I'm following the \"CMI 2025 Demo Submission\" guidelines, using inference_server with the predict function. I train my model in an external notebook, save it to a dataset, and then load it in the submission notebook for inference. The notebook runs successfully within 2 minutes, including the inference_server cell (copied from the official demo), and logs show no errors.
\nHowever, during scoring, I keep getting the error: \"Notebook Inference Server Never Started.\"
\nI'm using the open-source aeon library, which isn’t in the default Kaggle environment, so I added it via the Kaggle Packages dependency manager as instructed. Since the rules allow public external data and pre-trained models, I assumed this would be fine.
\nI’d appreciate clarification: is this an issue with the Kaggle dependency manager, or does inference_server not support external dependencies? If these can’t be used in this competition, it would be helpful to know so I don’t waste time trying to make it work.
\nHas anyone else run into similar issues with external libraries?
", "messages": [ { "id": "3220139", "postDate": "06/08/2025 21:55:58", "content": "I'm following the \"CMI 2025 Demo Submission\" guidelines, using inference_server with the predict function. I train my model in an external notebook, save it to a dataset, and then load it in the submission notebook for inference. The notebook runs successfully within 2 minutes, including the inference_server cell (copied from the official demo), and logs show no errors.
\nHowever, during scoring, I keep getting the error: \"Notebook Inference Server Never Started.\"
\nI'm using the open-source aeon library, which isn’t in the default Kaggle environment, so I added it via the Kaggle Packages dependency manager as instructed. Since the rules allow public external data and pre-trained models, I assumed this would be fine.
\nI’d appreciate clarification: is this an issue with the Kaggle dependency manager, or does inference_server not support external dependencies? If these can’t be used in this competition, it would be helpful to know so I don’t waste time trying to make it work.
\nHas anyone else run into similar issues with external libraries?
", "rawMarkdown": "I'm following the \"CMI 2025 Demo Submission\" guidelines, using inference_server with the predict function. I train my model in an external notebook, save it to a dataset, and then load it in the submission notebook for inference. The notebook runs successfully within 2 minutes, including the inference_server cell (copied from the official demo), and logs show no errors.\n\nHowever, during scoring, I keep getting the error: \"Notebook Inference Server Never Started.\"\n\nI'm using the open-source aeon library, which isn’t in the default Kaggle environment, so I added it via the Kaggle Packages dependency manager as [instructed](https://www.kaggle.com/docs/packages). Since the rules allow public external data and pre-trained models, I assumed this would be fine.\n\nI’d appreciate clarification: is this an issue with the Kaggle dependency manager, or does inference_server not support external dependencies? If these can’t be used in this competition, it would be helpful to know so I don’t waste time trying to make it work.\n\nHas anyone else run into similar issues with external libraries?", "votes": null }, { "id": "3220144", "postDate": "06/08/2025 22:11:09", "content": "Did you make sure to call inference_server.serve() in the notebook?
", "rawMarkdown": "Did you make sure to call inference_server.serve() in the notebook?", "votes": null }, { "id": "3220146", "postDate": "06/08/2025 22:15:52", "content": "uhm I did
\n\"
\nif os.getenv('KAGGLE_IS_COMPETITION_RERUN'):
\n inference_server.serve()
\n\"to be precise, because it was like that in demo submission. do we need to call it unconditionally? 🤔
I think your training in the first sequence call is exceeding 30 minutes causing this error @pelinkeskin
", "rawMarkdown": "I think your training in the first sequence call is exceeding 30 minutes causing this error @pelinkeskin", "votes": null }, { "id": "3220443", "postDate": "06/09/2025 10:16:01", "content": "thnx for your comment. Installing wheel files from a dataset using pip directly within notebook cell solved my submission scoring problem, so the issue was using kaggle packages dependency manager.
", "rawMarkdown": "thnx for your comment. Installing wheel files from a dataset using pip directly within notebook cell solved my submission scoring problem, so the issue was using kaggle packages dependency manager.", "votes": null }, { "id": "3245934", "postDate": "07/10/2025 04:10:26", "content": "where is inference_server? how to import that?
