test: optimize example test discovery and execution speed #372
+565
−21
Conversation
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Merge ProtectionsYour pull request matches the following merge protections and will not be merged until they are valid. 🟢 Enforce conventional commitWonderful, this rule succeeded.Make sure that we follow https://www.conventionalcommits.org/en/v1.0.0/
|
Contributor
|
The PR description has been updated. Please fill out the template for your PR to be reviewed. |
Contributor
Author
|
There's a few things to clarify from this PR
|
Contributor
Author
|
Here's an example output run after these fixes, just using the default This was run on a 32GB macbook m1 max - and takes around 4 minutes |
Member
|
Here's the results from a run on my 32GB RAM / 4GB VRAM Thinkpad |
Member
|
I don't think my issues are blocking, so I'm inclined to approve / merge but will wait a bit in case there are other opinions. |
Contributor
Author
-> A parsing error in the check. I'll fix. The URL is set in my environment
We have a number of models used in our tests. I looked at this one for validation as I didn't have it, but actually we'd need to do something on every test to check the model exists. Docs/upfront check can easily get out of date unless we can do it programatically. -> Suggest discussion, then additional issue/pr if action needed and multiple ones like are down to insufficient resources. 4GB vram is low for AI dev (though we'd always want to drive down min requirements for usage). The error at least is clear. A possible mitigation would be more fine grained markers to document memory usage - perhaps resulting in skipping all llm tests? Another alternative might be to allow cpu only (not gpu) -> Similar - let's discuss and plan |
planetf1
force-pushed
the
feat/optimize-example-test-discovery
branch
from
c43a53b to
1540a3c
Compare
Contributor
Author
|
The issue with watsonx should be fixed now. |
Member
|
I did not have the qwen model downloaded. Pulled it and trying again just to see what happens |
Contributor
Author
ollama probably intelligently split the layers across gpu and cpu. I have marked the test as large ram, as when I ran it the system was struggling somewhat since the model was ~16GB ram by itself. |
Member
qwen2.5vl:7b ? For me it's only 6GB (we're starting to approach building cross-platform container images levels of fun here 😄 ) |
psschwei
reviewed
Jan 28, 2026
ajbozarth
reviewed
Jan 28, 2026
planetf1
added a commit
to planetf1/mellea
that referenced
this pull request
Jan 29, 2026
…s by default Addresses PR generative-computing#372 review feedback: 1. Keep examples clean by using comment-based pytest markers - Examples now use single-line `# pytest: marker1, marker2` format - No pytest imports or pytestmark assignments in example files - Added pytest hooks to parse markers and skip heavy examples early - Prevents memory exhaustion from HuggingFace imports during collection 2. Run qualitative tests by default - Removed `-m "not qualitative"` from pytest addopts - Users can opt-out with: `pytest -m "not qualitative"` (~2 min) - Default `pytest` now runs full suite including quality checks Changes: - docs/examples/conftest.py: +266 lines for comment marker parsing - docs/examples/*.py: All 61 examples restored clean with single comment marker - pyproject.toml: Removed qualitative exclusion, added explanatory comments - AGENTS.md: Updated documentation to reflect new defaults All examples remain pure Python without test infrastructure pollution.
Contributor
|
@planetf1 I think you accidentally committed a bunch of temp files in your last commit |
planetf1
added a commit
to planetf1/mellea
that referenced
this pull request
Jan 29, 2026
- Add 'slow' marker for tests >5 minutes (e.g., dataset loading) - Skip slow tests by default via pyproject.toml addopts - Mark generative_gsm8k.py with slow marker - Exclude conftest.py from test collection - Update all documentation (AGENTS.md, README.md, test/MARKERS_GUIDE.md, PR_372_SUMMARY.md) Addresses PR generative-computing#372 review feedback: - Slow tests (like GSM8K) now skipped by default for fast experience - Qualitative tests still run by default (comprehensive) - conftest.py no longer collected as a test Default: pytest runs qualitative tests, skips slow (~4-6 min) Fast: pytest -m 'not qualitative' (~2 min) Slow: pytest -m slow All: pytest --co -q (bypass config)
Contributor
Author
ah. Fixed Still Refining the execution to ensure we can run all tests |
psschwei
reviewed
Jan 29, 2026
Contributor
Author
Test Suite Timing SummaryQuick Reference
Recommended Workflows
Key Insights
|
- Add pytest markers to example files for proper categorization - Wrap heavy examples with complex initialization to prevent collection hangs - Fix critical skip logic bug in example conftest (import failure) - Configure qualitative tests to be excluded by default - Update docs Tests now run significantly faster and reliably on constrained systems. Run full suite with 'pytest -m ""'.
- Modified 58 example files to conditionally import pytest - Allows examples to run standalone without pytest dependency - Maintains pytest marker functionality when run via pytest
- USDA website returns 403 Forbidden (likely rate limiting/user-agent filtering) - External network dependency makes test flaky - Marked as qualitative to exclude from default test runs
- Fix vision_openai_examples.py: simplified skip logic, added requirements docstring - Fix langchain_messages.py: enhanced conftest to detect pytest.skip() in subprocess - Fix test_vision_openai.py: added @pytest.mark.qualitative decorator - Configure pytest to run test/ before docs/ for fail-fast behavior All tests now pass/skip/deselect correctly with default configuration.
