[llm-d] Work on the KPI generation#100
Conversation
|
[APPROVALNOTIFIER] This PR is NOT APPROVED This pull-request has been approved by: The full list of commands accepted by this bot can be found here. DetailsNeeds approval from an approver in each of these files:Approvers can indicate their approval by writing |
|
Note Reviews pausedIt looks like this branch is under active development. To avoid overwhelming you with review comments due to an influx of new commits, CodeRabbit has automatically paused this review. You can configure this behavior by changing the Use the following commands to manage reviews:
Use the checkboxes below for quick actions:
📝 WalkthroughWalkthroughThis PR adds MLflow web-UI URL-based artifact import to the Caliper CLI, introduces a KPI decorator/discovery/catalog framework, refactors GuideLLM KPI computation and adds HTML/CSV KPI reporting, reworks postprocess orchestration into a class-based flow with step logging and AI-eval artifact export, changes replot's download path to shell out via CLI, and renames a shared CI decorator. ChangesCaliper and GuideLLM pipeline changes
Estimated code review effort: 4 (Complex) | ~75 minutes Postprocess orchestration, step logging, and AI-eval export
Estimated code review effort: 4 (Complex) | ~60 minutes Replot MLflow download refactor
Estimated code review effort: 3 (Moderate) | ~30 minutes CI entrypoint decorator rename
Estimated code review effort: 2 (Simple) | ~15 minutes KServe deployment capture
Estimated code review effort: 1 (Trivial) | ~5 minutes 🚥 Pre-merge checks | ✅ 5✅ Passed checks (5 passed)
✨ Finishing Touches🧪 Generate unit tests (beta)
Thanks for using CodeRabbit! It's free for OSS, and your support helps us grow. If you like it, consider giving us a shout-out. Comment |
|
/test fournos llm_d smoke |
🟢 Execution of
|
🟢 Submission of
|
|
/test fournos llm_d smoke |
🟢 Execution of
|
🟢 Submission of
|
|
/test fournos llm_d smoke |
![coderabbitai[bot]](https://avatars.githubusercontent.com/in/347564?s=60&v=4){kind=small}
There was a problem hiding this comment.
Actionable comments posted: 7
🧹 Nitpick comments (1)
projects/caliper/engine/kpi/decorators.py (1)
166-173: 📐 Maintainability & Code Quality | 🔵 Trivial | ⚡ Quick winDrop the hardcoded KPI name prefixes from discovery.
This engine-level helper only finds decorated functions whose names start with
guidellm_,matbench_, orkpi_, so a correctly decorated KPI silently disappears if it uses any other name. Prefer filtering by decorator metadata plusobj.__module__ == module.__name__instead of encoding downstream plugin prefixes here.♻️ Possible direction
- for name, obj in inspect.getmembers(module, inspect.isfunction): - if ( - name.startswith(("guidellm_", "matbench_", "kpi_")) - and hasattr(obj, "_kpi_help") - and hasattr(obj, "_kpi_unit") - and hasattr(obj, "_kpi_higher_is_better") - ): - kpi_functions[name] = obj + for name, obj in inspect.getmembers(module, inspect.isfunction): + if ( + obj.__module__ == module.__name__ + and hasattr(obj, "_kpi_help") + and hasattr(obj, "_kpi_unit") + and hasattr(obj, "_kpi_higher_is_better") + ): + kpi_functions[name] = obj🤖 Prompt for AI Agents
Verify each finding against current code. Fix only still-valid issues, skip the rest with a brief reason, keep changes minimal, and validate. In `@projects/caliper/engine/kpi/decorators.py` around lines 166 - 173, Remove the hardcoded name-prefix check from the KPI discovery loop in the module-scanning helper so decorated KPIs are found regardless of function name. Update the logic in the inspect.getmembers-based scan to rely on the decorator metadata fields (_kpi_help, _kpi_unit, _kpi_higher_is_better) and ensure the function belongs to the current module via obj.__module__ == module.__name__ before adding it to kpi_functions.
🤖 Prompt for all review comments with AI agents
Verify each finding against current code. Fix only still-valid issues, skip the
rest with a brief reason, keep changes minimal, and validate.
