Unit-3 Chapter-3 Vision Transformers Transfer learning for classification.

[Anindyadeep]

This PR is contributed by @shreydan @Anindyadeep @alanahmet @hanouticelina @asusevski @sezan92

In this PR, the following things are added:

1. The initial code for classification using transfer learning.

   1. Model: google/vit-base-patch16-224: this is the most downloaded and well-known base transformer model.
   2. Dataset: oxford-iiit pets dataset: hf-link - quite a popular dataset for image classification on the HF hub.
   3. Exclusively used HuggingFace API, that includes datasets, evaluate and transformers

2. This initial PR is merged in the forked repo after review by @Anindyadeep @sezan92 and @alanahmet

3. This also adds the code

Some questions to the reviewers in the main repo.

1. Should we only prioritize HuggingFace APIs or should we also include custom loops using PyTorch in the end.
2. For Transfer Learning / Fine-tuning, should we also add segmentation and object detection along with classification? (more explained in this issue)
3. Right now this PR only adds some codes for running, transfer learning for classification. But, how should we structure the folder? Should we follow this examples?

4. Transfer Learning
   1. Classification.mdx
   2. Object detection.mdx
   3. segmentation.mdx
      2. Fine-tuning
   4. Classification.mdx
   5. Object detection.mdx
   6. segmentation.mdx

Or, should we have follow a structure where we have two files for Transfer learning and fine-tuning and inside that, we include all of the tasks at once.

.
└── course/
    └── chapters/
        ├── chapter3/
        │   ├── Transfer Learning and Fine-tuning/
                        └── transfer-learning.mdx
                        └── fine-tuning.mdx
                        └── knowledge-distillation.mdx
        ├── chapter-n/
        │   └── ...mdx
        └── _toctree.yml

5. Right now this PR adds some code for Classification using Transfer Learning, but since this is going to be a tutorial, so how much explaination (in terms of natural words) is needed and diagrams etc.

Please note: This PR only contains the Jupyter Notebook, based on the feedback, we will start creating .mdx files for the same :)

[review-notebook-app[bot]]

Check out this pull request on 

See visual diffs & provide feedback on Jupyter Notebooks.

---

Powered by ReviewNB

[review-notebook-app[bot]]

View / edit / reply to this conversation on ReviewNB

merveenoyan commented on 2023-12-06T16:32:30Z ----------------------------------------------------------------

Maybe add some explanation on what is this notebook for and explain the end to end process that is happening in it.

Also instead of # Imports, just write "We will now import necessary modules and functions."

review-notebook-app[bot] commented 7 months ago

View / edit / reply to this conversation on ReviewNB

merveenoyan commented on 2023-12-06T16:32:31Z ----------------------------------------------------------------

For which task?

review-notebook-app[bot] commented 7 months ago

View / edit / reply to this conversation on ReviewNB

merveenoyan commented on 2023-12-06T16:32:32Z ----------------------------------------------------------------

maybe mention this is also necessary to pass to the model since model doesn't accept string classes. (you're mentioning briefly below, you can also have this section below instead if you're not using id2label yet).

review-notebook-app[bot] commented 7 months ago

View / edit / reply to this conversation on ReviewNB

merveenoyan commented on 2023-12-06T16:32:32Z ----------------------------------------------------------------

"Let's see some sample images from the dataset."

review-notebook-app[bot] commented 7 months ago

View / edit / reply to this conversation on ReviewNB

merveenoyan commented on 2023-12-06T16:32:33Z ----------------------------------------------------------------

I saw here you're doing evaluation split but not doing end of training validation with a separate split, maybe you could have a separate split for this here (if the dataset doesn't come with a final test split)

review-notebook-app[bot] commented 7 months ago

View / edit / reply to this conversation on ReviewNB

merveenoyan commented on 2023-12-06T16:32:34Z ----------------------------------------------------------------

suggestions for clarity:

"It might be possible that some images in your dataset will be grayscale or transparent (RGBA)."

"We pass the images through the processor to apply the transforms to process and convert them into PyTorch format."

also you could carry label2id here instead

review-notebook-app[bot] commented 7 months ago

View / edit / reply to this conversation on ReviewNB

merveenoyan commented on 2023-12-06T16:32:36Z ----------------------------------------------------------------

We can set push_to_hub to True and later call push_to_hub on Trainer after the save so people can learn how to use HF Hub as model registry.

MKhalusova commented on 2023-12-08T17:56:01Z ----------------------------------------------------------------

If we set push_to_hub to True, we also need to add a call with notebook_login:

from huggingface_hub import notebook_login
notebook_login()

In this case, the output_dir will also be the name of the repo where your model checkpoint will be pushed.

review-notebook-app[bot] commented 7 months ago

View / edit / reply to this conversation on ReviewNB

merveenoyan commented on 2023-12-06T16:32:37Z ----------------------------------------------------------------

We will now save the model.

review-notebook-app[bot] commented 7 months ago

View / edit / reply to this conversation on ReviewNB

merveenoyan commented on 2023-12-06T16:32:38Z ----------------------------------------------------------------

We can also push this repository to https://huggingface.co/hf-vision organization so people can checkout tensorboard logs and the model card.

shreydan commented 7 months ago

I checked the # trainable params with sum([p.numel() for p in model.parameters() if p.requires_grad])

and all the params were trainable. So had to freeze the weights manually.

---

View entire conversation on ReviewNB

shreydan commented 7 months ago

Quite end-to-end and nicely put 💜 left some comments on reviewNB to structure the notebook and also added some comments about using transformers 🤗

Thank you very much for the review @merveenoyan! Will keep all your detailed and helpful suggestions in mind, this was an initial PR to understand more about how everything should be arranged.

I will update this notebook accordingly and also add the mdx file. Thank you! 🤗

MKhalusova commented 7 months ago

If we set push_to_hub to True, we also need to add a call with notebook_login:

from huggingface_hub import notebook_login
notebook_login()

In this case, the output_dir will also be the name of the repo where your model checkpoint will be pushed.

---

View entire conversation on ReviewNB

johko commented 7 months ago

Nice Demo @shreydan, as Merve already suggested in some of the comments - feel free to make it a bit more conversational. The focus should be on the code, but there is no harm in writing some more text here and there :slightly_smiling_face:

Keep it up :+1:

shreydan commented 6 months ago

@MKhalusova @johko @merveenoyan I've made further updates to my PR as suggested.

Please check the image-classification notebooks:

• notebooks/Unit 3 - Vision Transformers/fine-tuning-multilabel-image-classification.ipynb
• notebooks/Unit 3 - Vision Transformers/transfer-learning-image-classification.ipynb

For transfer-learning:

• updated notebook as per the reviews before
• added pushing model to hub + inference visualization

For fine-tuning:

• new notebook with timm, accelerate, evaluate for multi-label image classification
• detailed comments and explanations on writing training loops
• inference visualization

Based on further reviews, I was thinking to then create a .mdx file which contains both transfer learning and fine-tuning parts.

merveenoyan commented 6 months ago

@Anindyadeep can you tell me which files are contributed through this PR since I remember I reviewed some of the notebooks in other PRs that seems like a change in this PR. It would be easier for us to review.

Anindyadeep commented 6 months ago

@Anindyadeep can you tell me which files are contributed through this PR since I remember I reviewed some of the notebooks in other PRs that seems like a change in this PR. It would be easier for us to review.

Yeah, so all the sub-topics this includes:

1. Fine-tuning for object detection
2. Lora for Image classification
3. Transfer Learning for Segmentation
4. Knowledge distillation
5. Transfer learning for segmentation
6. Fine-tuning multi class classification

All of these topics are included in the PR as a whole. So, some how I am not sure, but when we merged from @shreydan's main to this repo (after our initial peer reviews), it moved all the new files under this PR.

I got to know the reason, and this could be because the initial PR was set from @shreydan main to this main (for image classification only), and now that we were following the convention of first reviewing from our main and then merge for main review, it automatically collects the file and saves here.

shreydan commented 6 months ago

@MKhalusova @merveenoyan @johko

Sorry for this confusion, this PR is for the image classification part under unit-3 vision transformers : fine-tuning and transfer-learning.

We have other PRs for segmentation, object detection, LoRA and Knowledge distillation.

For this PR:

Please check the image-classification notebooks:

- notebooks/Unit 3 - Vision Transformers/fine-tuning-multilabel-image-classification.ipynb
- notebooks/Unit 3 - Vision Transformers/transfer-learning-image-classification.ipynb

For image classification, transfer-learning:

• updated notebook as per the reviews before
• added pushing model to hub + inference visualization

For image classification, fine-tuning:

• new notebook with timm, accelerate, evaluate for multi-label image classification
• detailed comments and explanations on writing training loops
• inference visualization

Based on further reviews, I was thinking to then create a .mdx file which contains both transfer learning and fine-tuning parts.

merveenoyan commented 6 months ago

Hey @shreydan 👋 We currently have transfer learning one and LoRA one already merged. Have you renamed the transfer learning one, it should've been in this PR somewhere. Moreover, can you check if your PR aligns with the merged notebooks? I will check which other PRs you have and merge them and come back to this if it's ok.

shreydan commented 6 months ago

@merveenoyan sorry didn't know the transfer learning file was merged, actually I have renamed it to transfer-learning-image-classification.ipynb since there were object detection and segmentation tasks as well.

merveenoyan commented 6 months ago

@shreydan if you can change the name of the file we can change it in another PR

merveenoyan commented 6 months ago

@shreydan we can first wait for other PRs to be merged and then merge this if it's ok for you

shreydan commented 6 months ago

@shreydan if you can change the name of the file we can change it in another PR @merveenoyan what should I rename the file to?