- Remove f-string formatting that converts None to 'None' string - Allows proper None handling by ibm_watsonx_ai library - Fixes WMLClientError when WATSONX_URL env var not set
- Add @pytest.mark.watsonx and @pytest.mark.requires_api_key markers - Ensures test is skipped at collection time when WATSONX_* env vars missing - Prevents misleading error messages about 'None' URLs
- Add @pytest.mark.requires_heavy_ram to vision_openai_examples.py - Model requires ~14GB RAM when loaded (too much for 32GB systems) - Updated docstring to document RAM requirement
…s by default Addresses PR generative-computing#372 review feedback: 1. Keep examples clean by using comment-based pytest markers - Examples now use single-line `# pytest: marker1, marker2` format - No pytest imports or pytestmark assignments in example files - Added pytest hooks to parse markers and skip heavy examples early - Prevents memory exhaustion from HuggingFace imports during collection 2. Run qualitative tests by default - Removed `-m "not qualitative"` from pytest addopts - Users can opt-out with: `pytest -m "not qualitative"` (~2 min) - Default `pytest` now runs full suite including quality checks Changes: - docs/examples/conftest.py: +266 lines for comment marker parsing - docs/examples/*.py: All 61 examples restored clean with single comment marker - pyproject.toml: Removed qualitative exclusion, added explanatory comments - AGENTS.md: Updated documentation to reflect new defaults All examples remain pure Python without test infrastructure pollution.
- Add 'slow' marker for tests >5 minutes (e.g., dataset loading) - Skip slow tests by default via pyproject.toml addopts - Mark generative_gsm8k.py with slow marker - Exclude conftest.py from test collection - Update all documentation (AGENTS.md, README.md, test/MARKERS_GUIDE.md, PR_372_SUMMARY.md) Addresses PR generative-computing#372 review feedback: - Slow tests (like GSM8K) now skipped by default for fast experience - Qualitative tests still run by default (comprehensive) - conftest.py no longer collected as a test Default: pytest runs qualitative tests, skips slow (~4-6 min) Fast: pytest -m 'not qualitative' (~2 min) Slow: pytest -m slow All: pytest --co -q (bypass config)
The example uses IBM_GRANITE_3_3_8B (8B model) which causes memory pressure on 32GB systems. Adding requires_heavy_ram marker to skip on systems with <48GB RAM. This example: - Downloads PDF from arXiv - Processes with Docling - Uses 8B model in Ollama - Runs 5 iterations Fixes test timeout at 97% on 32GB systems.
Removed files that were accidentally committed: - test_timings_non_qualitative.txt - test_full_run.txt - test_verification.txt - ISSUE_*.md files (debugging notes) - PR_SUMMARY.md (duplicate of PR_372_SUMMARY.md) These were temporary files used during development and should not be part of the PR.
These scripts were used during development to convert examples to comment-based markers but should not be part of the PR: - add_example_markers.py - convert_examples_to_comments.py - fix_pytest_imports.py - wrap_heavy_examples.py
PR_372_SUMMARY.md was a temporary summary document and should not be part of the PR. All relevant information is in the PR description and commit messages.
…ples with optional dependency detection - Add automatic detection for optional dependencies (langchain_core) - Fix import error in context_example.py (stdlib.base -> stdlib.context) - Add requires_heavy_ram marker to sofai_graph_coloring.py - Add 15-minute timeout to pytest configuration - Remove langchain_messages.py from manual skip list (now auto-detected) Resolves test failures while keeping examples clean with only marker comments.
planetf1
force-pushed
the
feat/optimize-example-test-discovery
branch
from
1cec30c to
4b1b773
Compare
Contributor
|
Running the tests myself and thought I'd share my version of your table:
|
Contributor
|
I updated my above table. I'm unclear on what is causing the skipped, deselected, and xpassed tests and if the warnings are just me or happening to everyone. I'll be taking a final look through the code changes after lunch. |
Member
|
Mine:
|
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
3 participants
Add this suggestion to a batch that can be applied as a single commit.
This suggestion is invalid because no changes were made to the code.
Suggestions cannot be applied while the pull request is closed.
Suggestions cannot be applied while viewing a subset of changes.
Only one suggestion per line can be applied in a batch.
Add this suggestion to a batch that can be applied as a single commit.
Applying suggestions on deleted lines is not supported.
You must change the existing code in this line in order to create a valid suggestion.
Outdated suggestions cannot be applied.
This suggestion has been applied or marked resolved.
Suggestions cannot be applied from pending reviews.
Suggestions cannot be applied on multi-line comments.
Suggestions cannot be applied while the pull request is queued to merge.
Suggestion cannot be applied right now. Please check back later.
Misc PR
Type of PR
Description
This PR optimizes the test infrastructure to improve test execution speed, reliability, and developer experience. The changes address test collection hangs, improve skip handling, and establish proper test categorization.
Key Changes:
Example Test Discovery Optimization (e1701c1)
@pytest.mark.ollama,@pytest.mark.qualitative, etc.) to 66 example filesdocs/examples/conftest.py(import failure handling)pytest -m ""for full suite)AGENTS.md,README.md,docs/tutorial.md,test/MARKERS_GUIDE.mdStandalone Example Execution (1cd9c7b)
Qualitative Test Marking (c5e36ef)
docs/examples/rag/mellea_pdf.pyas qualitative due to external PDF dependencyTest Failure Fixes (f03581e)
vision_openai_examples.py: Simplified skip logic, added requirements docstringdocs/examples/conftest.py: Detect pytest.skip() exceptions in subprocess stderrtest_vision_openai.py::test_image_block_in_chat: Added@pytest.mark.qualitativedecoratortestpaths = ["test", "docs"]inpyproject.tomlfor fail-fast behaviorImpact:
pytest -m ""Testing