Inline comments:
In `@projects/caliper/cli/main.py`:
- Around line 209-214: The secrets-based tracking URI is only applied inside the
verbose logging path, so `MLFLOW_TRACKING_URI` and
`MLFLOW_TRACKING_INSECURE_TLS` from `--mlflow-secrets` should be processed
regardless of `-v`. Update the flow in `main()` so the connection settings
(`final_tracking_uri`, `final_insecure_tls`) are populated before any
verbose-only exit/logic, and make sure the secrets handling branch that reads
`MLFLOW_TRACKING_URI` and `MLFLOW_TRACKING_INSECURE_TLS` is not gated by
`click.echo` or `verbose` state.
- Around line 58-60: The endpoint construction in main.py is stripping the path
from parsed URLs, which breaks path-prefixed MLflow tracking links. Update the
endpoint logic around the parsed.scheme/parsed.netloc handling to preserve any
non-empty parsed.path (while still removing fragment/query as needed) so URLs
like https://host/mlflow/#/... continue to point to the correct tracking URI.
- Around line 1037-1041: The --timeout option in main.py currently allows 0 or
negative values, which can disable the alarm and remove the download timeout.
Update the click.option definition for --timeout to enforce a minimum of 1,
using the existing timeout option in the CLI entrypoint so invalid values are
rejected before reaching the alarm setup.
In `@projects/guidellm/postprocess/guidellm/parsing/kpis.py`:
- Around line 137-145: The fallback in guidellm_throughput_curve() is
synthesizing fake 2D KPI points from request_rate when throughput_curve is
missing, which makes GuideLLMKpiHandler.compute_kpis() and the HTML report treat
invented data as real measurements. Remove the scalar-derived fallback in the
KPI parsing path (including the duplicate logic around the other referenced
section) and return an empty list whenever unified_record.metrics does not
contain throughput_curve, keeping only the real curve parsing from the existing
tuple data.
In `@projects/guidellm/postprocess/guidellm/plotting/kpi_report.py`:
- Around line 30-40: The KPI report is incorrectly treating a multi-record input
as a single record by reading only records[0] in the label/KPI assembly path.
Update GuideLLM KPI report generation to either explicitly aggregate/group the
records before building labels and values, or raise an error when more than one
record is passed; make the change in the code path that uses first_record,
LABEL_EXTRACTOR.extract, and all_labels so the output cannot silently ignore
extra records.
- Line 133: The HTML generation in kpi_report.py is interpolating untrusted
benchmark fields directly into the report, so escape every dynamic value before
rendering. Update the code around the report_title interpolation and the other
HTML builders that use labels, KPI descriptions, and test metadata so they all
pass through the same escaping helper before being inserted into f-strings. Use
the existing report rendering functions and any shared formatting helpers in
kpi_report.py to apply this consistently across all affected sections.
- Around line 64-80: The KPI rendering in kpi_report is re-deriving precision
instead of honoring the decorator metadata, so update the report generation and
2D rendering paths to use the format strings already attached to the KPI
function (_kpi_format, _kpi_x_format, and _kpi_y_format). In the kpi_record
assembly and the HTML/value formatting logic that consumes it, preserve and
apply those declared formats from the KPI function rather than forcing decimal
places, including the 2D-specific branches for x and y values.
---
Nitpick comments:
In `@projects/caliper/engine/kpi/decorators.py`:
- Around line 166-173: Remove the hardcoded name-prefix check from the KPI
discovery loop in the module-scanning helper so decorated KPIs are found
regardless of function name. Update the logic in the inspect.getmembers-based
scan to rely on the decorator metadata fields (_kpi_help, _kpi_unit,
_kpi_higher_is_better) and ensure the function belongs to the current module via
obj.__module__ == module.__name__ before adding it to kpi_functions.
🪄 Autofix (Beta)
Fix all unresolved CodeRabbit comments on this PR:
- Push a commit to this branch (recommended)
- Create a new PR with the fixes
ℹ️ Review info
⚙️ Run configuration
Configuration used: defaults
Review profile: CHILL
Plan: Pro
Run ID: 6f5ba352-2848-4e1b-9740-70a85402604b
📒 Files selected for processing (7)
projects/caliper/cli/main.pyprojects/caliper/engine/kpi/__init__.pyprojects/caliper/engine/kpi/decorators.pyprojects/guidellm/postprocess/guidellm/parsing/kpis.pyprojects/guidellm/postprocess/guidellm/plotting/kpi_report.pyprojects/guidellm/postprocess/guidellm/plugin.pyprojects/guidellm/postprocess/guidellm/visualize-groups.yaml
🟢 Execution of
|
🟢 Submission of
|
kpouget
force-pushed
the
llmd
branch
2 times, most recently
from
73e4ca4 to
b1bc7f7
Compare
|
/test fournos llm_d smoke |
There was a problem hiding this comment.