@shreydan we can first wait for other PRs to be merged and then merge this if it's ok for you

okay no issues in that. Sorry for this confusion.

merveenoyan commented 6 months ago

@Anindyadeep @shreydan I think we can merge this one. If you could sync the changes with this repository and solve merge conflicts I'll give a one last review 🤓

Anindyadeep commented 6 months ago

Sure we can do that, and then let's get it merged :)

merveenoyan commented 5 months ago

@Anindyadeep do you need help with this?

Anindyadeep commented 5 months ago

Hey @merveenoyan, so sorry for the delay, I am bit worked up with some priorities, give me few days I along with @shreydan will solve this very soon :)

Anindyadeep commented 5 months ago

Hi @merveenoyan, sorry for the delay, @shreydan has helped us by updating and removing the merge conflicts. Let us know what else is needed for the merge.

shreydan commented 5 months ago

hello @merveenoyan sorry for the conflicts, our unit had multiple PRs hence the problems. 1 old notebook was removed and has been updated with 2 new notebooks with regards to ViT image classification (transfer learning & fine tuning). Would really appreciate reviews on these 2 notebooks. Thank you for your patience

merveenoyan commented 5 months ago

@MKhalusova can you skim through the notebooks? it seemed ready to me

johko commented 5 months ago

Merging :rocket:

shreydan commented 5 months ago

@johko @MKhalusova @merveenoyan thank you very much!

• © Githubissues.
• Githubissues is a development platform for aggregating issues.

[review-notebook-app[bot]]

View / edit / reply to this conversation on ReviewNB

merveenoyan commented on 2023-12-06T16:32:31Z ----------------------------------------------------------------

For which task?

review-notebook-app[bot] commented 7 months ago

View / edit / reply to this conversation on ReviewNB

merveenoyan commented on 2023-12-06T16:32:32Z ----------------------------------------------------------------

maybe mention this is also necessary to pass to the model since model doesn't accept string classes. (you're mentioning briefly below, you can also have this section below instead if you're not using id2label yet).

review-notebook-app[bot] commented 7 months ago

View / edit / reply to this conversation on ReviewNB

merveenoyan commented on 2023-12-06T16:32:32Z ----------------------------------------------------------------

"Let's see some sample images from the dataset."

review-notebook-app[bot] commented 7 months ago

View / edit / reply to this conversation on ReviewNB

merveenoyan commented on 2023-12-06T16:32:33Z ----------------------------------------------------------------

I saw here you're doing evaluation split but not doing end of training validation with a separate split, maybe you could have a separate split for this here (if the dataset doesn't come with a final test split)

review-notebook-app[bot] commented 7 months ago

View / edit / reply to this conversation on ReviewNB

merveenoyan commented on 2023-12-06T16:32:34Z ----------------------------------------------------------------

suggestions for clarity:

"It might be possible that some images in your dataset will be grayscale or transparent (RGBA)."

"We pass the images through the processor to apply the transforms to process and convert them into PyTorch format."

also you could carry label2id here instead

review-notebook-app[bot] commented 7 months ago

View / edit / reply to this conversation on ReviewNB

merveenoyan commented on 2023-12-06T16:32:36Z ----------------------------------------------------------------

We can set push_to_hub to True and later call push_to_hub on Trainer after the save so people can learn how to use HF Hub as model registry.

MKhalusova commented on 2023-12-08T17:56:01Z ----------------------------------------------------------------

If we set push_to_hub to True, we also need to add a call with notebook_login:

from huggingface_hub import notebook_login
notebook_login()

In this case, the output_dir will also be the name of the repo where your model checkpoint will be pushed.

review-notebook-app[bot] commented 7 months ago

View / edit / reply to this conversation on ReviewNB

merveenoyan commented on 2023-12-06T16:32:37Z ----------------------------------------------------------------

We will now save the model.

review-notebook-app[bot] commented 7 months ago

View / edit / reply to this conversation on ReviewNB

merveenoyan commented on 2023-12-06T16:32:38Z ----------------------------------------------------------------

We can also push this repository to https://huggingface.co/hf-vision organization so people can checkout tensorboard logs and the model card.

shreydan commented 7 months ago

I checked the # trainable params with sum([p.numel() for p in model.parameters() if p.requires_grad])

and all the params were trainable. So had to freeze the weights manually.

---

View entire conversation on ReviewNB

shreydan commented 7 months ago

Quite end-to-end and nicely put 💜 left some comments on reviewNB to structure the notebook and also added some comments about using transformers 🤗

Thank you very much for the review @merveenoyan! Will keep all your detailed and helpful suggestions in mind, this was an initial PR to understand more about how everything should be arranged.

I will update this notebook accordingly and also add the mdx file. Thank you! 🤗

MKhalusova commented 7 months ago

If we set push_to_hub to True, we also need to add a call with notebook_login:

from huggingface_hub import notebook_login
notebook_login()

In this case, the output_dir will also be the name of the repo where your model checkpoint will be pushed.

---

View entire conversation on ReviewNB

johko commented 7 months ago

Nice Demo @shreydan, as Merve already suggested in some of the comments - feel free to make it a bit more conversational. The focus should be on the code, but there is no harm in writing some more text here and there :slightly_smiling_face:

Keep it up :+1:

shreydan commented 6 months ago

@MKhalusova @johko @merveenoyan I've made further updates to my PR as suggested.

Please check the image-classification notebooks:

• notebooks/Unit 3 - Vision Transformers/fine-tuning-multilabel-image-classification.ipynb
• notebooks/Unit 3 - Vision Transformers/transfer-learning-image-classification.ipynb

For transfer-learning:

• updated notebook as per the reviews before
• added pushing model to hub + inference visualization

For fine-tuning:

• new notebook with timm, accelerate, evaluate for multi-label image classification
• detailed comments and explanations on writing training loops
• inference visualization

Based on further reviews, I was thinking to then create a .mdx file which contains both transfer learning and fine-tuning parts.

merveenoyan commented 6 months ago

@Anindyadeep can you tell me which files are contributed through this PR since I remember I reviewed some of the notebooks in other PRs that seems like a change in this PR. It would be easier for us to review.

Anindyadeep commented 6 months ago

@Anindyadeep can you tell me which files are contributed through this PR since I remember I reviewed some of the notebooks in other PRs that seems like a change in this PR. It would be easier for us to review.

Yeah, so all the sub-topics this includes:

1. Fine-tuning for object detection
2. Lora for Image classification
3. Transfer Learning for Segmentation
4. Knowledge distillation
5. Transfer learning for segmentation
6. Fine-tuning multi class classification

All of these topics are included in the PR as a whole. So, some how I am not sure, but when we merged from @shreydan's main to this repo (after our initial peer reviews), it moved all the new files under this PR.

I got to know the reason, and this could be because the initial PR was set from @shreydan main to this main (for image classification only), and now that we were following the convention of first reviewing from our main and then merge for main review, it automatically collects the file and saves here.

shreydan commented 6 months ago

@MKhalusova @merveenoyan @johko

Sorry for this confusion, this PR is for the image classification part under unit-3 vision transformers : fine-tuning and transfer-learning.

We have other PRs for segmentation, object detection, LoRA and Knowledge distillation.

For this PR:

Please check the image-classification notebooks:

- notebooks/Unit 3 - Vision Transformers/fine-tuning-multilabel-image-classification.ipynb
- notebooks/Unit 3 - Vision Transformers/transfer-learning-image-classification.ipynb

For image classification, transfer-learning:

• updated notebook as per the reviews before
• added pushing model to hub + inference visualization

For image classification, fine-tuning:

• new notebook with timm, accelerate, evaluate for multi-label image classification
• detailed comments and explanations on writing training loops
• inference visualization

Based on further reviews, I was thinking to then create a .mdx file which contains both transfer learning and fine-tuning parts.

merveenoyan commented 6 months ago

Hey @shreydan 👋 We currently have transfer learning one and LoRA one already merged. Have you renamed the transfer learning one, it should've been in this PR somewhere. Moreover, can you check if your PR aligns with the merged notebooks? I will check which other PRs you have and merge them and come back to this if it's ok.

shreydan commented 6 months ago

@merveenoyan sorry didn't know the transfer learning file was merged, actually I have renamed it to transfer-learning-image-classification.ipynb since there were object detection and segmentation tasks as well.

merveenoyan commented 6 months ago

@shreydan if you can change the name of the file we can change it in another PR

merveenoyan commented 6 months ago

@shreydan we can first wait for other PRs to be merged and then merge this if it's ok for you

shreydan commented 6 months ago

@shreydan if you can change the name of the file we can change it in another PR @merveenoyan what should I rename the file to?

@shreydan we can first wait for other PRs to be merged and then merge this if it's ok for you

okay no issues in that. Sorry for this confusion.

merveenoyan commented 6 months ago

@Anindyadeep @shreydan I think we can merge this one. If you could sync the changes with this repository and solve merge conflicts I'll give a one last review 🤓

Anindyadeep commented 6 months ago

Sure we can do that, and then let's get it merged :)

merveenoyan commented 5 months ago

@Anindyadeep do you need help with this?

Anindyadeep commented 5 months ago

Hey @merveenoyan, so sorry for the delay, I am bit worked up with some priorities, give me few days I along with @shreydan will solve this very soon :)

Anindyadeep commented 5 months ago

Hi @merveenoyan, sorry for the delay, @shreydan has helped us by updating and removing the merge conflicts. Let us know what else is needed for the merge.

shreydan commented 5 months ago

hello @merveenoyan sorry for the conflicts, our unit had multiple PRs hence the problems. 1 old notebook was removed and has been updated with 2 new notebooks with regards to ViT image classification (transfer learning & fine tuning). Would really appreciate reviews on these 2 notebooks. Thank you for your patience

merveenoyan commented 5 months ago

@MKhalusova can you skim through the notebooks? it seemed ready to me

johko commented 5 months ago

Merging :rocket:

shreydan commented 5 months ago

@johko @MKhalusova @merveenoyan thank you very much!

