{
  "id": 494665,
  "title": "MultiThread with OpenVino for 5 Folds Submission",
  "url": "/competitions/birdclef-2024/discussion/494665",
  "author_name": "",
  "post_date": "2024-04-18T01:41:28.630983Z",
  "votes": 6,
  "comment_count": 5,
  "views": 0,
  "content": "<blockquote>\n  <p>“Sharing is one of good ways to learn”</p>\n</blockquote>\n<p>Hello everyone ! Hope we are doing good and enjoy this competition. </p>\n<p>I would like to share my simple code for MultiThread implement Pytorch model in OpenVino runtime. With a few line of code, I was be able to submit 5 Folds model in the accepted time.</p>\n<p><a href=\"https://www.kaggle.com/code/lmhongkhnh/birdclef2024-multithread-with-openvino-inference\" target=\"_blank\">Inference Notebook</a></p>\n<p>Please comment if you find any mistake in the code, as that’s the way we learn.</p>\n<p>Thanks !<br>\nHappy code Kaggler 🫡</p>",
  "messages": [
    {
      "id": "2758160",
      "postDate": "04/18/2024 01:41:28",
      "content": "<blockquote>\n  <p>“Sharing is one of good ways to learn”</p>\n</blockquote>\n<p>Hello everyone ! Hope we are doing good and enjoy this competition. </p>\n<p>I would like to share my simple code for MultiThread implement Pytorch model in OpenVino runtime. With a few line of code, I was be able to submit 5 Folds model in the accepted time.</p>\n<p><a href=\"https://www.kaggle.com/code/lmhongkhnh/birdclef2024-multithread-with-openvino-inference\" target=\"_blank\">Inference Notebook</a></p>\n<p>Please comment if you find any mistake in the code, as that’s the way we learn.</p>\n<p>Thanks !<br>\nHappy code Kaggler 🫡</p>",
      "rawMarkdown": ">“Sharing is one of good ways to learn”\n\nHello everyone ! Hope we are doing good and enjoy this competition. \n\nI would like to share my simple code for MultiThread implement Pytorch model in OpenVino runtime. With a few line of code, I was be able to submit 5 Folds model in the accepted time.\n\n[Inference Notebook](https://www.kaggle.com/code/lmhongkhnh/birdclef2024-multithread-with-openvino-inference )\n\nPlease comment if you find any mistake in the code, as that’s the way we learn.\n\nThanks !\nHappy code Kaggler 🫡",
      "votes": null
    },
    {
      "id": "2758210",
      "postDate": "04/18/2024 02:58:59",
      "content": "<p>How much faster is openvino compared to just onnx? Have you measured it with your code?</p>",
      "rawMarkdown": "How much faster is openvino compared to just onnx? Have you measured it with your code?",
      "votes": null
    },
    {
      "id": "2758370",
      "postDate": "04/18/2024 06:18:44",
      "content": "<p>Yes, I did.<br>\nMy code shows that ONNX is slower than OpenVINO. Around 40s for 5 folds with 10 samples.</p>",
      "rawMarkdown": "Yes, I did.\nMy code shows that ONNX is slower than OpenVINO. Around 40s for 5 folds with 10 samples.",
      "votes": null
    },
    {
      "id": "2759012",
      "postDate": "04/18/2024 12:45:14",
      "content": "<p>I have a lower score with openvino vs onnx, 0.64 vs 0.65 for my model. Even if I stick to fp32 in openvino.</p>\n<p>Maybe I did something wrong, I will double check tomorrow, but I am sharing in case others have don't see the same difference as I do.</p>\n<p>I generate the openvino models from the onnx model I use, therefore there should not be much room for change.</p>",
      "rawMarkdown": "I have a lower score with openvino vs onnx, 0.64 vs 0.65 for my model. Even if I stick to fp32 in openvino.\n\nMaybe I did something wrong, I will double check tomorrow, but I am sharing in case others have don't see the same difference as I do.\n\nI generate the openvino models from the onnx model I use, therefore there should not be much room for change.",
      "votes": null
    },
    {
      "id": "2759091",
      "postDate": "04/18/2024 13:40:06",
