{ "id": 674439, "title": "Model usage and VRAM limitations", "url": "/competitions/accident/discussion/674439", "author_name": "", "post_date": "2026-02-20T13:25:15.613626300Z", "votes": 2, "comment_count": 5, "views": 0, "content": "

Hello,

\n

Thank you for organizing this competition.

\n

I would like to clarify a couple of points regarding model usage and computational constraints.

\n
    \n

  1. Public pretrained models\nIs it allowed to use any publicly available pretrained models (e.g., from Hugging Face), including multimodal or video models?

    2. \n

  2. VRAM / resource limits\nAre there any recommended or enforced limits regarding GPU memory (VRAM) or computational resources for Kaggle notebooks or submissions?

    3. \n
\n

Thanks

", "messages": [ { "id": "3408371", "postDate": "02/20/2026 13:25:15", "content": "

Hello,

\n

Thank you for organizing this competition.

\n

I would like to clarify a couple of points regarding model usage and computational constraints.

\n
    \n

  1. Public pretrained models\nIs it allowed to use any publicly available pretrained models (e.g., from Hugging Face), including multimodal or video models?

    2. \n

  2. VRAM / resource limits\nAre there any recommended or enforced limits regarding GPU memory (VRAM) or computational resources for Kaggle notebooks or submissions?

    3. \n
\n

Thanks

", "rawMarkdown": "Hello,\n\nThank you for organizing this competition.\n\nI would like to clarify a couple of points regarding model usage and computational constraints.\n\n1. Public pretrained models\nIs it allowed to use any publicly available pretrained models (e.g., from Hugging Face), including multimodal or video models?\n\n2. VRAM / resource limits\nAre there any recommended or enforced limits regarding GPU memory (VRAM) or computational resources for Kaggle notebooks or submissions?\n\nThanks", "votes": null }, { "id": "3408374", "postDate": "02/20/2026 13:35:23", "content": "

Hi @chancekim ,

\n

Thank you for asking!\nThere are no specific restrictions on either of your points, but we encourage everyone to always ask upfront about a specific model.

\n

Good Luck!

\n

Best,\nLukas

", "rawMarkdown": "Hi @chancekim ,\n\nThank you for asking!\nThere are no specific restrictions on either of your points, but we encourage everyone to always ask upfront about a specific model.\n\nGood Luck!\n\nBest,\nLukas", "votes": null }, { "id": "3418732", "postDate": "03/09/2026 05:15:25", "content": "

Hi @picekl, there's a host of VLMs on huggingface that may potentially be used. For reference purposes, 7B models or higher for example. These may occassionally require more VRAM that that available on Kaggle and also run for longer. Is there an expected inference time that the pipeline is supposed to meet?

", "rawMarkdown": "Hi @picekl, there's a host of VLMs on huggingface that may potentially be used. For reference purposes, 7B models or higher for example. These may occassionally require more VRAM that that available on Kaggle and also run for longer. Is there an expected inference time that the pipeline is supposed to meet?", "votes": null }, { "id": "3418872", "postDate": "03/09/2026 12:35:53", "content": "

Hi @prashanttandon ,

\n

Not really. Use what ever you want :)

\n

Best,\nLukas

", "rawMarkdown": "Hi @prashanttandon ,\n\nNot really. Use what ever you want :)\n\nBest,\nLukas", "votes": null }, { "id": "3429688", "postDate": "03/27/2026 04:48:48", "content": "

Hi @picekl ,

\n

Thanks for the clarification so far.

\n

Just one quick follow-up should we assume submissions need to run within typical Kaggle resource limits, or are higher-resource pipelines (with optimizations or precomputed outputs) acceptable as long as results are reproducible?

\n

Appreciate the guidance!

", "rawMarkdown": "Hi @picekl ,\n\nThanks for the clarification so far.\n\nJust one quick follow-up should we assume submissions need to run within typical Kaggle resource limits, or are higher-resource pipelines (with optimizations or precomputed outputs) acceptable as long as results are reproducible?\n\nAppreciate the guidance!", "votes": null }, { "id": "3430091", "postDate": "03/27/2026 20:02:04", "content": "

Hi @ladiposamson ,

\n

No restrictions related to that :)

\n

LP

", "rawMarkdown": "Hi @ladiposamson ,\n\nNo restrictions related to that :)\n\nLP", "votes": null } ], "comments": [ { "id": 3408374, "author_name": "picekl", "author_url": "", "post_date": "02/20/2026 13:35:23", "content": "

Hi @chancekim ,

\n

Thank you for asking!\nThere are no specific restrictions on either of your points, but we encourage everyone to always ask upfront about a specific model.

\n

Good Luck!

\n

Best,\nLukas

", "votes": null, "replies": [ { "id": 3418732, "author_name": "prashanttandon", "author_url": "", "post_date": "03/09/2026 05:15:25", "content": "

Hi @picekl, there's a host of VLMs on huggingface that may potentially be used. For reference purposes, 7B models or higher for example. These may occassionally require more VRAM that that available on Kaggle and also run for longer. Is there an expected inference time that the pipeline is supposed to meet?

", "votes": null, "replies": [ { "id": 3418872, "author_name": "picekl", "author_url": "", "post_date": "03/09/2026 12:35:53", "content": "

Hi @prashanttandon ,

\n

Not really. Use what ever you want :)

\n

Best,\nLukas

", "votes": null, "replies": [] } ] }, { "id": 3429688, "author_name": "ladiposamson", "author_url": "", "post_date": "03/27/2026 04:48:48", "content": "

Hi @picekl ,

\n

Thanks for the clarification so far.

\n

Just one quick follow-up should we assume submissions need to run within typical Kaggle resource limits, or are higher-resource pipelines (with optimizations or precomputed outputs) acceptable as long as results are reproducible?

\n

Appreciate the guidance!

", "votes": null, "replies": [ { "id": 3430091, "author_name": "picekl", "author_url": "", "post_date": "03/27/2026 20:02:04", "content": "

Hi @ladiposamson ,

\n

No restrictions related to that :)

\n

LP

", "votes": null, "replies": [] } ] } ] } ], "raw_markdown_by_id": { "3408371": "Hello,\n\nThank you for organizing this competition.\n\nI would like to clarify a couple of points regarding model usage and computational constraints.\n\n1. Public pretrained models\nIs it allowed to use any publicly available pretrained models (e.g., from Hugging Face), including multimodal or video models?\n\n2. VRAM / resource limits\nAre there any recommended or enforced limits regarding GPU memory (VRAM) or computational resources for Kaggle notebooks or submissions?\n\nThanks", "3408374": "Hi @chancekim ,\n\nThank you for asking!\nThere are no specific restrictions on either of your points, but we encourage everyone to always ask upfront about a specific model.\n\nGood Luck!\n\nBest,\nLukas", "3418732": "Hi @picekl, there's a host of VLMs on huggingface that may potentially be used. For reference purposes, 7B models or higher for example. These may occassionally require more VRAM that that available on Kaggle and also run for longer. Is there an expected inference time that the pipeline is supposed to meet?", "3418872": "Hi @prashanttandon ,\n\nNot really. Use what ever you want :)\n\nBest,\nLukas", "3429688": "Hi @picekl ,\n\nThanks for the clarification so far.\n\nJust one quick follow-up should we assume submissions need to run within typical Kaggle resource limits, or are higher-resource pipelines (with optimizations or precomputed outputs) acceptable as long as results are reproducible?\n\nAppreciate the guidance!", "3430091": "Hi @ladiposamson ,\n\nNo restrictions related to that :)\n\nLP" }, "source": "meta" }