• © Githubissues.
• Githubissues is a development platform for aggregating issues.

[review-notebook-app[bot]]

View / edit / reply to this conversation on ReviewNB

merveenoyan commented on 2023-12-06T16:32:32Z ----------------------------------------------------------------

maybe mention this is also necessary to pass to the model since model doesn't accept string classes. (you're mentioning briefly below, you can also have this section below instead if you're not using id2label yet).

review-notebook-app[bot] commented 7 months ago

View / edit / reply to this conversation on ReviewNB

merveenoyan commented on 2023-12-06T16:32:32Z ----------------------------------------------------------------

"Let's see some sample images from the dataset."

review-notebook-app[bot] commented 7 months ago

View / edit / reply to this conversation on ReviewNB

merveenoyan commented on 2023-12-06T16:32:33Z ----------------------------------------------------------------

I saw here you're doing evaluation split but not doing end of training validation with a separate split, maybe you could have a separate split for this here (if the dataset doesn't come with a final test split)

review-notebook-app[bot] commented 7 months ago

View / edit / reply to this conversation on ReviewNB

merveenoyan commented on 2023-12-06T16:32:34Z ----------------------------------------------------------------

suggestions for clarity:

"It might be possible that some images in your dataset will be grayscale or transparent (RGBA)."

"We pass the images through the processor to apply the transforms to process and convert them into PyTorch format."

also you could carry label2id here instead

review-notebook-app[bot] commented 7 months ago

View / edit / reply to this conversation on ReviewNB

merveenoyan commented on 2023-12-06T16:32:36Z ----------------------------------------------------------------

We can set push_to_hub to True and later call push_to_hub on Trainer after the save so people can learn how to use HF Hub as model registry.

MKhalusova commented on 2023-12-08T17:56:01Z ----------------------------------------------------------------

If we set push_to_hub to True, we also need to add a call with notebook_login:

from huggingface_hub import notebook_login
notebook_login()

In this case, the output_dir will also be the name of the repo where your model checkpoint will be pushed.

review-notebook-app[bot] commented 7 months ago

View / edit / reply to this conversation on ReviewNB

merveenoyan commented on 2023-12-06T16:32:37Z ----------------------------------------------------------------

We will now save the model.

review-notebook-app[bot] commented 7 months ago

View / edit / reply to this conversation on ReviewNB

merveenoyan commented on 2023-12-06T16:32:38Z ----------------------------------------------------------------

We can also push this repository to https://huggingface.co/hf-vision organization so people can checkout tensorboard logs and the model card.

shreydan commented 7 months ago

I checked the # trainable params with sum([p.numel() for p in model.parameters() if p.requires_grad])

and all the params were trainable. So had to freeze the weights manually.

---

View entire conversation on ReviewNB

shreydan commented 7 months ago

Quite end-to-end and nicely put 💜 left some comments on reviewNB to structure the notebook and also added some comments about using transformers 🤗

Thank you very much for the review @merveenoyan! Will keep all your detailed and helpful suggestions in mind, this was an initial PR to understand more about how everything should be arranged.

I will update this notebook accordingly and also add the mdx file. Thank you! 🤗

MKhalusova commented 7 months ago

If we set push_to_hub to True, we also need to add a call with notebook_login:

from huggingface_hub import notebook_login
notebook_login()

In this case, the output_dir will also be the name of the repo where your model checkpoint will be pushed.

---

View entire conversation on ReviewNB

johko commented 7 months ago

Nice Demo @shreydan, as Merve already suggested in some of the comments - feel free to make it a bit more conversational. The focus should be on the code, but there is no harm in writing some more text here and there :slightly_smiling_face:

Keep it up :+1:

shreydan commented 6 months ago

@MKhalusova @johko @merveenoyan I've made further updates to my PR as suggested.

Please check the image-classification notebooks:

• notebooks/Unit 3 - Vision Transformers/fine-tuning-multilabel-image-classification.ipynb
• notebooks/Unit 3 - Vision Transformers/transfer-learning-image-classification.ipynb

For transfer-learning:

• updated notebook as per the reviews before
• added pushing model to hub + inference visualization

For fine-tuning:

• new notebook with timm, accelerate, evaluate for multi-label image classification
• detailed comments and explanations on writing training loops
• inference visualization

Based on further reviews, I was thinking to then create a .mdx file which contains both transfer learning and fine-tuning parts.

merveenoyan commented 6 months ago

@Anindyadeep can you tell me which files are contributed through this PR since I remember I reviewed some of the notebooks in other PRs that seems like a change in this PR. It would be easier for us to review.

Anindyadeep commented 6 months ago

@Anindyadeep can you tell me which files are contributed through this PR since I remember I reviewed some of the notebooks in other PRs that seems like a change in this PR. It would be easier for us to review.

Yeah, so all the sub-topics this includes:

1. Fine-tuning for object detection
2. Lora for Image classification
3. Transfer Learning for Segmentation
4. Knowledge distillation
5. Transfer learning for segmentation
6. Fine-tuning multi class classification

All of these topics are included in the PR as a whole. So, some how I am not sure, but when we merged from @shreydan's main to this repo (after our initial peer reviews), it moved all the new files under this PR.

I got to know the reason, and this could be because the initial PR was set from @shreydan main to this main (for image classification only), and now that we were following the convention of first reviewing from our main and then merge for main review, it automatically collects the file and saves here.

shreydan commented 6 months ago

@MKhalusova @merveenoyan @johko

Sorry for this confusion, this PR is for the image classification part under unit-3 vision transformers : fine-tuning and transfer-learning.

We have other PRs for segmentation, object detection, LoRA and Knowledge distillation.

For this PR:

Please check the image-classification notebooks:

- notebooks/Unit 3 - Vision Transformers/fine-tuning-multilabel-image-classification.ipynb
- notebooks/Unit 3 - Vision Transformers/transfer-learning-image-classification.ipynb

For image classification, transfer-learning:

• updated notebook as per the reviews before
• added pushing model to hub + inference visualization

For image classification, fine-tuning:

• new notebook with timm, accelerate, evaluate for multi-label image classification
• detailed comments and explanations on writing training loops
• inference visualization

Based on further reviews, I was thinking to then create a .mdx file which contains both transfer learning and fine-tuning parts.

merveenoyan commented 6 months ago

Hey @shreydan 👋 We currently have transfer learning one and LoRA one already merged. Have you renamed the transfer learning one, it should've been in this PR somewhere. Moreover, can you check if your PR aligns with the merged notebooks? I will check which other PRs you have and merge them and come back to this if it's ok.

shreydan commented 6 months ago

@merveenoyan sorry didn't know the transfer learning file was merged, actually I have renamed it to transfer-learning-image-classification.ipynb since there were object detection and segmentation tasks as well.

merveenoyan commented 6 months ago

@shreydan if you can change the name of the file we can change it in another PR

merveenoyan commented 6 months ago

@shreydan we can first wait for other PRs to be merged and then merge this if it's ok for you

shreydan commented 6 months ago

@shreydan if you can change the name of the file we can change it in another PR @merveenoyan what should I rename the file to?

@shreydan we can first wait for other PRs to be merged and then merge this if it's ok for you

okay no issues in that. Sorry for this confusion.

merveenoyan commented 6 months ago

@Anindyadeep @shreydan I think we can merge this one. If you could sync the changes with this repository and solve merge conflicts I'll give a one last review 🤓

Anindyadeep commented 6 months ago

Sure we can do that, and then let's get it merged :)

merveenoyan commented 5 months ago

@Anindyadeep do you need help with this?

Anindyadeep commented 5 months ago

Hey @merveenoyan, so sorry for the delay, I am bit worked up with some priorities, give me few days I along with @shreydan will solve this very soon :)

Anindyadeep commented 5 months ago

Hi @merveenoyan, sorry for the delay, @shreydan has helped us by updating and removing the merge conflicts. Let us know what else is needed for the merge.

shreydan commented 5 months ago

hello @merveenoyan sorry for the conflicts, our unit had multiple PRs hence the problems. 1 old notebook was removed and has been updated with 2 new notebooks with regards to ViT image classification (transfer learning & fine tuning). Would really appreciate reviews on these 2 notebooks. Thank you for your patience

merveenoyan commented 5 months ago

@MKhalusova can you skim through the notebooks? it seemed ready to me

johko commented 5 months ago

Merging :rocket:

shreydan commented 5 months ago

@johko @MKhalusova @merveenoyan thank you very much!

• © Githubissues.
• Githubissues is a development platform for aggregating issues.

[review-notebook-app[bot]]

View / edit / reply to this conversation on ReviewNB

merveenoyan commented on 2023-12-06T16:32:32Z ----------------------------------------------------------------

"Let's see some sample images from the dataset."

review-notebook-app[bot] commented 7 months ago

View / edit / reply to this conversation on ReviewNB

merveenoyan commented on 2023-12-06T16:32:33Z ----------------------------------------------------------------

I saw here you're doing evaluation split but not doing end of training validation with a separate split, maybe you could have a separate split for this here (if the dataset doesn't come with a final test split)

review-notebook-app[bot] commented 7 months ago

View / edit / reply to this conversation on ReviewNB

merveenoyan commented on 2023-12-06T16:32:34Z ----------------------------------------------------------------

suggestions for clarity:

"It might be possible that some images in your dataset will be grayscale or transparent (RGBA)."

