{ "id": 417886, "title": "What's the best public starter notebook for MatCha", "url": "/competitions/benetech-making-graphs-accessible/discussion/417886", "author_name": "Bo Peng", "post_date": "2023-06-17T16:49:24.656000", "votes": 0, "comment_count": 6, "views": 0, "content": "

What's the best public starter notebook to use for Matcha?

\n

Thank you.

", "messages": [ { "id": 2307084, "postDate": "2023-06-17T19:45:00.297Z", "content": "

There is not one specific to this competition. I wanted to share one but I felt it would be unfair because the matcha model would be in medal contention.

", "rawMarkdown": "There is not one specific to this competition. I wanted to share one but I felt it would be unfair because the matcha model would be in medal contention.", "votes": 1, "replies": [ { "id": 2307490, "postDate": "2023-06-18T08:26:13.330Z", "content": "

Hi Cody, did you start with https://www.kaggle.com/code/nbroad/donut-train-benetech as Sinan mentioned?

\n

Do you have any suggestions on what to do starting from the donut notebook? So I switch out for MatCha, should I train for more epochs, do you have any suggestions for hyperparameters, do I need colab pro or another GPU or is Kaggle kernel able to handle it?
\nShould I use an additional dataset?

\n

Appreciate the help.

\n

Thanks.

", "rawMarkdown": "Hi Cody, did you start with https://www.kaggle.com/code/nbroad/donut-train-benetech as Sinan mentioned? \n\nDo you have any suggestions on what to do starting from the donut notebook? So I switch out for MatCha, should I train for more epochs, do you have any suggestions for hyperparameters, do I need colab pro or another GPU or is Kaggle kernel able to handle it?\nShould I use an additional dataset? \n\nAppreciate the help. \n\nThanks.", "replies": [ { "id": 2307800, "postDate": "2023-06-18T12:14:06.580Z", "rawMarkdown": "", "isDeleted": true } ] } ] }, { "id": 2306940, "postDate": "2023-06-17T17:36:05.587Z", "content": "

I haven't seen a matcha specific notebook(s) but these two can give you a pretty solid baseline to start with:

\n\n

Update the model name and it pretty much works.

", "rawMarkdown": "I haven't seen a matcha specific notebook(s) but these two can give you a pretty solid baseline to start with:\n\n- https://www.kaggle.com/code/nbroad/donut-train-benetech\n- https://www.kaggle.com/code/heyytanay/training-pix2struct-pytorch-amp-w-b\n\nUpdate the model name and it pretty much works.", "votes": 1, "replies": [ { "id": 2307469, "postDate": "2023-06-18T08:07:13.747Z", "content": "

Ok appreciate it thank you. Is that what you did Sinan?

\n

Thanks for the help.

", "rawMarkdown": "Ok appreciate it thank you. Is that what you did Sinan? \n\nThanks for the help. ", "replies": [ { "id": 2308252, "postDate": "2023-06-18T19:02:14.727Z", "content": "

Kinda yes, I combined the two and followed some of the suggestions in the discussions like total number of epochs, lr, etc.

", "rawMarkdown": "Kinda yes, I combined the two and followed some of the suggestions in the discussions like total number of epochs, lr, etc." } ] } ] }, { "id": 2306865, "postDate": "2023-06-17T16:49:24.657Z", "content": "

What's the best public starter notebook to use for Matcha?

\n

Thank you.

", "rawMarkdown": "What's the best public starter notebook to use for Matcha? \n\nThank you. " } ], "comments": [ { "id": 2307084, "author_name": "Cody_Null", "author_url": "", "post_date": "2023-06-17T19:45:00.297000", "content": "

There is not one specific to this competition. I wanted to share one but I felt it would be unfair because the matcha model would be in medal contention.

", "votes": 1, "replies": [ { "id": 2307490, "author_name": "Bo Peng", "author_url": "", "post_date": "2023-06-18T08:26:13.330000", "content": "

Hi Cody, did you start with https://www.kaggle.com/code/nbroad/donut-train-benetech as Sinan mentioned?

\n

Do you have any suggestions on what to do starting from the donut notebook? So I switch out for MatCha, should I train for more epochs, do you have any suggestions for hyperparameters, do I need colab pro or another GPU or is Kaggle kernel able to handle it?
\nShould I use an additional dataset?

\n

Appreciate the help.

\n

Thanks.

", "votes": 0, "replies": [ { "id": 2307800, "author_name": "", "author_url": "", "post_date": "2023-06-18T12:14:06.580000", "content": "", "votes": 0, "replies": [] } ] } ] }, { "id": 2306940, "author_name": "Sinan Calisir", "author_url": "", "post_date": "2023-06-17T17:36:05.587000", "content": "

I haven't seen a matcha specific notebook(s) but these two can give you a pretty solid baseline to start with:

\n\n

Update the model name and it pretty much works.

", "votes": 1, "replies": [ { "id": 2307469, "author_name": "Bo Peng", "author_url": "", "post_date": "2023-06-18T08:07:13.747000", "content": "

Ok appreciate it thank you. Is that what you did Sinan?

\n

Thanks for the help.

", "votes": 0, "replies": [ { "id": 2308252, "author_name": "Sinan Calisir", "author_url": "", "post_date": "2023-06-18T19:02:14.727000", "content": "

Kinda yes, I combined the two and followed some of the suggestions in the discussions like total number of epochs, lr, etc.

", "votes": 0, "replies": [] } ] } ] } ], "raw_markdown_by_id": { "2307084": "There is not one specific to this competition. I wanted to share one but I felt it would be unfair because the matcha model would be in medal contention.", "2306940": "I haven't seen a matcha specific notebook(s) but these two can give you a pretty solid baseline to start with:\n\n- https://www.kaggle.com/code/nbroad/donut-train-benetech\n- https://www.kaggle.com/code/heyytanay/training-pix2struct-pytorch-amp-w-b\n\nUpdate the model name and it pretty much works.", "2306865": "What's the best public starter notebook to use for Matcha? \n\nThank you. " } }