Conversation
🔗 Helpful Links🧪 See artifacts and rendered test results at hud.pytorch.org/pr/pytorch/executorch/16890
Note: Links to docs will display an error until the docs builds have been completed. ❌ 2 New Failures, 81 PendingAs of commit 89d2c80 with merge base f7dfcf4 ( NEW FAILURES - The following jobs have failed:
This comment was automatically generated by Dr. CI and updates every 15 minutes. |
c7ae74d to
8b4a4f9
Compare
This PR needs a
|
There was a problem hiding this comment.
Pull request overview
This PR relaxes the validation check for LoRA foundation weight files from a byte-for-byte comparison using diff -q to a size-only comparison. The change is motivated by intermittent test failures on CI that may be related to file truncation or memory issues, as referenced in PR #16876.
Changes:
- Replace
diff -qcommand with size comparison usingstatfor checking foundation weight files - Add TODO comment acknowledging that proper content validation (deserializing and checking fields) should be implemented in the future
💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.
| # Confirm that qwen_foundation_lora_q.ptd and qwen_foundation_q.ptd are the same size. | ||
| # TODO(lfq): confirm they are the same (deserialize and check fields) | ||
| size1=$(stat -c %s qwen_foundation_lora_q.ptd) | ||
| size2=$(stat -c %s qwen_foundation_q.ptd) | ||
|
|
||
| if [ "$size1" -eq "$size2" ]; then | ||
| echo "qwen_foundation_lora_q.ptd and qwen_foundation_q.ptd are the same size: $size1." | ||
| else | ||
| echo "qwen_foundation_lora_q.ptd and qwen_foundation_q.ptd are not identical." | ||
| ls -la qwen_foundation_lora_q.ptd | ||
| ls -la qwen_foundation_q.ptd | ||
| cleanup_files | ||
| exit 1 | ||
| echo "qwen_foundation_lora_q.ptd and qwen_foundation_q.ptd have different sizes." | ||
| ls -la qwen_foundation_lora_q.ptd | ||
| ls -la qwen_foundation_q.ptd | ||
| cleanup_files | ||
| exit 1 | ||
| fi |
There was a problem hiding this comment.
Relaxing the file comparison from byte-for-byte equality (using diff -q) to only size equality significantly weakens the validation. Files of the same size can have completely different content. While the TODO comment acknowledges this should be improved, this change means the test will pass even if the foundation weights are corrupted or different, as long as they're the same size. Consider whether this adequately addresses the root cause of the intermittent failures, or if it's merely masking a real problem. If file truncation or corruption is suspected (as mentioned in PR #16876), a size-only check won't catch that if both files are truncated to the same size.
8b4a4f9 to
89d2c80
Compare
|
@pytorchbot cherry-pick --onto release/1.1 -c docs |
|
❌ 🤖 pytorchbot command failed: Try |
|
@pytorchbot cherry-pick --onto release/1.1 -c docs |
Cherry picking #16890The cherry pick PR is at #16893 The following tracker issues are updated: Details for Dev Infra teamRaised by workflow job |
see: #16876
still seeing intermittent failures here.