{ "id": 673621, "title": "3 public protenix inference notebook for RNA 3D Folding Part 2", "url": "/competitions/stanford-rna-3d-folding-2/discussion/673621", "author_name": "Tony Li", "post_date": "2026-02-15T23:41:43.777000", "votes": 9, "comment_count": 16, "views": 0, "content": "

Hi all โ€” does anyone have a public, working Protenix inference notebook for Stanford RNA 3D Folding (Part 2) (end-to-end โ†’ generates a valid submission)?

\n

Iโ€™m interested because Protenix looks like a promising diversity source for ensembles. Repo:

\n

https://github.com/bytedance/Protenix

\n

Also saw their new release note:

\n
\n

2026-02-05: Protenix-v1 Released ๐Ÿ’ช [Technical Report]\n Supported Template/RNA MSA features and improved training dynamics, along with further inference-time performance enhancements.

\n
\n

Iโ€™m planning to spend some time exploring this comp and this model โ€” If anyone has a notebook link (or a minimal inference example adapted to this comp) or info about this model , Iโ€™d really appreciate it. Thanks!

\n

Update โ€” thanks 5 kindly sharing notebooks:

\n

LB 0.408 : https://www.kaggle.com/code/llkh0a/protenix-tbm

\n

LB 0.406: https://www.kaggle.com/code/rluethy/template-or-protenix-lb-0-4-ish?scriptVersionId=298270706

\n

LB 0.250: https://www.kaggle.com/code/zoushuxian/aido-rna-augmented-protenix

\n

Draft version: https://www.kaggle.com/code/qiweiyin/protenix-v1-inference-2026

\n

version 2: https://www.kaggle.com/datasets/qiweiyin/protenix-v1-adjusted?select=Protenix-v1-adjust-v2

", "messages": [ { "id": 3406541, "postDate": "2026-02-15T23:41:43.777Z", "content": "

Hi all โ€” does anyone have a public, working Protenix inference notebook for Stanford RNA 3D Folding (Part 2) (end-to-end โ†’ generates a valid submission)?

\n

Iโ€™m interested because Protenix looks like a promising diversity source for ensembles. Repo:

\n

https://github.com/bytedance/Protenix

\n

Also saw their new release note:

\n
\n

2026-02-05: Protenix-v1 Released ๐Ÿ’ช [Technical Report]\n Supported Template/RNA MSA features and improved training dynamics, along with further inference-time performance enhancements.

\n
\n

Iโ€™m planning to spend some time exploring this comp and this model โ€” If anyone has a notebook link (or a minimal inference example adapted to this comp) or info about this model , Iโ€™d really appreciate it. Thanks!

\n

Update โ€” thanks 5 kindly sharing notebooks:

\n

LB 0.408 : https://www.kaggle.com/code/llkh0a/protenix-tbm

\n

LB 0.406: https://www.kaggle.com/code/rluethy/template-or-protenix-lb-0-4-ish?scriptVersionId=298270706

\n

LB 0.250: https://www.kaggle.com/code/zoushuxian/aido-rna-augmented-protenix

\n

Draft version: https://www.kaggle.com/code/qiweiyin/protenix-v1-inference-2026

\n

version 2: https://www.kaggle.com/datasets/qiweiyin/protenix-v1-adjusted?select=Protenix-v1-adjust-v2

", "rawMarkdown": "Hi all โ€” does anyone have a **public, working Protenix inference notebook** for **Stanford RNA 3D Folding (Part 2)** (end-to-end โ†’ generates a valid submission)?\n\nIโ€™m interested because Protenix looks like a **promising diversity source** for ensembles. Repo:\n\nhttps://github.com/bytedance/Protenix \n\nAlso saw their new release note:\n\n> **2026-02-05: Protenix-v1 Released ๐Ÿ’ช [Technical Report]**\n> Supported Template/RNA MSA features and improved training dynamics, along with further inference-time performance enhancements.\n\nIโ€™m planning to spend some time exploring this comp and this model โ€” If anyone has a notebook link (or a minimal inference example adapted to this comp) or info about this model , Iโ€™d really appreciate it. Thanks!\n\n**Update โ€” thanks 5 kindly sharing notebooks:**\n\nLB 0.408 : https://www.kaggle.com/code/llkh0a/protenix-tbm\n\nLB 0.406: https://www.kaggle.com/code/rluethy/template-or-protenix-lb-0-4-ish?scriptVersionId=298270706\n\nLB 0.250: https://www.kaggle.com/code/zoushuxian/aido-rna-augmented-protenix\n\nDraft version: https://www.kaggle.com/code/qiweiyin/protenix-v1-inference-2026\n\nversion 2: https://www.kaggle.com/datasets/qiweiyin/protenix-v1-adjusted?select=Protenix-v1-adjust-v2 \n\n", "votes": 9 }, { "id": 3406622, "postDate": "2026-02-16T08:03:35.047Z", "content": "