"We pass the images through the processor to apply the transforms to process and convert them into PyTorch format."

also you could carry label2id here instead

review-notebook-app[bot] commented 7 months ago

View / edit / reply to this conversation on ReviewNB

merveenoyan commented on 2023-12-06T16:32:36Z ----------------------------------------------------------------

We can set push_to_hub to True and later call push_to_hub on Trainer after the save so people can learn how to use HF Hub as model registry.

MKhalusova commented on 2023-12-08T17:56:01Z ----------------------------------------------------------------

If we set push_to_hub to True, we also need to add a call with notebook_login:

from huggingface_hub import notebook_login
notebook_login()

In this case, the output_dir will also be the name of the repo where your model checkpoint will be pushed.

review-notebook-app[bot] commented 7 months ago

View / edit / reply to this conversation on ReviewNB

merveenoyan commented on 2023-12-06T16:32:37Z ----------------------------------------------------------------

We will now save the model.

review-notebook-app[bot] commented 7 months ago

View / edit / reply to this conversation on ReviewNB

merveenoyan commented on 2023-12-06T16:32:38Z ----------------------------------------------------------------

We can also push this repository to https://huggingface.co/hf-vision organization so people can checkout tensorboard logs and the model card.

shreydan commented 7 months ago

I checked the # trainable params with sum([p.numel() for p in model.parameters() if p.requires_grad])

and all the params were trainable. So had to freeze the weights manually.

---

View entire conversation on ReviewNB

shreydan commented 7 months ago

Quite end-to-end and nicely put 💜 left some comments on reviewNB to structure the notebook and also added some comments about using transformers 🤗

Thank you very much for the review @merveenoyan! Will keep all your detailed and helpful suggestions in mind, this was an initial PR to understand more about how everything should be arranged.

I will update this notebook accordingly and also add the mdx file. Thank you! 🤗

MKhalusova commented 7 months ago

If we set push_to_hub to True, we also need to add a call with notebook_login:

from huggingface_hub import notebook_login
notebook_login()

In this case, the output_dir will also be the name of the repo where your model checkpoint will be pushed.

---

View entire conversation on ReviewNB

johko commented 7 months ago

Nice Demo @shreydan, as Merve already suggested in some of the comments - feel free to make it a bit more conversational. The focus should be on the code, but there is no harm in writing some more text here and there :slightly_smiling_face:

Keep it up :+1:

shreydan commented 6 months ago

@MKhalusova @johko @merveenoyan I've made further updates to my PR as suggested.

Please check the image-classification notebooks:

• notebooks/Unit 3 - Vision Transformers/fine-tuning-multilabel-image-classification.ipynb
• notebooks/Unit 3 - Vision Transformers/transfer-learning-image-classification.ipynb

For transfer-learning:

• updated notebook as per the reviews before
• added pushing model to hub + inference visualization

For fine-tuning:

• new notebook with timm, accelerate, evaluate for multi-label image classification
• detailed comments and explanations on writing training loops
• inference visualization

Based on further reviews, I was thinking to then create a .mdx file which contains both transfer learning and fine-tuning parts.

merveenoyan commented 6 months ago

@Anindyadeep can you tell me which files are contributed through this PR since I remember I reviewed some of the notebooks in other PRs that seems like a change in this PR. It would be easier for us to review.

Anindyadeep commented 6 months ago

@Anindyadeep can you tell me which files are contributed through this PR since I remember I reviewed some of the notebooks in other PRs that seems like a change in this PR. It would be easier for us to review.

Yeah, so all the sub-topics this includes:

1. Fine-tuning for object detection
2. Lora for Image classification
3. Transfer Learning for Segmentation
4. Knowledge distillation
5. Transfer learning for segmentation
6. Fine-tuning multi class classification

All of these topics are included in the PR as a whole. So, some how I am not sure, but when we merged from @shreydan's main to this repo (after our initial peer reviews), it moved all the new files under this PR.

I got to know the reason, and this could be because the initial PR was set from @shreydan main to this main (for image classification only), and now that we were following the convention of first reviewing from our main and then merge for main review, it automatically collects the file and saves here.

shreydan commented 6 months ago

@MKhalusova @merveenoyan @johko

Sorry for this confusion, this PR is for the image classification part under unit-3 vision transformers : fine-tuning and transfer-learning.

We have other PRs for segmentation, object detection, LoRA and Knowledge distillation.

For this PR:

Please check the image-classification notebooks:

- notebooks/Unit 3 - Vision Transformers/fine-tuning-multilabel-image-classification.ipynb
- notebooks/Unit 3 - Vision Transformers/transfer-learning-image-classification.ipynb

For image classification, transfer-learning:

• updated notebook as per the reviews before
• added pushing model to hub + inference visualization

For image classification, fine-tuning:

• new notebook with timm, accelerate, evaluate for multi-label image classification
• detailed comments and explanations on writing training loops
• inference visualization

Based on further reviews, I was thinking to then create a .mdx file which contains both transfer learning and fine-tuning parts.

merveenoyan commented 6 months ago

Hey @shreydan 👋 We currently have transfer learning one and LoRA one already merged. Have you renamed the transfer learning one, it should've been in this PR somewhere. Moreover, can you check if your PR aligns with the merged notebooks? I will check which other PRs you have and merge them and come back to this if it's ok.

shreydan commented 6 months ago

@merveenoyan sorry didn't know the transfer learning file was merged, actually I have renamed it to transfer-learning-image-classification.ipynb since there were object detection and segmentation tasks as well.

merveenoyan commented 6 months ago

@shreydan if you can change the name of the file we can change it in another PR

merveenoyan commented 6 months ago

@shreydan we can first wait for other PRs to be merged and then merge this if it's ok for you

shreydan commented 6 months ago

@shreydan if you can change the name of the file we can change it in another PR @merveenoyan what should I rename the file to?

@shreydan we can first wait for other PRs to be merged and then merge this if it's ok for you

okay no issues in that. Sorry for this confusion.

merveenoyan commented 6 months ago

@Anindyadeep @shreydan I think we can merge this one. If you could sync the changes with this repository and solve merge conflicts I'll give a one last review 🤓

Anindyadeep commented 6 months ago

Sure we can do that, and then let's get it merged :)

merveenoyan commented 5 months ago

@Anindyadeep do you need help with this?

Anindyadeep commented 5 months ago

Hey @merveenoyan, so sorry for the delay, I am bit worked up with some priorities, give me few days I along with @shreydan will solve this very soon :)

Anindyadeep commented 5 months ago

Hi @merveenoyan, sorry for the delay, @shreydan has helped us by updating and removing the merge conflicts. Let us know what else is needed for the merge.

shreydan commented 5 months ago

hello @merveenoyan sorry for the conflicts, our unit had multiple PRs hence the problems. 1 old notebook was removed and has been updated with 2 new notebooks with regards to ViT image classification (transfer learning & fine tuning). Would really appreciate reviews on these 2 notebooks. Thank you for your patience

merveenoyan commented 5 months ago

@MKhalusova can you skim through the notebooks? it seemed ready to me

johko commented 5 months ago

Merging :rocket:

shreydan commented 5 months ago

@johko @MKhalusova @merveenoyan thank you very much!

• © Githubissues.
• Githubissues is a development platform for aggregating issues.

[review-notebook-app[bot]]

View / edit / reply to this conversation on ReviewNB

merveenoyan commented on 2023-12-06T16:32:33Z ----------------------------------------------------------------

I saw here you're doing evaluation split but not doing end of training validation with a separate split, maybe you could have a separate split for this here (if the dataset doesn't come with a final test split)

review-notebook-app[bot] commented 7 months ago

View / edit / reply to this conversation on ReviewNB

merveenoyan commented on 2023-12-06T16:32:34Z ----------------------------------------------------------------

suggestions for clarity:

"It might be possible that some images in your dataset will be grayscale or transparent (RGBA)."

"We pass the images through the processor to apply the transforms to process and convert them into PyTorch format."

also you could carry label2id here instead

review-notebook-app[bot] commented 7 months ago

View / edit / reply to this conversation on ReviewNB

merveenoyan commented on 2023-12-06T16:32:36Z ----------------------------------------------------------------

We can set push_to_hub to True and later call push_to_hub on Trainer after the save so people can learn how to use HF Hub as model registry.

MKhalusova commented on 2023-12-08T17:56:01Z ----------------------------------------------------------------

If we set push_to_hub to True, we also need to add a call with notebook_login:

from huggingface_hub import notebook_login
notebook_login()

In this case, the output_dir will also be the name of the repo where your model checkpoint will be pushed.

review-notebook-app[bot] commented 7 months ago

View / edit / reply to this conversation on ReviewNB

merveenoyan commented on 2023-12-06T16:32:37Z ----------------------------------------------------------------

We will now save the model.

review-notebook-app[bot] commented 7 months ago

View / edit / reply to this conversation on ReviewNB

merveenoyan commented on 2023-12-06T16:32:38Z ----------------------------------------------------------------

We can also push this repository to https://huggingface.co/hf-vision organization so people can checkout tensorboard logs and the model card.

shreydan commented 7 months ago

I checked the # trainable params with sum([p.numel() for p in model.parameters() if p.requires_grad])

and all the params were trainable. So had to freeze the weights manually.

---

View entire conversation on ReviewNB

shreydan commented 7 months ago

Quite end-to-end and nicely put 💜 left some comments on reviewNB to structure the notebook and also added some comments about using transformers 🤗

Thank you very much for the review @merveenoyan! Will keep all your detailed and helpful suggestions in mind, this was an initial PR to understand more about how everything should be arranged.