", "rawMarkdown": "where is inference_server? how to import that?", "votes": null }, { "id": "3277846", "postDate": "08/28/2025 18:19:55", "content": "could you provide more detail about how to export a wheel files from kaggle packages.I meet the same question,and need your help please
", "rawMarkdown": "could you provide more detail about how to export a wheel files from kaggle packages.I meet the same question,and need your help please", "votes": null }, { "id": "3278158", "postDate": "08/29/2025 10:56:16", "content": "hi @ovrtnsc. this post is useful for learning how to do it https://www.kaggle.com/competitions/severstal-steel-defect-detection/discussion/113195
", "rawMarkdown": "hi @ovrtnsc. this post is useful for learning how to do it https://www.kaggle.com/competitions/severstal-steel-defect-detection/discussion/113195", "votes": null }, { "id": "3278232", "postDate": "08/29/2025 14:15:18", "content": "Thanks for sharing, that's really helpful!😄
", "rawMarkdown": "Thanks for sharing, that's really helpful!😄", "votes": null } ], "comments": [ { "id": 3220144, "author_name": "mertbayraktar", "author_url": "", "post_date": "06/08/2025 22:11:09", "content": "Did you make sure to call inference_server.serve() in the notebook?
", "votes": null, "replies": [ { "id": 3220146, "author_name": "pelinkeskin", "author_url": "", "post_date": "06/08/2025 22:15:52", "content": "uhm I did
\n\"
\nif os.getenv('KAGGLE_IS_COMPETITION_RERUN'):
\n inference_server.serve()
\n\"to be precise, because it was like that in demo submission. do we need to call it unconditionally? 🤔
where is inference_server? how to import that?
", "votes": null, "replies": [] } ] } ] }, { "id": 3220267, "author_name": "ravi20076", "author_url": "", "post_date": "06/09/2025 04:58:53", "content": "I think your training in the first sequence call is exceeding 30 minutes causing this error @pelinkeskin
", "votes": null, "replies": [ { "id": 3220443, "author_name": "pelinkeskin", "author_url": "", "post_date": "06/09/2025 10:16:01", "content": "thnx for your comment. Installing wheel files from a dataset using pip directly within notebook cell solved my submission scoring problem, so the issue was using kaggle packages dependency manager.
", "votes": null, "replies": [ { "id": 3277846, "author_name": "ovrtnsc", "author_url": "", "post_date": "08/28/2025 18:19:55", "content": "could you provide more detail about how to export a wheel files from kaggle packages.I meet the same question,and need your help please
", "votes": null, "replies": [ { "id": 3278158, "author_name": "pelinkeskin", "author_url": "", "post_date": "08/29/2025 10:56:16", "content": "hi @ovrtnsc. this post is useful for learning how to do it https://www.kaggle.com/competitions/severstal-steel-defect-detection/discussion/113195
", "votes": null, "replies": [ { "id": 3278232, "author_name": "ovrtnsc", "author_url": "", "post_date": "08/29/2025 14:15:18", "content": "Thanks for sharing, that's really helpful!😄
", "votes": null, "replies": [] } ] } ] } ] } ] } ], "raw_markdown_by_id": { "3220139": "I'm following the \"CMI 2025 Demo Submission\" guidelines, using inference_server with the predict function. I train my model in an external notebook, save it to a dataset, and then load it in the submission notebook for inference. The notebook runs successfully within 2 minutes, including the inference_server cell (copied from the official demo), and logs show no errors.\n\nHowever, during scoring, I keep getting the error: \"Notebook Inference Server Never Started.\"\n\nI'm using the open-source aeon library, which isn’t in the default Kaggle environment, so I added it via the Kaggle Packages dependency manager as [instructed](https://www.kaggle.com/docs/packages). Since the rules allow public external data and pre-trained models, I assumed this would be fine.\n\nI’d appreciate clarification: is this an issue with the Kaggle dependency manager, or does inference_server not support external dependencies? If these can’t be used in this competition, it would be helpful to know so I don’t waste time trying to make it work.\n\nHas anyone else run into similar issues with external libraries?", "3220144": "Did you make sure to call inference_server.serve() in the notebook?", "3220146": "uhm I did \n\"\nif os.getenv('KAGGLE_IS_COMPETITION_RERUN'):\n inference_server.serve()\n\"to be precise, because it was like that in demo submission. do we need to call it unconditionally? 🤔", "3220267": "I think your training in the first sequence call is exceeding 30 minutes causing this error @pelinkeskin", "3220443": "thnx for your comment. Installing wheel files from a dataset using pip directly within notebook cell solved my submission scoring problem, so the issue was using kaggle packages dependency manager.", "3245934": "where is inference_server? how to import that?", "3277846": "could you provide more detail about how to export a wheel files from kaggle packages.I meet the same question,and need your help please", "3278158": "hi @ovrtnsc. this post is useful for learning how to do it https://www.kaggle.com/competitions/severstal-steel-defect-detection/discussion/113195", "3278232": "Thanks for sharing, that's really helpful!😄" }, "source": "meta" }