Actionable comments posted: 9
Caution
Some comments are outside the diff and can’t be posted inline due to platform limitations.
⚠️ Outside diff range comments (1)
projects/caliper/orchestration/step_logging.py (1)
1-422: 📐 Maintainability & Code Quality | 🟡 Minor | ⚡ Quick winRun
ruff formaton this file before merge.The current formatting is already failing both Ruff format checks in CI.
🤖 Prompt for AI Agents
Verify each finding against current code. Fix only still-valid issues, skip the rest with a brief reason, keep changes minimal, and validate. In `@projects/caliper/orchestration/step_logging.py` around lines 1 - 422, The file’s formatting is not Ruff-compliant, so run ruff format on step_logging.py and keep the resulting style changes only. Focus on preserving the existing behavior in functions like _get_next_step_index, step_logging_indexed, step_logging, and the log_*_command helpers while aligning spacing, wrapping, and line breaks with Ruff’s formatter so CI passes.Source: Pipeline failures
🤖 Prompt for all review comments with AI agents
Verify each finding against current code. Fix only still-valid issues, skip the
rest with a brief reason, keep changes minimal, and validate.
Inline comments:
In `@projects/caliper/orchestration/postprocess_config.py`:
- Around line 81-84: The csv_export output field currently allows null in
postprocess_config, which later breaks _run_kpi_csv_export() when it combines
output_dir with postprocess_config.kpi.csv_export.output. Update the
CsvExport/PostprocessConfig field definition so csv_export.output is required to
be a non-null string, keeping the default of "kpis.csv" and removing any None
allowance from the type or validation.
In `@projects/caliper/orchestration/postprocess.py`:
- Around line 583-600: The HTML report generation in the postprocess flow is
using the optional visualize path instead of the resolved visualize output
directory, so the final pages can be skipped when callers rely on
caliper.postprocess.visualize.output_dir. Update the report-writing block in
postprocess.py to use the resolved output directory chosen by run_visualize(),
and pass that resolved path into generate_caliper_reports_index and
generate_postprocess_status_report so both reports are always written to the
actual visualization output location.
- Around line 228-230: The CSV export path in postprocess output handling is not
sandboxed, so absolute paths or path traversal can escape the artifact
directory. Update the output path logic in the postprocess flow around
output_file creation to resolve the configured CSV export path against
output_dir and enforce that the final path stays within that directory, using
the existing postprocess_config.kpi.csv_export.output and output_file handling.
Reject or normalize any config value that would write outside the postprocess
output dir.
- Around line 503-510: The KPI CSV export step in postprocess.py is logging an
input path from postprocess_config.kpi.generate.output even when the KPI
generate step did not run or failed, so adjust the _run_kpi_csv_export call path
handling to only reference a real JSONL file. Use the existing identifiers
step_logging, kpi_csv_export, and postprocess_config.kpi.generate to determine
whether the generate output exists; if it does not, avoid emitting a
reproduction banner with a fake --input value and instead pass a valid created
path or skip logging that command.
In `@projects/guidellm/postprocess/guidellm/csv_export/kpi_csv_exporter.py`:
- Around line 49-53: The KPI CSV export flow in KpiCsvExporter currently treats
2D records as scalar data when include_2d_kpis is enabled, which silently
corrupts the export. Update the 2D handling branch in the exporter logic to fail
fast instead of appending those records to scalar_records, and surface a clear
error from the export path so callers know 2D KPI CSV export is not supported
yet. Keep the fix localized to the record-processing code that checks
record.get("is_2d", False) and the subsequent CSV conversion path.
In `@projects/guidellm/postprocess/guidellm/csv_export/kpi_csv_model.py`:
- Around line 205-208: The KPI CSV export header is using the column name
instead of the actual schema version value. Update the header generation in the
KPI CSV model so the Schema Version comment reads from the stored schema version
data rather than looking up the 'schema_version' entry in self.columns, and keep
the change localized to the header-building logic.