I will update this notebook accordingly and also add the mdx file. Thank you! 🤗

MKhalusova commented 7 months ago

If we set push_to_hub to True, we also need to add a call with notebook_login:

from huggingface_hub import notebook_login
notebook_login()

In this case, the output_dir will also be the name of the repo where your model checkpoint will be pushed.

---

View entire conversation on ReviewNB

johko commented 7 months ago

Nice Demo @shreydan, as Merve already suggested in some of the comments - feel free to make it a bit more conversational. The focus should be on the code, but there is no harm in writing some more text here and there :slightly_smiling_face:

Keep it up :+1:

shreydan commented 6 months ago

@MKhalusova @johko @merveenoyan I've made further updates to my PR as suggested.

Please check the image-classification notebooks:

• notebooks/Unit 3 - Vision Transformers/fine-tuning-multilabel-image-classification.ipynb
• notebooks/Unit 3 - Vision Transformers/transfer-learning-image-classification.ipynb

For transfer-learning:

• updated notebook as per the reviews before
• added pushing model to hub + inference visualization

For fine-tuning:

• new notebook with timm, accelerate, evaluate for multi-label image classification
• detailed comments and explanations on writing training loops
• inference visualization

Based on further reviews, I was thinking to then create a .mdx file which contains both transfer learning and fine-tuning parts.

merveenoyan commented 6 months ago

@Anindyadeep can you tell me which files are contributed through this PR since I remember I reviewed some of the notebooks in other PRs that seems like a change in this PR. It would be easier for us to review.

Anindyadeep commented 6 months ago

@Anindyadeep can you tell me which files are contributed through this PR since I remember I reviewed some of the notebooks in other PRs that seems like a change in this PR. It would be easier for us to review.

Yeah, so all the sub-topics this includes:

1. Fine-tuning for object detection
2. Lora for Image classification
3. Transfer Learning for Segmentation
4. Knowledge distillation
5. Transfer learning for segmentation
6. Fine-tuning multi class classification

All of these topics are included in the PR as a whole. So, some how I am not sure, but when we merged from @shreydan's main to this repo (after our initial peer reviews), it moved all the new files under this PR.

I got to know the reason, and this could be because the initial PR was set from @shreydan main to this main (for image classification only), and now that we were following the convention of first reviewing from our main and then merge for main review, it automatically collects the file and saves here.

shreydan commented 6 months ago

@MKhalusova @merveenoyan @johko

Sorry for this confusion, this PR is for the image classification part under unit-3 vision transformers : fine-tuning and transfer-learning.

We have other PRs for segmentation, object detection, LoRA and Knowledge distillation.

For this PR:

Please check the image-classification notebooks:

- notebooks/Unit 3 - Vision Transformers/fine-tuning-multilabel-image-classification.ipynb
- notebooks/Unit 3 - Vision Transformers/transfer-learning-image-classification.ipynb

For image classification, transfer-learning:

• updated notebook as per the reviews before
• added pushing model to hub + inference visualization

For image classification, fine-tuning:

• new notebook with timm, accelerate, evaluate for multi-label image classification
• detailed comments and explanations on writing training loops
• inference visualization

Based on further reviews, I was thinking to then create a .mdx file which contains both transfer learning and fine-tuning parts.

merveenoyan commented 6 months ago

Hey @shreydan 👋 We currently have transfer learning one and LoRA one already merged. Have you renamed the transfer learning one, it should've been in this PR somewhere. Moreover, can you check if your PR aligns with the merged notebooks? I will check which other PRs you have and merge them and come back to this if it's ok.

shreydan commented 6 months ago

@merveenoyan sorry didn't know the transfer learning file was merged, actually I have renamed it to transfer-learning-image-classification.ipynb since there were object detection and segmentation tasks as well.

merveenoyan commented 6 months ago

@shreydan if you can change the name of the file we can change it in another PR

merveenoyan commented 6 months ago

@shreydan we can first wait for other PRs to be merged and then merge this if it's ok for you

shreydan commented 6 months ago

@shreydan if you can change the name of the file we can change it in another PR @merveenoyan what should I rename the file to?

@shreydan we can first wait for other PRs to be merged and then merge this if it's ok for you

okay no issues in that. Sorry for this confusion.

merveenoyan commented 6 months ago

@Anindyadeep @shreydan I think we can merge this one. If you could sync the changes with this repository and solve merge conflicts I'll give a one last review 🤓

Anindyadeep commented 6 months ago

Sure we can do that, and then let's get it merged :)

merveenoyan commented 5 months ago

@Anindyadeep do you need help with this?

Anindyadeep commented 5 months ago

Hey @merveenoyan, so sorry for the delay, I am bit worked up with some priorities, give me few days I along with @shreydan will solve this very soon :)

Anindyadeep commented 5 months ago

Hi @merveenoyan, sorry for the delay, @shreydan has helped us by updating and removing the merge conflicts. Let us know what else is needed for the merge.

shreydan commented 5 months ago

hello @merveenoyan sorry for the conflicts, our unit had multiple PRs hence the problems. 1 old notebook was removed and has been updated with 2 new notebooks with regards to ViT image classification (transfer learning & fine tuning). Would really appreciate reviews on these 2 notebooks. Thank you for your patience

merveenoyan commented 5 months ago

@MKhalusova can you skim through the notebooks? it seemed ready to me

johko commented 5 months ago

Merging :rocket:

shreydan commented 5 months ago

@johko @MKhalusova @merveenoyan thank you very much!

• © Githubissues.
• Githubissues is a development platform for aggregating issues.

[review-notebook-app[bot]]

View / edit / reply to this conversation on ReviewNB

merveenoyan commented on 2023-12-06T16:32:34Z ----------------------------------------------------------------

suggestions for clarity:

"It might be possible that some images in your dataset will be grayscale or transparent (RGBA)."

"We pass the images through the processor to apply the transforms to process and convert them into PyTorch format."

also you could carry label2id here instead

review-notebook-app[bot] commented 7 months ago

View / edit / reply to this conversation on ReviewNB

merveenoyan commented on 2023-12-06T16:32:36Z ----------------------------------------------------------------

We can set push_to_hub to True and later call push_to_hub on Trainer after the save so people can learn how to use HF Hub as model registry.

MKhalusova commented on 2023-12-08T17:56:01Z ----------------------------------------------------------------

If we set push_to_hub to True, we also need to add a call with notebook_login:

from huggingface_hub import notebook_login
notebook_login()

In this case, the output_dir will also be the name of the repo where your model checkpoint will be pushed.

review-notebook-app[bot] commented 7 months ago

View / edit / reply to this conversation on ReviewNB

merveenoyan commented on 2023-12-06T16:32:37Z ----------------------------------------------------------------

We will now save the model.

review-notebook-app[bot] commented 7 months ago

View / edit / reply to this conversation on ReviewNB

merveenoyan commented on 2023-12-06T16:32:38Z ----------------------------------------------------------------

We can also push this repository to https://huggingface.co/hf-vision organization so people can checkout tensorboard logs and the model card.

shreydan commented 7 months ago

I checked the # trainable params with sum([p.numel() for p in model.parameters() if p.requires_grad])

and all the params were trainable. So had to freeze the weights manually.

---

View entire conversation on ReviewNB

shreydan commented 7 months ago

Quite end-to-end and nicely put 💜 left some comments on reviewNB to structure the notebook and also added some comments about using transformers 🤗

Thank you very much for the review @merveenoyan! Will keep all your detailed and helpful suggestions in mind, this was an initial PR to understand more about how everything should be arranged.

I will update this notebook accordingly and also add the mdx file. Thank you! 🤗

MKhalusova commented 7 months ago

If we set push_to_hub to True, we also need to add a call with notebook_login:

from huggingface_hub import notebook_login
notebook_login()

In this case, the output_dir will also be the name of the repo where your model checkpoint will be pushed.

---

View entire conversation on ReviewNB

johko commented 7 months ago

Nice Demo @shreydan, as Merve already suggested in some of the comments - feel free to make it a bit more conversational. The focus should be on the code, but there is no harm in writing some more text here and there :slightly_smiling_face:

Keep it up :+1:

shreydan commented 6 months ago

@MKhalusova @johko @merveenoyan I've made further updates to my PR as suggested.

Please check the image-classification notebooks:

• notebooks/Unit 3 - Vision Transformers/fine-tuning-multilabel-image-classification.ipynb
• notebooks/Unit 3 - Vision Transformers/transfer-learning-image-classification.ipynb

For transfer-learning:

• updated notebook as per the reviews before
• added pushing model to hub + inference visualization

For fine-tuning:

• new notebook with timm, accelerate, evaluate for multi-label image classification
• detailed comments and explanations on writing training loops
• inference visualization

Based on further reviews, I was thinking to then create a .mdx file which contains both transfer learning and fine-tuning parts.

merveenoyan commented 6 months ago

@Anindyadeep can you tell me which files are contributed through this PR since I remember I reviewed some of the notebooks in other PRs that seems like a change in this PR. It would be easier for us to review.

Anindyadeep commented 6 months ago

@Anindyadeep can you tell me which files are contributed through this PR since I remember I reviewed some of the notebooks in other PRs that seems like a change in this PR. It would be easier for us to review.

Yeah, so all the sub-topics this includes:

1. Fine-tuning for object detection
2. Lora for Image classification
3. Transfer Learning for Segmentation
4. Knowledge distillation
5. Transfer learning for segmentation
6. Fine-tuning multi class classification

All of these topics are included in the PR as a whole. So, some how I am not sure, but when we merged from @shreydan's main to this repo (after our initial peer reviews), it moved all the new files under this PR.

