloads/ai_dynamo/ai_dynamo.py, src/cloudai/workloads/ai_dynamo/report_generation_strategy.py |
AIDynamoTestDefinition.parse_benchmark_csv parses CSV metrics into AIDynamoBenchReport. It returns None for read or CSV errors, or when required mean TTFT, mean TPOT, or throughput values are missing. The report-generation strategy passes the selected CSV and maximum concurrency to this method. |
Estimated code review effort: 3 (Moderate) | ~20 minutes
Change: Refactor
Possibly related PRs
- NVIDIA/cloudai#1018: Adds AI Dynamo multi-node execution and AIPerf multi-phase scenarios, which this change builds on by moving phase resolution into
AIDynamoCmdArgs. - NVIDIA/cloudai#1121: Moves serving benchmark parsing into test definitions and adds shared LLM-serving report infrastructure, which this change applies to AI Dynamo.
Merge Risk: 
Minimal · up to 21a86
-
No actionable merge-blocking issue was identified.
Pre-merge checks: 
4
-
Full details: Title check
-
Explanation
The title names the two main themes of the changeset: moving Dynamo benchmark CSV parsing and the benchmark result model into the test-definition module, and sharing AIPerf phase resolution between execution and reporting. It is concise and specific, and a reviewer scanning history can identify the primary change.
Full details: Description check
-
Explanation
The description states that the Dynamo CSV parser and benchmark result model move into the test-definition module and that AIPerf phase resolution is shared between execution and reporting. This matches the raw summary, which shows parse_benchmark_csv, AIDynamoBenchReport, and resolve_aiperf_phase changes. It also covers the unchanged behavior and test plan.
-
Finishing Touches
-
📝 Generate docstrings
-
- Commit to this branch
- Create a new PR
🧪 Generate unit tests (beta)
-
- Commit to this branch
- Create a new PR
-
- Autofix · Keep fixing CodeRabbit findings and required CI, and resolving merge conflicts
Comment @coderabbitai help to get the list of available commands.
|
/build |
47c92ab to
c4a19a4
Compare
|
/build |
c4a19a4 to
349a225
Compare
Signed-off-by: Ivan Podkidyshev <ipodkidyshev@nvidia.com>
349a225 to
21a8615
Compare
|
/build |
Summary
Move the existing Dynamo CSV parser and benchmark result model into the test-definition module. Share AIPerf phase resolution between execution and reporting so the following Pareto PR can reuse it. Existing parsing requirements, phase overrides, and artifact paths stay unchanged.
Test Plan
Additional Notes
Stacked on #1121. The Dynamo Pareto feature in #1115 stacks on this PR. All branches include the latest main and its CI fixes.