a draft version: https://www.kaggle.com/code/qiweiyin/protenix-v1-inference-2026

", "rawMarkdown": "a draft version: https://www.kaggle.com/code/qiweiyin/protenix-v1-inference-2026 ", "votes": 3, "replies": [ { "id": 3406665, "postDate": "2026-02-16T09:42:52.380Z", "content": "

There is significant room for memory optimization in Protenix's code.

", "rawMarkdown": "There is significant room for memory optimization in Protenix's code.", "votes": 1 }, { "id": 3406739, "postDate": "2026-02-16T14:25:12.873Z", "content": "
\n

A draft version: https://www.kaggle.com/code/qiweiyin/protenix-v1-inference-2026

\n
\n

Thank you for sharing this โ€” it saves me a lot of time. Iโ€™ll dive into it soon ๐Ÿ™‚\nAlso, I noticed two other working Protenix notebooks:

\n", "rawMarkdown": "> A draft version: [https://www.kaggle.com/code/qiweiyin/protenix-v1-inference-2026](https://www.kaggle.com/code/qiweiyin/protenix-v1-inference-2026)\n\n\n\nThank you for sharing this โ€” it saves me a lot of time. Iโ€™ll dive into it soon ๐Ÿ™‚\nAlso, I noticed two other working Protenix notebooks:\n\n* **LB 0.404:** [https://www.kaggle.com/code/rluethy/template-or-protenix-lb-0-4-ish](https://www.kaggle.com/code/rluethy/template-or-protenix-lb-0-4-ish)\n* **LB 0.250:** [https://www.kaggle.com/code/zoushuxian/aido-rna-augmented-protenix](https://www.kaggle.com/code/zoushuxian/aido-rna-augmented-protenix)\n\n", "votes": 1, "replies": [ { "id": 3406775, "postDate": "2026-02-16T15:59:27.550Z", "content": "

The draft version https://www.kaggle.com/code/qiweiyin/protenix-v1-inference-2026 was based on the latest 2026-02-05: Protenix-v1 Released

", "rawMarkdown": "The draft version [https://www.kaggle.com/code/qiweiyin/protenix-v1-inference-2026](https://www.kaggle.com/code/qiweiyin/protenix-v1-inference-2026) was based on the latest 2026-02-05: Protenix-v1 Released", "votes": 2 } ] } ] }, { "id": 3406592, "postDate": "2026-02-16T05:32:23.560Z", "content": "

I'm trying this. but always OOM issues!

", "rawMarkdown": "I'm trying this. but always OOM issues!", "votes": 2, "replies": [ { "id": 3407294, "postDate": "2026-02-18T03:52:03.067Z", "content": "

Hi!\nI slightly modified your script (https://www.kaggle.com/code/alexxanderlarko/protenix-v1) and the problem is solved.\nCould you please add the model?\n(protenix_base_20250630_v1.0.0 in checkpoint dir.)

", "rawMarkdown": "Hi!\nI slightly modified your script (https://www.kaggle.com/code/alexxanderlarko/protenix-v1) and the problem is solved.\nCould you please add the model?\n(protenix_base_20250630_v1.0.0 in checkpoint dir.)", "votes": 1, "replies": [ { "id": 3407316, "postDate": "2026-02-18T05:07:12.507Z", "content": "

Hi, thanks!~ I created a new version Dataset named \"Protenix-v1-adjust-v2\" (https://www.kaggle.com/datasets/qiweiyin/protenix-v1-adjusted?select=Protenix-v1-adjust-v2 ) The protenix_base_20250630_v1.0.0 is in its checkpoint dir.