In `@projects/guidellm/postprocess/guidellm/parsing/kpis.py`:
- Around line 344-350: `performance_tier` is comparing
`record.metrics.get("total_tok/sec", 0)` directly, so string values can raise
TypeError and break `TestLabelExtractor.extract()` fallback behavior. Update the
KPI derivation in `compute_kpis()`/the `performance_tier` mapping to normalize
`total_tok/sec` to a numeric type before the threshold checks, and make sure any
access through `record.metrics` handles missing or non-numeric values
consistently.
- Around line 245-250: `guidellm_request_latency_max` is incorrectly annotated
as higher-is-better, which reverses the ranking semantics for this latency KPI.
Update the decorators on `guidellm_request_latency_max` in `kpis.py` to use the
lower-is-better equivalent instead of `@HigherBetter()`, while keeping the
existing `@Format("{:.4f}")` and `@KPIMetadata` metadata unchanged so consumers
interpret higher latency as worse.
- Around line 424-432: The KPI record built in compute_kpis is missing decorator
metadata such as help, so create_csv_row_from_kpi_record cannot populate the CSV
kpi_help column. Update the kpi_record assembly in compute_kpis to carry through
the existing decorator fields from kpi_func, especially the help text (and any
other relevant metadata already available on the KPI definition), so downstream
CSV export reads them instead of emitting empty values.
---
Outside diff comments:
In `@projects/caliper/orchestration/step_logging.py`:
- Around line 1-422: The file’s formatting is not Ruff-compliant, so run ruff
format on step_logging.py and keep the resulting style changes only. Focus on
preserving the existing behavior in functions like _get_next_step_index,
step_logging_indexed, step_logging, and the log_*_command helpers while aligning
spacing, wrapping, and line breaks with Ruff’s formatter so CI passes.
🪄 Autofix (Beta)
Fix all unresolved CodeRabbit comments on this PR:
- Push a commit to this branch (recommended)
- Create a new PR with the fixes
ℹ️ Review info
⚙️ Run configuration
Configuration used: defaults
Review profile: CHILL
Plan: Pro
Run ID: 69d184da-e489-49ef-9e9b-27f5dbfc9f81
📒 Files selected for processing (13)
projects/caliper/cli/main.pyprojects/caliper/engine/kpi/__init__.pyprojects/caliper/engine/kpi/decorators.pyprojects/caliper/orchestration/postprocess.pyprojects/caliper/orchestration/postprocess_config.pyprojects/caliper/orchestration/step_logging.pyprojects/guidellm/postprocess/guidellm/csv_export/__init__.pyprojects/guidellm/postprocess/guidellm/csv_export/kpi_csv_exporter.pyprojects/guidellm/postprocess/guidellm/csv_export/kpi_csv_model.pyprojects/guidellm/postprocess/guidellm/parsing/kpis.pyprojects/guidellm/postprocess/guidellm/plotting/kpi_report.pyprojects/guidellm/postprocess/guidellm/plugin.pyprojects/guidellm/postprocess/guidellm/visualize-groups.yaml
✅ Files skipped from review due to trivial changes (1)
- projects/caliper/engine/kpi/init.py
🚧 Files skipped from review as they are similar to previous changes (5)
- projects/guidellm/postprocess/guidellm/plugin.py
- projects/guidellm/postprocess/guidellm/visualize-groups.yaml
- projects/guidellm/postprocess/guidellm/plotting/kpi_report.py
- projects/caliper/engine/kpi/decorators.py
- projects/caliper/cli/main.py
| output: str | None = Field( | ||
| default="kpis.csv", | ||
| description="CSV filename or path; relative paths resolve under the post-processing artifact dir.", | ||
| ) |
There was a problem hiding this comment.
🎯 Functional Correctness | 🟠 Major | ⚡ Quick win
Don't allow csv_export.output to be null.
_run_kpi_csv_export() later does output_dir / postprocess_config.kpi.csv_export.output, so output: null passes validation here and then fails at runtime with TypeError as soon as the step is enabled.
🤖 Prompt for AI Agents
Verify each finding against current code. Fix only still-valid issues, skip the
rest with a brief reason, keep changes minimal, and validate.
In `@projects/caliper/orchestration/postprocess_config.py` around lines 81 - 84,
The csv_export output field currently allows null in postprocess_config, which
later breaks _run_kpi_csv_export() when it combines output_dir with
postprocess_config.kpi.csv_export.output. Update the CsvExport/PostprocessConfig
field definition so csv_export.output is required to be a non-null string,
keeping the default of "kpis.csv" and removing any None allowance from the type
or validation.
| # Create output file path | ||
| output_file = output_dir / postprocess_config.kpi.csv_export.output | ||
| output_file.parent.mkdir(parents=True, exist_ok=True) |
There was a problem hiding this comment.