      "content": "<p>Oh, really looking forward to hear your results.<br>\nbtw, I convert from PyTorch direct to OpenVino and the result is the same !</p>",
      "rawMarkdown": "Oh, really looking forward to hear your results.\nbtw, I convert from PyTorch direct to OpenVino and the result is the same !",
      "votes": null
    },
    {
      "id": "2759118",
      "postDate": "04/18/2024 13:49:57",
      "content": "<p>I had an errer when converting from pytorch.</p>",
      "rawMarkdown": "I had an errer when converting from pytorch.",
      "votes": null
    }
  ],
  "comments": [
    {
      "id": 2758210,
      "author_name": "cpmpml",
      "author_url": "",
      "post_date": "04/18/2024 02:58:59",
      "content": "<p>How much faster is openvino compared to just onnx? Have you measured it with your code?</p>",
      "votes": null,
      "replies": [
        {
          "id": 2758370,
          "author_name": "lmhongkhnh",
          "author_url": "",
          "post_date": "04/18/2024 06:18:44",
          "content": "<p>Yes, I did.<br>\nMy code shows that ONNX is slower than OpenVINO. Around 40s for 5 folds with 10 samples.</p>",
          "votes": null,
          "replies": []
        }
      ]
    },
    {
      "id": 2759012,
      "author_name": "cpmpml",
      "author_url": "",
      "post_date": "04/18/2024 12:45:14",
      "content": "<p>I have a lower score with openvino vs onnx, 0.64 vs 0.65 for my model. Even if I stick to fp32 in openvino.</p>\n<p>Maybe I did something wrong, I will double check tomorrow, but I am sharing in case others have don't see the same difference as I do.</p>\n<p>I generate the openvino models from the onnx model I use, therefore there should not be much room for change.</p>",
      "votes": null,
      "replies": [
        {
          "id": 2759091,
          "author_name": "lmhongkhnh",
          "author_url": "",
          "post_date": "04/18/2024 13:40:06",
          "content": "<p>Oh, really looking forward to hear your results.<br>\nbtw, I convert from PyTorch direct to OpenVino and the result is the same !</p>",
          "votes": null,
          "replies": [
            {
              "id": 2759118,
              "author_name": "cpmpml",
              "author_url": "",
              "post_date": "04/18/2024 13:49:57",
              "content": "<p>I had an errer when converting from pytorch.</p>",
              "votes": null,
              "replies": []
            }
          ]
        }
      ]
    }
  ],
  "raw_markdown_by_id": {
    "2758160": ">“Sharing is one of good ways to learn”\n\nHello everyone ! Hope we are doing good and enjoy this competition. \n\nI would like to share my simple code for MultiThread implement Pytorch model in OpenVino runtime. With a few line of code, I was be able to submit 5 Folds model in the accepted time.\n\n[Inference Notebook](https://www.kaggle.com/code/lmhongkhnh/birdclef2024-multithread-with-openvino-inference )\n\nPlease comment if you find any mistake in the code, as that’s the way we learn.\n\nThanks !\nHappy code Kaggler 🫡",
    "2758210": "How much faster is openvino compared to just onnx? Have you measured it with your code?",
    "2758370": "Yes, I did.\nMy code shows that ONNX is slower than OpenVINO. Around 40s for 5 folds with 10 samples.",
    "2759012": "I have a lower score with openvino vs onnx, 0.64 vs 0.65 for my model. Even if I stick to fp32 in openvino.\n\nMaybe I did something wrong, I will double check tomorrow, but I am sharing in case others have don't see the same difference as I do.\n\nI generate the openvino models from the onnx model I use, therefore there should not be much room for change.",
    "2759091": "Oh, really looking forward to hear your results.\nbtw, I convert from PyTorch direct to OpenVino and the result is the same !",
    "2759118": "I had an errer when converting from pytorch."
  },
  "source": "meta"
}