", "rawMarkdown": "Hi, thanks!~ I created a new version Dataset named \"Protenix-v1-adjust-v2\" (https://www.kaggle.com/datasets/qiweiyin/protenix-v1-adjusted?select=Protenix-v1-adjust-v2 ) The protenix_base_20250630_v1.0.0 is in its checkpoint dir.", "votes": 2 }, { "id": 3407318, "postDate": "2026-02-18T05:13:24.323Z", "content": "

THANK YOU!๐Ÿ™‚

", "rawMarkdown": "THANK YOU!๐Ÿ™‚", "votes": 1 }, { "id": 3411930, "postDate": "2026-02-24T00:31:35.270Z", "rawMarkdown": "", "isDeleted": true }, { "id": 3413023, "postDate": "2026-02-24T04:50:30.877Z", "content": "

may i ask where do you find the checkpoint files?

", "rawMarkdown": "may i ask where do you find the checkpoint files?" }, { "id": 3413037, "postDate": "2026-02-24T05:27:50.823Z", "rawMarkdown": "", "isDeleted": true }, { "id": 3413039, "postDate": "2026-02-24T05:29:22.117Z", "content": "

I searched the official website: https://github.com/bytedance/Protenix and found the official download links in the documents and codes. You can also download the full contents of https://github.com/bytedance/Protenix to your local machine and then search it :)

", "rawMarkdown": "I searched the official website: https://github.com/bytedance/Protenix and found the official download links in the documents and codes. You can also download the full contents of https://github.com/bytedance/Protenix to your local machine and then search it :)", "votes": 1 } ] }, { "id": 3407733, "postDate": "2026-02-19T02:05:02.017Z", "content": "

based on Aleksandr Larko's notebook and protenix lastest checkpoint, i have made a complete inference notebook that utilize both TBM and protenix

\n

Pure protenix scored 0.249 and scored 0.408 when mixed with TBM

", "rawMarkdown": "based on [Aleksandr Larko's](https://www.kaggle.com/code/alexxanderlarko/protenix-v1) notebook and [protenix lastest checkpoint](https://www.kaggle.com/datasets/qiweiyin/protenix-v1-adjusted), i have made a complete [inference notebook ](https://www.kaggle.com/code/llkh0a/protenix-tbm) that utilize both TBM and protenix\n\nPure protenix scored 0.249 and scored 0.408 when mixed with TBM\n\n", "votes": 4, "replies": [ { "id": 3407741, "postDate": "2026-02-19T02:31:51.413Z", "content": "

Great work๏ผ

", "rawMarkdown": "Great work๏ผ", "votes": 2 } ] } ] }, { "id": 3406669, "postDate": "2026-02-16T09:52:21.100Z", "rawMarkdown": "", "isDeleted": true }, { "id": 3407377, "postDate": "2026-02-18T07:21:55.320Z", "content": "

Thank you for this post @tonylica

", "rawMarkdown": "Thank you for this post @tonylica ", "votes": 1 } ], "comments": [ { "id": 3406622, "author_name": "Qiwei", "author_url": "", "post_date": "2026-02-16T08:03:35.047000", "content": "

a draft version: https://www.kaggle.com/code/qiweiyin/protenix-v1-inference-2026

", "votes": 3, "replies": [ { "id": 3406665, "author_name": "DECEM", "author_url": "", "post_date": "2026-02-16T09:42:52.380000", "content": "

There is significant room for memory optimization in Protenix's code.

", "votes": 1, "replies": [] }, { "id": 3406739, "author_name": "Tony Li", "author_url": "", "post_date": "2026-02-16T14:25:12.873000", "content": "
\n

A draft version: https://www.kaggle.com/code/qiweiyin/protenix-v1-inference-2026

\n
\n

Thank you for sharing this โ€” it saves me a lot of time. Iโ€™ll dive into it soon ๐Ÿ™‚\nAlso, I noticed two other working Protenix notebooks:

\n", "votes": 1, "replies": [ { "id": 3406775, "author_name": "Qiwei", "author_url": "", "post_date": "2026-02-16T15:59:27.550000", "content": "

The draft version https://www.kaggle.com/code/qiweiyin/protenix-v1-inference-2026 was based on the latest 2026-02-05: Protenix-v1 Released

", "votes": 2, "replies": [] } ] } ] }, { "id": 3406592, "author_name": "Qiwei", "author_url": "", "post_date": "2026-02-16T05:32:23.560000", "content": "

I'm trying this. but always OOM issues!

", "votes": 2, "replies": [ { "id": 3407294, "author_name": "Aleksandr Larko", "author_url": "", "post_date": "2026-02-18T03:52:03.067000", "content": "

Hi!\nI slightly modified your script (https://www.kaggle.com/code/alexxanderlarko/protenix-v1) and the problem is solved.\nCould you please add the model?\n(protenix_base_20250630_v1.0.0 in checkpoint dir.)