🔒 Security & Privacy | 🟠 Major | ⚡ Quick win
Keep CSV exports inside the postprocess output dir.
output_dir / postprocess_config.kpi.csv_export.output is not a sandbox: an absolute value discards output_dir, and .. segments still escape it after resolution. A bad config can therefore write outside the artifact tree despite the config contract saying relative paths stay under the post-processing dir.
🤖 Prompt for AI Agents
Verify each finding against current code. Fix only still-valid issues, skip the
rest with a brief reason, keep changes minimal, and validate.
In `@projects/caliper/orchestration/postprocess.py` around lines 228 - 230, The
CSV export path in postprocess output handling is not sandboxed, so absolute
paths or path traversal can escape the artifact directory. Update the output
path logic in the postprocess flow around output_file creation to resolve the
configured CSV export path against output_dir and enforce that the final path
stays within that directory, using the existing
postprocess_config.kpi.csv_export.output and output_file handling. Reject or
normalize any config value that would write outside the postprocess output dir.
| # KPI CSV export (uses the same computed KPIs) | ||
| if postprocess_config.kpi.csv_export.enabled: | ||
| with step_logging("caliper_kpi_csv_export", step_logs_dir): | ||
| # Path to the JSONL file for reference in command logging | ||
| kpi_jsonl_path = output_dir / postprocess_config.kpi.generate.output | ||
| kpi_results["kpi_csv_export"] = _run_kpi_csv_export( | ||
| postprocess_config, plugin, model, output_dir, kpi_jsonl_path | ||
| ) |
There was a problem hiding this comment.
🎯 Functional Correctness | 🟡 Minor | ⚡ Quick win
Don't log a CSV-export input file that was never created.
This always points the reproduction banner at output_dir / postprocess_config.kpi.generate.output, even when kpi.generate was disabled or failed. The logged caliper kpi csv-export --input ... command is then not actually reproducible.
🤖 Prompt for AI Agents
Verify each finding against current code. Fix only still-valid issues, skip the
rest with a brief reason, keep changes minimal, and validate.
In `@projects/caliper/orchestration/postprocess.py` around lines 503 - 510, The
KPI CSV export step in postprocess.py is logging an input path from
postprocess_config.kpi.generate.output even when the KPI generate step did not
run or failed, so adjust the _run_kpi_csv_export call path handling to only
reference a real JSONL file. Use the existing identifiers step_logging,
kpi_csv_export, and postprocess_config.kpi.generate to determine whether the
generate output exists; if it does not, avoid emitting a reproduction banner
with a fake --input value and instead pass a valid created path or skip logging
that command.
| if record.get("is_2d", False): | ||
| if self.include_2d_kpis: | ||
| # TODO: Implement 2D KPI export | ||
| scalar_records.append(record) | ||
| else: |
There was a problem hiding this comment.
🗄️ Data Integrity & Integration | 🟠 Major | ⚡ Quick win
Fail fast when include_2d_kpis=True.
That branch currently appends raw 2D KPI records into a scalar CSV schema. The later conversion just stringifies the point list into one cell and drops the x/y metadata, so callers get a “successful” export with corrupted 2D data.
Proposed fix
if record.get("is_2d", False):
if self.include_2d_kpis:
- # TODO: Implement 2D KPI export
- scalar_records.append(record)
+ raise NotImplementedError("2D KPI CSV export is not implemented yet")
else:
skipped_2d_count += 1📝 Committable suggestion
‼️ IMPORTANT
Carefully review the code before committing. Ensure that it accurately replaces the highlighted code, contains no missing lines, and has no issues with indentation. Thoroughly test & benchmark the code to ensure it meets the requirements.
| if record.get("is_2d", False): | |
| if self.include_2d_kpis: | |
| # TODO: Implement 2D KPI export | |
| scalar_records.append(record) | |
| else: | |
| if record.get("is_2d", False): | |
| if self.include_2d_kpis: | |
| raise NotImplementedError("2D KPI CSV export is not implemented yet") | |
| else: |
🤖 Prompt for AI Agents
Verify each finding against current code. Fix only still-valid issues, skip the
rest with a brief reason, keep changes minimal, and validate.