I got to know the reason, and this could be because the initial PR was set from @shreydan main to this main (for image classification only), and now that we were following the convention of first reviewing from our main and then merge for main review, it automatically collects the file and saves here.

shreydan commented 6 months ago

@MKhalusova @merveenoyan @johko

Sorry for this confusion, this PR is for the image classification part under unit-3 vision transformers : fine-tuning and transfer-learning.

We have other PRs for segmentation, object detection, LoRA and Knowledge distillation.

For this PR:

Please check the image-classification notebooks:

- notebooks/Unit 3 - Vision Transformers/fine-tuning-multilabel-image-classification.ipynb
- notebooks/Unit 3 - Vision Transformers/transfer-learning-image-classification.ipynb

For image classification, transfer-learning:

• updated notebook as per the reviews before
• added pushing model to hub + inference visualization

For image classification, fine-tuning:

• new notebook with timm, accelerate, evaluate for multi-label image classification
• detailed comments and explanations on writing training loops
• inference visualization

Based on further reviews, I was thinking to then create a .mdx file which contains both transfer learning and fine-tuning parts.

merveenoyan commented 6 months ago

Hey @shreydan 👋 We currently have transfer learning one and LoRA one already merged. Have you renamed the transfer learning one, it should've been in this PR somewhere. Moreover, can you check if your PR aligns with the merged notebooks? I will check which other PRs you have and merge them and come back to this if it's ok.

shreydan commented 6 months ago

@merveenoyan sorry didn't know the transfer learning file was merged, actually I have renamed it to transfer-learning-image-classification.ipynb since there were object detection and segmentation tasks as well.

merveenoyan commented 6 months ago

@shreydan if you can change the name of the file we can change it in another PR

merveenoyan commented 6 months ago

@shreydan we can first wait for other PRs to be merged and then merge this if it's ok for you

shreydan commented 6 months ago

@shreydan if you can change the name of the file we can change it in another PR @merveenoyan what should I rename the file to?

@shreydan we can first wait for other PRs to be merged and then merge this if it's ok for you

okay no issues in that. Sorry for this confusion.

merveenoyan commented 6 months ago

@Anindyadeep @shreydan I think we can merge this one. If you could sync the changes with this repository and solve merge conflicts I'll give a one last review 🤓

Anindyadeep commented 6 months ago

Sure we can do that, and then let's get it merged :)

merveenoyan commented 5 months ago

@Anindyadeep do you need help with this?

Anindyadeep commented 5 months ago

Hey @merveenoyan, so sorry for the delay, I am bit worked up with some priorities, give me few days I along with @shreydan will solve this very soon :)

Anindyadeep commented 5 months ago

Hi @merveenoyan, sorry for the delay, @shreydan has helped us by updating and removing the merge conflicts. Let us know what else is needed for the merge.

shreydan commented 5 months ago

hello @merveenoyan sorry for the conflicts, our unit had multiple PRs hence the problems. 1 old notebook was removed and has been updated with 2 new notebooks with regards to ViT image classification (transfer learning & fine tuning). Would really appreciate reviews on these 2 notebooks. Thank you for your patience

merveenoyan commented 5 months ago

@MKhalusova can you skim through the notebooks? it seemed ready to me

johko commented 5 months ago

Merging :rocket:

shreydan commented 5 months ago

@johko @MKhalusova @merveenoyan thank you very much!

• © Githubissues.
• Githubissues is a development platform for aggregating issues.

[review-notebook-app[bot]]

View / edit / reply to this conversation on ReviewNB

merveenoyan commented on 2023-12-06T16:32:36Z ----------------------------------------------------------------

We can set push_to_hub to True and later call push_to_hub on Trainer after the save so people can learn how to use HF Hub as model registry.

MKhalusova commented on 2023-12-08T17:56:01Z ----------------------------------------------------------------

If we set push_to_hub to True, we also need to add a call with notebook_login:

from huggingface_hub import notebook_login
notebook_login()

In this case, the output_dir will also be the name of the repo where your model checkpoint will be pushed.

review-notebook-app[bot] commented 7 months ago

View / edit / reply to this conversation on ReviewNB

merveenoyan commented on 2023-12-06T16:32:37Z ----------------------------------------------------------------

We will now save the model.

review-notebook-app[bot] commented 7 months ago

View / edit / reply to this conversation on ReviewNB

merveenoyan commented on 2023-12-06T16:32:38Z ----------------------------------------------------------------

We can also push this repository to https://huggingface.co/hf-vision organization so people can checkout tensorboard logs and the model card.

shreydan commented 7 months ago

I checked the # trainable params with sum([p.numel() for p in model.parameters() if p.requires_grad])

and all the params were trainable. So had to freeze the weights manually.

---

View entire conversation on ReviewNB

shreydan commented 7 months ago

Quite end-to-end and nicely put 💜 left some comments on reviewNB to structure the notebook and also added some comments about using transformers 🤗

Thank you very much for the review @merveenoyan! Will keep all your detailed and helpful suggestions in mind, this was an initial PR to understand more about how everything should be arranged.

I will update this notebook accordingly and also add the mdx file. Thank you! 🤗

MKhalusova commented 7 months ago

If we set push_to_hub to True, we also need to add a call with notebook_login:

from huggingface_hub import notebook_login
notebook_login()

In this case, the output_dir will also be the name of the repo where your model checkpoint will be pushed.

---

View entire conversation on ReviewNB

johko commented 7 months ago

Nice Demo @shreydan, as Merve already suggested in some of the comments - feel free to make it a bit more conversational. The focus should be on the code, but there is no harm in writing some more text here and there :slightly_smiling_face:

Keep it up :+1:

shreydan commented 6 months ago

@MKhalusova @johko @merveenoyan I've made further updates to my PR as suggested.

Please check the image-classification notebooks:

• notebooks/Unit 3 - Vision Transformers/fine-tuning-multilabel-image-classification.ipynb
• notebooks/Unit 3 - Vision Transformers/transfer-learning-image-classification.ipynb

For transfer-learning:

• updated notebook as per the reviews before
• added pushing model to hub + inference visualization

For fine-tuning:

• new notebook with timm, accelerate, evaluate for multi-label image classification
• detailed comments and explanations on writing training loops
• inference visualization

Based on further reviews, I was thinking to then create a .mdx file which contains both transfer learning and fine-tuning parts.

merveenoyan commented 6 months ago

@Anindyadeep can you tell me which files are contributed through this PR since I remember I reviewed some of the notebooks in other PRs that seems like a change in this PR. It would be easier for us to review.

Anindyadeep commented 6 months ago

@Anindyadeep can you tell me which files are contributed through this PR since I remember I reviewed some of the notebooks in other PRs that seems like a change in this PR. It would be easier for us to review.

Yeah, so all the sub-topics this includes:

1. Fine-tuning for object detection
2. Lora for Image classification
3. Transfer Learning for Segmentation
4. Knowledge distillation
5. Transfer learning for segmentation
6. Fine-tuning multi class classification

All of these topics are included in the PR as a whole. So, some how I am not sure, but when we merged from @shreydan's main to this repo (after our initial peer reviews), it moved all the new files under this PR.

I got to know the reason, and this could be because the initial PR was set from @shreydan main to this main (for image classification only), and now that we were following the convention of first reviewing from our main and then merge for main review, it automatically collects the file and saves here.

shreydan commented 6 months ago

@MKhalusova @merveenoyan @johko

Sorry for this confusion, this PR is for the image classification part under unit-3 vision transformers : fine-tuning and transfer-learning.

We have other PRs for segmentation, object detection, LoRA and Knowledge distillation.

For this PR:

Please check the image-classification notebooks:

- notebooks/Unit 3 - Vision Transformers/fine-tuning-multilabel-image-classification.ipynb
- notebooks/Unit 3 - Vision Transformers/transfer-learning-image-classification.ipynb

For image classification, transfer-learning:

• updated notebook as per the reviews before
• added pushing model to hub + inference visualization

For image classification, fine-tuning:

• new notebook with timm, accelerate, evaluate for multi-label image classification
• detailed comments and explanations on writing training loops
• inference visualization

Based on further reviews, I was thinking to then create a .mdx file which contains both transfer learning and fine-tuning parts.

merveenoyan commented 6 months ago

Hey @shreydan 👋 We currently have transfer learning one and LoRA one already merged. Have you renamed the transfer learning one, it should've been in this PR somewhere. Moreover, can you check if your PR aligns with the merged notebooks? I will check which other PRs you have and merge them and come back to this if it's ok.

shreydan commented 6 months ago

@merveenoyan sorry didn't know the transfer learning file was merged, actually I have renamed it to transfer-learning-image-classification.ipynb since there were object detection and segmentation tasks as well.

merveenoyan commented 6 months ago

@shreydan if you can change the name of the file we can change it in another PR

merveenoyan commented 6 months ago

@shreydan we can first wait for other PRs to be merged and then merge this if it's ok for you

shreydan commented 6 months ago

@shreydan if you can change the name of the file we can change it in another PR @merveenoyan what should I rename the file to?

@shreydan we can first wait for other PRs to be merged and then merge this if it's ok for you

okay no issues in that. Sorry for this confusion.

merveenoyan commented 6 months ago

@Anindyadeep @shreydan I think we can merge this one. If you could sync the changes with this repository and solve merge conflicts I'll give a one last review 🤓

Anindyadeep commented 6 months ago

Sure we can do that, and then let's get it merged :)

merveenoyan commented 5 months ago

@Anindyadeep do you need help with this?

