I am trying to submit a TorchScript model (.pt file) for the CSIRO Biomass competition. My pipeline works perfectly on the Verification (Public) test set (predicts 1 image, generates valid CSV). However, it consistently fails on the Hidden Test Set with Notebook Threw Exception.
\nWhat I have tried:
\nThe Problem\nPublic Run: Success. submission.csv is generated with valid floats.\nSubmission: Fails. Notebook Threw Exception (likely non-zero exit code). Since standard logs are not available for failed submissions, I am stuck.
\nLink to public notebook: https://www.kaggle.com/code/aditya1711/csiro-p2-19-ref-logic/notebook?scriptVersionId=292320463
", "messages": [ { "id": "3392563", "postDate": "01/17/2026 06:23:58", "content": "I am trying to submit a TorchScript model (.pt file) for the CSIRO Biomass competition. My pipeline works perfectly on the Verification (Public) test set (predicts 1 image, generates valid CSV). However, it consistently fails on the Hidden Test Set with Notebook Threw Exception.
\nWhat I have tried:
\nThe Problem\nPublic Run: Success. submission.csv is generated with valid floats.\nSubmission: Fails. Notebook Threw Exception (likely non-zero exit code). Since standard logs are not available for failed submissions, I am stuck.
\nLink to public notebook: https://www.kaggle.com/code/aditya1711/csiro-p2-19-ref-logic/notebook?scriptVersionId=292320463
", "rawMarkdown": "I am trying to submit a TorchScript model (.pt file) for the CSIRO Biomass competition. My pipeline works perfectly on the Verification (Public) test set (predicts 1 image, generates valid CSV). However, it consistently fails on the Hidden Test Set with Notebook Threw Exception.\n\n*What I have tried:*\n- Memory Leak Check: Ran a stress test loop (2500 iterations on public image). Memory usage is stable. No OOM.\n- Device Mismatch: Model is traced on CUDA and inferred on CUDA. Works locally and on public runner.\n- Missing Files: Added robust checks for test_images/ directory and extensions (.jpg, .jpeg). Did not fix it.\n- Path Logic: Reverted to the standard direct path concatenation used in successful public kernels. (This notebook).\n\n**The Problem**\nPublic Run: Success. submission.csv is generated with valid floats.\nSubmission: Fails. Notebook Threw Exception (likely non-zero exit code). Since standard logs are not available for failed submissions, I am stuck.\n\nLink to public notebook: https://www.kaggle.com/code/aditya1711/csiro-p2-19-ref-logic/notebook?scriptVersionId=292320463", "votes": null }, { "id": "3392642", "postDate": "01/17/2026 09:34:13", "content": "I'm not sure if this solves the problem, but here is what's wrong with your code:\nfor i in range(len(submission)):\n rel_path = test_df.iloc[i]['image_path']\nYou should use sample_id for matching. Why use sys.exit(0)? Don't rely on sample_submission.csv to build the output; build it from scratch.
Avoid loading sample_submission or train dataframe csv files. It appears they are missing in submited notebook.
\nOnly generate your submission file from test.csv provided
", "rawMarkdown": "Avoid loading sample_submission or train dataframe csv files. It appears they are missing in submited notebook.\n\nOnly generate your submission file from test.csv provided", "votes": null }, { "id": "3393711", "postDate": "01/19/2026 16:26:00", "content": "100% certain this will cause failure. There are a dozen posts that tell you this is bad. Replace it with test.csv and it will probably work as intended. Replace it with train.csv and you can probably memory stress test your notebook.
\nsample_sub = pd.read_csv(os.path.join(DATA_DIR, 'sample_submission.csv'))
When you stress test it with train.csv you will get a huge error message. Not going to attempt to fix it for you but good luck.
", "rawMarkdown": "When you stress test it with train.csv you will get a huge error message. Not going to attempt to fix it for you but good luck.", "votes": null } ], "comments": [ { "id": 3392642, "author_name": "giovannyrodrguez", "author_url": "", "post_date": "01/17/2026 09:34:13", "content": "I'm not sure if this solves the problem, but here is what's wrong with your code:\nfor i in range(len(submission)):\n rel_path = test_df.iloc[i]['image_path']\nYou should use sample_id for matching. Why use sys.exit(0)? Don't rely on sample_submission.csv to build the output; build it from scratch.
Avoid loading sample_submission or train dataframe csv files. It appears they are missing in submited notebook.
\nOnly generate your submission file from test.csv provided
", "votes": null, "replies": [] }, { "id": 3393711, "author_name": "pcjimmmy", "author_url": "", "post_date": "01/19/2026 16:26:00", "content": "100% certain this will cause failure. There are a dozen posts that tell you this is bad. Replace it with test.csv and it will probably work as intended. Replace it with train.csv and you can probably memory stress test your notebook.
\nsample_sub = pd.read_csv(os.path.join(DATA_DIR, 'sample_submission.csv'))
When you stress test it with train.csv you will get a huge error message. Not going to attempt to fix it for you but good luck.
", "votes": null, "replies": [] } ] } ], "raw_markdown_by_id": { "3392563": "I am trying to submit a TorchScript model (.pt file) for the CSIRO Biomass competition. My pipeline works perfectly on the Verification (Public) test set (predicts 1 image, generates valid CSV). However, it consistently fails on the Hidden Test Set with Notebook Threw Exception.\n\n*What I have tried:*\n- Memory Leak Check: Ran a stress test loop (2500 iterations on public image). Memory usage is stable. No OOM.\n- Device Mismatch: Model is traced on CUDA and inferred on CUDA. Works locally and on public runner.\n- Missing Files: Added robust checks for test_images/ directory and extensions (.jpg, .jpeg). Did not fix it.\n- Path Logic: Reverted to the standard direct path concatenation used in successful public kernels. (This notebook).\n\n**The Problem**\nPublic Run: Success. submission.csv is generated with valid floats.\nSubmission: Fails. Notebook Threw Exception (likely non-zero exit code). Since standard logs are not available for failed submissions, I am stuck.\n\nLink to public notebook: https://www.kaggle.com/code/aditya1711/csiro-p2-19-ref-logic/notebook?scriptVersionId=292320463", "3392642": "I'm not sure if this solves the problem, but here is what's wrong with your code:\n` for i in range(len(submission)):\n rel_path = test_df.iloc[i]['image_path'] `\nYou should use sample_id for matching. Why use sys.exit(0)? Don't rely on sample_submission.csv to build the output; build it from scratch.", "3392995": "Avoid loading sample_submission or train dataframe csv files. It appears they are missing in submited notebook.\n\nOnly generate your submission file from test.csv provided", "3393711": "100% certain this will cause failure. There are a dozen posts that tell you this is bad. Replace it with test.csv and it will probably work as intended. Replace it with train.csv and you can probably memory stress test your notebook.\n\n`sample_sub = pd.read_csv(os.path.join(DATA_DIR, 'sample_submission.csv'))`", "3393718": "When you stress test it with train.csv you will get a huge error message. Not going to attempt to fix it for you but good luck." }, "source": "meta" }