In `@projects/guidellm/postprocess/guidellm/csv_export/kpi_csv_exporter.py` around
lines 49 - 53, The KPI CSV export flow in KpiCsvExporter currently treats 2D
records as scalar data when include_2d_kpis is enabled, which silently corrupts
the export. Update the 2D handling branch in the exporter logic to fail fast
instead of appending those records to scalar_records, and surface a clear error
from the export path so callers know 2D KPI CSV export is not supported yet.
Keep the fix localized to the record-processing code that checks
record.get("is_2d", False) and the subsequent CSV conversion path.
| header_lines.append("# GuideLLM KPI Export") | ||
| header_lines.append( | ||
| f"# Schema Version: {self.columns[self.columns.index('schema_version')]}" | ||
| ) |
There was a problem hiding this comment.
🎯 Functional Correctness | 🟡 Minor | ⚡ Quick win
Write the actual schema version in the header comment.
This currently emits # Schema Version: schema_version because it reads the column name from self.columns instead of the version value. The exported header is misleading as written.
Proposed fix
- header_lines.append(
- f"# Schema Version: {self.columns[self.columns.index('schema_version')]}"
- )
+ header_lines.append("# Schema Version: 1")📝 Committable suggestion
‼️ IMPORTANT
Carefully review the code before committing. Ensure that it accurately replaces the highlighted code, contains no missing lines, and has no issues with indentation. Thoroughly test & benchmark the code to ensure it meets the requirements.
| header_lines.append("# GuideLLM KPI Export") | |
| header_lines.append( | |
| f"# Schema Version: {self.columns[self.columns.index('schema_version')]}" | |
| ) | |
| header_lines.append("# GuideLLM KPI Export") | |
| header_lines.append("# Schema Version: 1") |
🤖 Prompt for AI Agents
Verify each finding against current code. Fix only still-valid issues, skip the
rest with a brief reason, keep changes minimal, and validate.
In `@projects/guidellm/postprocess/guidellm/csv_export/kpi_csv_model.py` around
lines 205 - 208, The KPI CSV export header is using the column name instead of
the actual schema version value. Update the header generation in the KPI CSV
model so the Schema Version comment reads from the stored schema version data
rather than looking up the 'schema_version' entry in self.columns, and keep the
change localized to the header-building logic.
| "performance_tier": ( | ||
| "high" | ||
| if record.metrics.get("total_tok/sec", 0) > 1000 | ||
| else "medium" | ||
| if record.metrics.get("total_tok/sec", 0) > 100 | ||
| else "low" | ||
| ), |
There was a problem hiding this comment.
🎯 Functional Correctness | 🟠 Major | ⚡ Quick win
Cast total_tok/sec before deriving performance_tier.
If record.metrics["total_tok/sec"] arrives as a string, these > comparisons raise TypeError. TestLabelExtractor.extract() then falls back to {}, which drops every extracted test-condition label for that record in both compute_kpis() and the HTML report.
🤖 Prompt for AI Agents
Verify each finding against current code. Fix only still-valid issues, skip the
rest with a brief reason, keep changes minimal, and validate.
In `@projects/guidellm/postprocess/guidellm/parsing/kpis.py` around lines 344 -
350, `performance_tier` is comparing `record.metrics.get("total_tok/sec", 0)`
directly, so string values can raise TypeError and break
`TestLabelExtractor.extract()` fallback behavior. Update the KPI derivation in
`compute_kpis()`/the `performance_tier` mapping to normalize `total_tok/sec` to
a numeric type before the threshold checks, and make sure any access through
`record.metrics` handles missing or non-numeric values consistently.
🟢 Execution of
|
🔴 Submission of
|
|
/test fournos llm_d smoke |
🟢 Execution of
|
🟢 Submission of
|
|
/test fournos llm_d smoke |
🟢 Execution of
|
🟢 Submission of
|
…the generated files
|
/test fournos llm_d smoke |
🟢 Execution of
|
🟢 Submission of
|
|
/test fournos llm_d smoke |
🟢 Execution of
|
🔴 Submission of
|
|
@kpouget: The following test failed, say
Full PR test history. Your PR dashboard. DetailsInstructions for interacting with me using PR comments are available here. If you have questions or suggestions related to my behavior, please file an issue against the kubernetes-sigs/prow repository. I understand the commands that are listed here. |
1 participant
Summary by CodeRabbit
--keep-downloadoption.