Anindyadeep commented 5 months ago

Hey @merveenoyan, so sorry for the delay, I am bit worked up with some priorities, give me few days I along with @shreydan will solve this very soon :)

Anindyadeep commented 5 months ago

Hi @merveenoyan, sorry for the delay, @shreydan has helped us by updating and removing the merge conflicts. Let us know what else is needed for the merge.

shreydan commented 5 months ago

hello @merveenoyan sorry for the conflicts, our unit had multiple PRs hence the problems. 1 old notebook was removed and has been updated with 2 new notebooks with regards to ViT image classification (transfer learning & fine tuning). Would really appreciate reviews on these 2 notebooks. Thank you for your patience

merveenoyan commented 5 months ago

@MKhalusova can you skim through the notebooks? it seemed ready to me

johko commented 5 months ago

Merging :rocket:

shreydan commented 5 months ago

@johko @MKhalusova @merveenoyan thank you very much!

• © Githubissues.
• Githubissues is a development platform for aggregating issues.

[review-notebook-app[bot]]

View / edit / reply to this conversation on ReviewNB

merveenoyan commented on 2023-12-06T16:32:37Z ----------------------------------------------------------------

We will now save the model.

review-notebook-app[bot] commented 7 months ago

View / edit / reply to this conversation on ReviewNB

merveenoyan commented on 2023-12-06T16:32:38Z ----------------------------------------------------------------

We can also push this repository to https://huggingface.co/hf-vision organization so people can checkout tensorboard logs and the model card.

shreydan commented 7 months ago

I checked the # trainable params with sum([p.numel() for p in model.parameters() if p.requires_grad])

and all the params were trainable. So had to freeze the weights manually.

---

View entire conversation on ReviewNB

shreydan commented 7 months ago

Quite end-to-end and nicely put 💜 left some comments on reviewNB to structure the notebook and also added some comments about using transformers 🤗

Thank you very much for the review @merveenoyan! Will keep all your detailed and helpful suggestions in mind, this was an initial PR to understand more about how everything should be arranged.

I will update this notebook accordingly and also add the mdx file. Thank you! 🤗

MKhalusova commented 7 months ago

If we set push_to_hub to True, we also need to add a call with notebook_login:

from huggingface_hub import notebook_login
notebook_login()

In this case, the output_dir will also be the name of the repo where your model checkpoint will be pushed.

---

View entire conversation on ReviewNB

johko commented 7 months ago

Nice Demo @shreydan, as Merve already suggested in some of the comments - feel free to make it a bit more conversational. The focus should be on the code, but there is no harm in writing some more text here and there :slightly_smiling_face:

Keep it up :+1:

shreydan commented 6 months ago

@MKhalusova @johko @merveenoyan I've made further updates to my PR as suggested.

Please check the image-classification notebooks:

• notebooks/Unit 3 - Vision Transformers/fine-tuning-multilabel-image-classification.ipynb
• notebooks/Unit 3 - Vision Transformers/transfer-learning-image-classification.ipynb

For transfer-learning:

• updated notebook as per the reviews before
• added pushing model to hub + inference visualization

For fine-tuning:

• new notebook with timm, accelerate, evaluate for multi-label image classification
• detailed comments and explanations on writing training loops
• inference visualization

Based on further reviews, I was thinking to then create a .mdx file which contains both transfer learning and fine-tuning parts.

merveenoyan commented 6 months ago

@Anindyadeep can you tell me which files are contributed through this PR since I remember I reviewed some of the notebooks in other PRs that seems like a change in this PR. It would be easier for us to review.

Anindyadeep commented 6 months ago

@Anindyadeep can you tell me which files are contributed through this PR since I remember I reviewed some of the notebooks in other PRs that seems like a change in this PR. It would be easier for us to review.

Yeah, so all the sub-topics this includes:

1. Fine-tuning for object detection
2. Lora for Image classification
3. Transfer Learning for Segmentation
4. Knowledge distillation
5. Transfer learning for segmentation
6. Fine-tuning multi class classification

All of these topics are included in the PR as a whole. So, some how I am not sure, but when we merged from @shreydan's main to this repo (after our initial peer reviews), it moved all the new files under this PR.

I got to know the reason, and this could be because the initial PR was set from @shreydan main to this main (for image classification only), and now that we were following the convention of first reviewing from our main and then merge for main review, it automatically collects the file and saves here.

shreydan commented 6 months ago

@MKhalusova @merveenoyan @johko

Sorry for this confusion, this PR is for the image classification part under unit-3 vision transformers : fine-tuning and transfer-learning.

We have other PRs for segmentation, object detection, LoRA and Knowledge distillation.

For this PR:

Please check the image-classification notebooks:

- notebooks/Unit 3 - Vision Transformers/fine-tuning-multilabel-image-classification.ipynb
- notebooks/Unit 3 - Vision Transformers/transfer-learning-image-classification.ipynb

For image classification, transfer-learning:

• updated notebook as per the reviews before
• added pushing model to hub + inference visualization

For image classification, fine-tuning:

• new notebook with timm, accelerate, evaluate for multi-label image classification
• detailed comments and explanations on writing training loops
• inference visualization

Based on further reviews, I was thinking to then create a .mdx file which contains both transfer learning and fine-tuning parts.

merveenoyan commented 6 months ago

Hey @shreydan 👋 We currently have transfer learning one and LoRA one already merged. Have you renamed the transfer learning one, it should've been in this PR somewhere. Moreover, can you check if your PR aligns with the merged notebooks? I will check which other PRs you have and merge them and come back to this if it's ok.

shreydan commented 6 months ago

@merveenoyan sorry didn't know the transfer learning file was merged, actually I have renamed it to transfer-learning-image-classification.ipynb since there were object detection and segmentation tasks as well.

merveenoyan commented 6 months ago

@shreydan if you can change the name of the file we can change it in another PR

merveenoyan commented 6 months ago

@shreydan we can first wait for other PRs to be merged and then merge this if it's ok for you

shreydan commented 6 months ago

@shreydan if you can change the name of the file we can change it in another PR @merveenoyan what should I rename the file to?

@shreydan we can first wait for other PRs to be merged and then merge this if it's ok for you

okay no issues in that. Sorry for this confusion.

merveenoyan commented 6 months ago

@Anindyadeep @shreydan I think we can merge this one. If you could sync the changes with this repository and solve merge conflicts I'll give a one last review 🤓

Anindyadeep commented 6 months ago

Sure we can do that, and then let's get it merged :)

merveenoyan commented 5 months ago

@Anindyadeep do you need help with this?

Anindyadeep commented 5 months ago

Hey @merveenoyan, so sorry for the delay, I am bit worked up with some priorities, give me few days I along with @shreydan will solve this very soon :)

Anindyadeep commented 5 months ago

Hi @merveenoyan, sorry for the delay, @shreydan has helped us by updating and removing the merge conflicts. Let us know what else is needed for the merge.

shreydan commented 5 months ago

hello @merveenoyan sorry for the conflicts, our unit had multiple PRs hence the problems. 1 old notebook was removed and has been updated with 2 new notebooks with regards to ViT image classification (transfer learning & fine tuning). Would really appreciate reviews on these 2 notebooks. Thank you for your patience

merveenoyan commented 5 months ago

@MKhalusova can you skim through the notebooks? it seemed ready to me

johko commented 5 months ago

Merging :rocket:

shreydan commented 5 months ago

@johko @MKhalusova @merveenoyan thank you very much!

• © Githubissues.
• Githubissues is a development platform for aggregating issues.

[review-notebook-app[bot]]

View / edit / reply to this conversation on ReviewNB

merveenoyan commented on 2023-12-06T16:32:38Z ----------------------------------------------------------------

We can also push this repository to https://huggingface.co/hf-vision organization so people can checkout tensorboard logs and the model card.

shreydan commented 7 months ago

I checked the # trainable params with sum([p.numel() for p in model.parameters() if p.requires_grad])

and all the params were trainable. So had to freeze the weights manually.

---

View entire conversation on ReviewNB

shreydan commented 7 months ago

Quite end-to-end and nicely put 💜 left some comments on reviewNB to structure the notebook and also added some comments about using transformers 🤗

Thank you very much for the review @merveenoyan! Will keep all your detailed and helpful suggestions in mind, this was an initial PR to understand more about how everything should be arranged.

I will update this notebook accordingly and also add the mdx file. Thank you! 🤗

MKhalusova commented 7 months ago

If we set push_to_hub to True, we also need to add a call with notebook_login:

from huggingface_hub import notebook_login
notebook_login()

In this case, the output_dir will also be the name of the repo where your model checkpoint will be pushed.

---

View entire conversation on ReviewNB

johko commented 7 months ago

Nice Demo @shreydan, as Merve already suggested in some of the comments - feel free to make it a bit more conversational. The focus should be on the code, but there is no harm in writing some more text here and there :slightly_smiling_face:

Keep it up :+1:

shreydan commented 6 months ago

@MKhalusova @johko @merveenoyan I've made further updates to my PR as suggested.

Please check the image-classification notebooks:

• notebooks/Unit 3 - Vision Transformers/fine-tuning-multilabel-image-classification.ipynb
• notebooks/Unit 3 - Vision Transformers/transfer-learning-image-classification.ipynb

For transfer-learning:

• updated notebook as per the reviews before
• added pushing model to hub + inference visualization

For fine-tuning:

• new notebook with timm, accelerate, evaluate for multi-label image classification
• detailed comments and explanations on writing training loops
• inference visualization

Based on further reviews, I was thinking to then create a .mdx file which contains both transfer learning and fine-tuning parts.

merveenoyan commented 6 months ago

@Anindyadeep can you tell me which files are contributed through this PR since I remember I reviewed some of the notebooks in other PRs that seems like a change in this PR. It would be easier for us to review.