", "votes": 1, "replies": [ { "id": 3407316, "author_name": "Qiwei", "author_url": "", "post_date": "2026-02-18T05:07:12.507000", "content": "

Hi, thanks!~ I created a new version Dataset named \"Protenix-v1-adjust-v2\" (https://www.kaggle.com/datasets/qiweiyin/protenix-v1-adjusted?select=Protenix-v1-adjust-v2 ) The protenix_base_20250630_v1.0.0 is in its checkpoint dir.

", "votes": 2, "replies": [] }, { "id": 3407318, "author_name": "Aleksandr Larko", "author_url": "", "post_date": "2026-02-18T05:13:24.323000", "content": "

THANK YOU!๐Ÿ™‚

", "votes": 1, "replies": [] }, { "id": 3411930, "author_name": "", "author_url": "", "post_date": "2026-02-24T00:31:35.270000", "content": "", "votes": 0, "replies": [] }, { "id": 3413023, "author_name": "Kh0a", "author_url": "", "post_date": "2026-02-24T04:50:30.877000", "content": "

may i ask where do you find the checkpoint files?

", "votes": 0, "replies": [] }, { "id": 3413037, "author_name": "", "author_url": "", "post_date": "2026-02-24T05:27:50.823000", "content": "", "votes": 0, "replies": [] }, { "id": 3413039, "author_name": "Qiwei", "author_url": "", "post_date": "2026-02-24T05:29:22.117000", "content": "

I searched the official website: https://github.com/bytedance/Protenix and found the official download links in the documents and codes. You can also download the full contents of https://github.com/bytedance/Protenix to your local machine and then search it :)

", "votes": 1, "replies": [] } ] }, { "id": 3407733, "author_name": "Kh0a", "author_url": "", "post_date": "2026-02-19T02:05:02.017000", "content": "

based on Aleksandr Larko's notebook and protenix lastest checkpoint, i have made a complete inference notebook that utilize both TBM and protenix

\n

Pure protenix scored 0.249 and scored 0.408 when mixed with TBM

", "votes": 4, "replies": [ { "id": 3407741, "author_name": "Qiwei", "author_url": "", "post_date": "2026-02-19T02:31:51.413000", "content": "

Great work๏ผ

", "votes": 2, "replies": [] } ] } ] }, { "id": 3406669, "author_name": "", "author_url": "", "post_date": "2026-02-16T09:52:21.100000", "content": "", "votes": 0, "replies": [] }, { "id": 3407377, "author_name": "Navneet", "author_url": "", "post_date": "2026-02-18T07:21:55.320000", "content": "

Thank you for this post @tonylica

", "votes": 1, "replies": [] } ], "raw_markdown_by_id": { "3406541": "Hi all โ€” does anyone have a **public, working Protenix inference notebook** for **Stanford RNA 3D Folding (Part 2)** (end-to-end โ†’ generates a valid submission)?\n\nIโ€™m interested because Protenix looks like a **promising diversity source** for ensembles. Repo:\n\nhttps://github.com/bytedance/Protenix \n\nAlso saw their new release note:\n\n> **2026-02-05: Protenix-v1 Released ๐Ÿ’ช [Technical Report]**\n> Supported Template/RNA MSA features and improved training dynamics, along with further inference-time performance enhancements.\n\nIโ€™m planning to spend some time exploring this comp and this model โ€” If anyone has a notebook link (or a minimal inference example adapted to this comp) or info about this model , Iโ€™d really appreciate it. Thanks!\n\n**Update โ€” thanks 5 kindly sharing notebooks:**\n\nLB 0.408 : https://www.kaggle.com/code/llkh0a/protenix-tbm\n\nLB 0.406: https://www.kaggle.com/code/rluethy/template-or-protenix-lb-0-4-ish?scriptVersionId=298270706\n\nLB 0.250: https://www.kaggle.com/code/zoushuxian/aido-rna-augmented-protenix\n\nDraft version: https://www.kaggle.com/code/qiweiyin/protenix-v1-inference-2026\n\nversion 2: https://www.kaggle.com/datasets/qiweiyin/protenix-v1-adjusted?select=Protenix-v1-adjust-v2 \n\n", "3406622": "a draft version: https://www.kaggle.com/code/qiweiyin/protenix-v1-inference-2026 ", "3406592": "I'm trying this. but always OOM issues!", "3406669": "", "3407377": "Thank you for this post @tonylica " } }