{
  "id": 613229,
  "title": "anyone have face memory issue ",
  "url": "/competitions/cafa-6-protein-function-prediction/discussion/613229",
  "author_name": "",
  "post_date": "2025-10-25T07:15:11.766592700Z",
  "votes": 2,
  "comment_count": 3,
  "views": 0,
  "content": "<p>i stuck on memory issue problem that persist my training issue have any one face this issue</p>",
  "messages": [
    {
      "id": "3306755",
      "postDate": "10/25/2025 07:15:11",
      "content": "<p>i stuck on memory issue problem that persist my training issue have any one face this issue</p>",
      "rawMarkdown": "i stuck on memory issue problem that persist my training issue have any one face this issue",
      "votes": null
    },
    {
      "id": "3312037",
      "postDate": "11/06/2025 09:23:00",
      "content": "<p>I also have faced memory issue. After 68% the session automatically restarts.</p>",
      "rawMarkdown": "I also have faced memory issue. After 68% the session automatically restarts.",
      "votes": null
    },
    {
      "id": "3319543",
      "postDate": "11/11/2025 20:39:01",
      "content": "<p>I am spending hours to get my nvidia graphical card GPU acceleration ready 😂,  I am a biomedical scientist and was an AI engineer in 2020, now 2 days ago get to programming again. Chatgpt5 is a hugh difference to do things in short time. Constantly iterate between code generation of chatgpt5, test and debug, I do a manual check with the eye for each script to be sure 😅  I prepared my features in 5 hours with nice graphs and thresholds to delete certain protein entries in the fasta training data because they are less value for computetime vs result. I am now struggling for 8 hours with getting my GPU ready with cuda and tensorflow but I am almost there I think.  I assume you can use a CNN model because a protein consists of proteindomains. And proteindomains are probably correlating with biological functions. It's like finding cats in pictures but now Go terms in protein sequences I guess, I am planning to evaluate different neural network architectures. Intrested in Co-op? Let me know <a href=\"mailto:ertryckx.joost@gmail.com\">ertryckx.joost@gmail.com</a></p>",
      "rawMarkdown": "I am spending hours to get my nvidia graphical card GPU acceleration ready 😂,  I am a biomedical scientist and was an AI engineer in 2020, now 2 days ago get to programming again. Chatgpt5 is a hugh difference to do things in short time. Constantly iterate between code generation of chatgpt5, test and debug, I do a manual check with the eye for each script to be sure 😅  I prepared my features in 5 hours with nice graphs and thresholds to delete certain protein entries in the fasta training data because they are less value for computetime vs result. I am now struggling for 8 hours with getting my GPU ready with cuda and tensorflow but I am almost there I think.  I assume you can use a CNN model because a protein consists of proteindomains. And proteindomains are probably correlating with biological functions. It's like finding cats in pictures but now Go terms in protein sequences I guess, I am planning to evaluate different neural network architectures. Intrested in Co-op? Let me know ertryckx.joost@gmail.com",
      "votes": null
    },
    {
      "id": "3344385",
      "postDate": "11/22/2025 14:17:04",
      "content": "<p>I resolved my issues with out-of-memory errors by gradually reducing the batch size in powers of 2 (from 64 → 32 → 16) until inference succeeded.  A batch size of 32 worked well on a P100 GPU.  Monitoring the logs and using Claude.ai to troubleshoot based on the error messages was particularly helpful.</p>",
      "rawMarkdown": "I resolved my issues with out-of-memory errors by gradually reducing the batch size in powers of 2 (from 64 → 32 → 16) until inference succeeded.  A batch size of 32 worked well on a P100 GPU.  Monitoring the logs and using Claude.ai to troubleshoot based on the error messages was particularly helpful.",
      "votes": null
    }
  ],
  "comments": [
    {
      "id": 3312037,
      "author_name": "mdshamimaktar",
      "author_url": "",
      "post_date": "11/06/2025 09:23:00",
      "content": "<p>I also have faced memory issue. After 68% the session automatically restarts.</p>",
      "votes": null,
      "replies": []
    },
    {
      "id": 3319543,
      "author_name": "joostertryckx",
      "author_url": "",
      "post_date": "11/11/2025 20:39:01",
      "content": "<p>I am spending hours to get my nvidia graphical card GPU acceleration ready 😂,  I am a biomedical scientist and was an AI engineer in 2020, now 2 days ago get to programming again. Chatgpt5 is a hugh difference to do things in short time. Constantly iterate between code generation of chatgpt5, test and debug, I do a manual check with the eye for each script to be sure 😅  I prepared my features in 5 hours with nice graphs and thresholds to delete certain protein entries in the fasta training data because they are less value for computetime vs result. I am now struggling for 8 hours with getting my GPU ready with cuda and tensorflow but I am almost there I think.  I assume you can use a CNN model because a protein consists of proteindomains. And proteindomains are probably correlating with biological functions. It's like finding cats in pictures but now Go terms in protein sequences I guess, I am planning to evaluate different neural network architectures. Intrested in Co-op? Let me know <a href=\"mailto:ertryckx.joost@gmail.com\">ertryckx.joost@gmail.com</a></p>",
      "votes": null,
      "replies": []
    },
    {
      "id": 3344385,
      "author_name": "christie",
      "author_url": "",
      "post_date": "11/22/2025 14:17:04",
      "content": "<p>I resolved my issues with out-of-memory errors by gradually reducing the batch size in powers of 2 (from 64 → 32 → 16) until inference succeeded.  A batch size of 32 worked well on a P100 GPU.  Monitoring the logs and using Claude.ai to troubleshoot based on the error messages was particularly helpful.</p>",
      "votes": null,
      "replies": []
    }
  ],
  "raw_markdown_by_id": {
    "3306755": "i stuck on memory issue problem that persist my training issue have any one face this issue",
    "3312037": "I also have faced memory issue. After 68% the session automatically restarts.",
    "3319543": "I am spending hours to get my nvidia graphical card GPU acceleration ready 😂,  I am a biomedical scientist and was an AI engineer in 2020, now 2 days ago get to programming again. Chatgpt5 is a hugh difference to do things in short time. Constantly iterate between code generation of chatgpt5, test and debug, I do a manual check with the eye for each script to be sure 😅  I prepared my features in 5 hours with nice graphs and thresholds to delete certain protein entries in the fasta training data because they are less value for computetime vs result. I am now struggling for 8 hours with getting my GPU ready with cuda and tensorflow but I am almost there I think.  I assume you can use a CNN model because a protein consists of proteindomains. And proteindomains are probably correlating with biological functions. It's like finding cats in pictures but now Go terms in protein sequences I guess, I am planning to evaluate different neural network architectures. Intrested in Co-op? Let me know ertryckx.joost@gmail.com",
    "3344385": "I resolved my issues with out-of-memory errors by gradually reducing the batch size in powers of 2 (from 64 → 32 → 16) until inference succeeded.  A batch size of 32 worked well on a P100 GPU.  Monitoring the logs and using Claude.ai to troubleshoot based on the error messages was particularly helpful."
  },
  "source": "meta"
}