Anindyadeep commented 6 months ago

@Anindyadeep can you tell me which files are contributed through this PR since I remember I reviewed some of the notebooks in other PRs that seems like a change in this PR. It would be easier for us to review.

Yeah, so all the sub-topics this includes:

1. Fine-tuning for object detection
2. Lora for Image classification
3. Transfer Learning for Segmentation
4. Knowledge distillation
5. Transfer learning for segmentation
6. Fine-tuning multi class classification

All of these topics are included in the PR as a whole. So, some how I am not sure, but when we merged from @shreydan's main to this repo (after our initial peer reviews), it moved all the new files under this PR.

I got to know the reason, and this could be because the initial PR was set from @shreydan main to this main (for image classification only), and now that we were following the convention of first reviewing from our main and then merge for main review, it automatically collects the file and saves here.

shreydan commented 6 months ago

@MKhalusova @merveenoyan @johko

Sorry for this confusion, this PR is for the image classification part under unit-3 vision transformers : fine-tuning and transfer-learning.

We have other PRs for segmentation, object detection, LoRA and Knowledge distillation.

For this PR:

Please check the image-classification notebooks:

- notebooks/Unit 3 - Vision Transformers/fine-tuning-multilabel-image-classification.ipynb
- notebooks/Unit 3 - Vision Transformers/transfer-learning-image-classification.ipynb

For image classification, transfer-learning:

• updated notebook as per the reviews before
• added pushing model to hub + inference visualization

For image classification, fine-tuning:

• new notebook with timm, accelerate, evaluate for multi-label image classification
• detailed comments and explanations on writing training loops
• inference visualization

Based on further reviews, I was thinking to then create a .mdx file which contains both transfer learning and fine-tuning parts.

merveenoyan commented 6 months ago

Hey @shreydan 👋 We currently have transfer learning one and LoRA one already merged. Have you renamed the transfer learning one, it should've been in this PR somewhere. Moreover, can you check if your PR aligns with the merged notebooks? I will check which other PRs you have and merge them and come back to this if it's ok.

shreydan commented 6 months ago

@merveenoyan sorry didn't know the transfer learning file was merged, actually I have renamed it to transfer-learning-image-classification.ipynb since there were object detection and segmentation tasks as well.

merveenoyan commented 6 months ago

@shreydan if you can change the name of the file we can change it in another PR

merveenoyan commented 6 months ago

@shreydan we can first wait for other PRs to be merged and then merge this if it's ok for you

shreydan commented 6 months ago

@shreydan if you can change the name of the file we can change it in another PR @merveenoyan what should I rename the file to?

@shreydan we can first wait for other PRs to be merged and then merge this if it's ok for you

okay no issues in that. Sorry for this confusion.

merveenoyan commented 6 months ago

@Anindyadeep @shreydan I think we can merge this one. If you could sync the changes with this repository and solve merge conflicts I'll give a one last review 🤓

Anindyadeep commented 6 months ago

Sure we can do that, and then let's get it merged :)

merveenoyan commented 5 months ago

@Anindyadeep do you need help with this?

Anindyadeep commented 5 months ago

Hey @merveenoyan, so sorry for the delay, I am bit worked up with some priorities, give me few days I along with @shreydan will solve this very soon :)

Anindyadeep commented 5 months ago

Hi @merveenoyan, sorry for the delay, @shreydan has helped us by updating and removing the merge conflicts. Let us know what else is needed for the merge.

shreydan commented 5 months ago

hello @merveenoyan sorry for the conflicts, our unit had multiple PRs hence the problems. 1 old notebook was removed and has been updated with 2 new notebooks with regards to ViT image classification (transfer learning & fine tuning). Would really appreciate reviews on these 2 notebooks. Thank you for your patience

merveenoyan commented 5 months ago

@MKhalusova can you skim through the notebooks? it seemed ready to me

johko commented 5 months ago

Merging :rocket:

shreydan commented 5 months ago

@johko @MKhalusova @merveenoyan thank you very much!

• © Githubissues.
• Githubissues is a development platform for aggregating issues.

[shreydan]

I checked the # trainable params with sum([p.numel() for p in model.parameters() if p.requires_grad])

and all the params were trainable. So had to freeze the weights manually.

---

View entire conversation on ReviewNB

[shreydan]

Quite end-to-end and nicely put 💜 left some comments on reviewNB to structure the notebook and also added some comments about using transformers 🤗

Thank you very much for the review @merveenoyan! Will keep all your detailed and helpful suggestions in mind, this was an initial PR to understand more about how everything should be arranged.

I will update this notebook accordingly and also add the mdx file. Thank you! 🤗

[MKhalusova]

If we set push_to_hub to True, we also need to add a call with notebook_login:

from huggingface_hub import notebook_login
notebook_login()

In this case, the output_dir will also be the name of the repo where your model checkpoint will be pushed.

---

View entire conversation on ReviewNB

[johko]

Nice Demo @shreydan, as Merve already suggested in some of the comments - feel free to make it a bit more conversational. The focus should be on the code, but there is no harm in writing some more text here and there :slightly_smiling_face:

Keep it up :+1:

[shreydan]

@MKhalusova @johko @merveenoyan I've made further updates to my PR as suggested.

Please check the image-classification notebooks:

• notebooks/Unit 3 - Vision Transformers/fine-tuning-multilabel-image-classification.ipynb
• notebooks/Unit 3 - Vision Transformers/transfer-learning-image-classification.ipynb

For transfer-learning:

• updated notebook as per the reviews before
• added pushing model to hub + inference visualization

For fine-tuning:

• new notebook with timm, accelerate, evaluate for multi-label image classification
• detailed comments and explanations on writing training loops
• inference visualization

Based on further reviews, I was thinking to then create a .mdx file which contains both transfer learning and fine-tuning parts.

[merveenoyan]

@Anindyadeep can you tell me which files are contributed through this PR since I remember I reviewed some of the notebooks in other PRs that seems like a change in this PR. It would be easier for us to review.

[Anindyadeep]

@Anindyadeep can you tell me which files are contributed through this PR since I remember I reviewed some of the notebooks in other PRs that seems like a change in this PR. It would be easier for us to review.

Yeah, so all the sub-topics this includes:

1. Fine-tuning for object detection
2. Lora for Image classification
3. Transfer Learning for Segmentation
4. Knowledge distillation
5. Transfer learning for segmentation
6. Fine-tuning multi class classification

All of these topics are included in the PR as a whole. So, some how I am not sure, but when we merged from @shreydan's main to this repo (after our initial peer reviews), it moved all the new files under this PR.

I got to know the reason, and this could be because the initial PR was set from @shreydan main to this main (for image classification only), and now that we were following the convention of first reviewing from our main and then merge for main review, it automatically collects the file and saves here.

[shreydan]

@MKhalusova @merveenoyan @johko

Sorry for this confusion, this PR is for the image classification part under unit-3 vision transformers : fine-tuning and transfer-learning.

We have other PRs for segmentation, object detection, LoRA and Knowledge distillation.

For this PR:

Please check the image-classification notebooks:

- notebooks/Unit 3 - Vision Transformers/fine-tuning-multilabel-image-classification.ipynb
- notebooks/Unit 3 - Vision Transformers/transfer-learning-image-classification.ipynb

For image classification, transfer-learning:

• updated notebook as per the reviews before
• added pushing model to hub + inference visualization

For image classification, fine-tuning:

• new notebook with timm, accelerate, evaluate for multi-label image classification
• detailed comments and explanations on writing training loops
• inference visualization

Based on further reviews, I was thinking to then create a .mdx file which contains both transfer learning and fine-tuning parts.

[merveenoyan]

Hey @shreydan 👋 We currently have transfer learning one and LoRA one already merged. Have you renamed the transfer learning one, it should've been in this PR somewhere. Moreover, can you check if your PR aligns with the merged notebooks? I will check which other PRs you have and merge them and come back to this if it's ok.

[shreydan]

@merveenoyan sorry didn't know the transfer learning file was merged, actually I have renamed it to transfer-learning-image-classification.ipynb since there were object detection and segmentation tasks as well.

[merveenoyan]

@shreydan if you can change the name of the file we can change it in another PR

[merveenoyan]

@shreydan we can first wait for other PRs to be merged and then merge this if it's ok for you

[shreydan]

@shreydan if you can change the name of the file we can change it in another PR @merveenoyan what should I rename the file to?

@shreydan we can first wait for other PRs to be merged and then merge this if it's ok for you

okay no issues in that. Sorry for this confusion.

[merveenoyan]

@Anindyadeep @shreydan I think we can merge this one. If you could sync the changes with this repository and solve merge conflicts I'll give a one last review 🤓

[Anindyadeep]

Sure we can do that, and then let's get it merged :)

[merveenoyan]

@Anindyadeep do you need help with this?

[Anindyadeep]

Hey @merveenoyan, so sorry for the delay, I am bit worked up with some priorities, give me few days I along with @shreydan will solve this very soon :)

[Anindyadeep]

Hi @merveenoyan, sorry for the delay, @shreydan has helped us by updating and removing the merge conflicts. Let us know what else is needed for the merge.

[shreydan]

hello @merveenoyan sorry for the conflicts, our unit had multiple PRs hence the problems. 1 old notebook was removed and has been updated with 2 new notebooks with regards to ViT image classification (transfer learning & fine tuning). Would really appreciate reviews on these 2 notebooks. Thank you for your patience

[merveenoyan]

@MKhalusova can you skim through the notebooks? it seemed ready to me

[johko]

Merging :rocket:

[shreydan]

@johko @MKhalusova @merveenoyan thank you very much!