Adapt custom AI models to the trainer API and to π€
β’
15
LAin
not-lain
AI & ML interests
custom architectures with huggingface
Articles
Organizations
not-lain's activity
posted
an
update
2 days ago
Post
646
If you're a researcher or developing your own model π you might need to take a look at huggingface's ModelHubMixin classes.
They are used to seamlessly integrate your AI model with huggingface and to save/ load your model easily π
1οΈβ£ make sure you're using the appropriate library version
2οΈβ£ inherit from the appropriate class
4οΈβ£ push the model to the hub (or use save_pretrained method to save locally)
5οΈβ£ Load and initialize the model from the hub using the original class
They are used to seamlessly integrate your AI model with huggingface and to save/ load your model easily π
1οΈβ£ make sure you're using the appropriate library version
pip install -qU "huggingface_hub>=0.22"2οΈβ£ inherit from the appropriate class
from huggingface_hub import PyTorchModelHubMixin
from torch import nn
class MyModel(nn.Module,PyTorchModelHubMixin):
def __init__(self, a, b):
super().__init__()
self.layer = nn.Linear(a,b)
def forward(self,inputs):
return self.layer(inputs)
first_model = MyModel(3,1)4οΈβ£ push the model to the hub (or use save_pretrained method to save locally)
first_model.push_to_hub("not-lain/test")5οΈβ£ Load and initialize the model from the hub using the original class
pretrained_model = MyModel.from_pretrained("not-lain/test")
posted
an
update
5 days ago
Post
782
I'm looking for open-source image embedding models for RAG applications and/or multimodel embedding models if they exist in the first place.
if you have any extra resources about using, creating, or finetuning them feel free to share them below π€
if you have any extra resources about using, creating, or finetuning them feel free to share them below π€
reel
replied to
their
post
9 days ago
thank you sooooo much β€οΈβ€οΈ
posted
an
update
9 days ago
Post
1165
π₯³celebrating 5K readers in one of my blog posts π₯³
I came back with another one this time π€
in this blog you will learn π :
* How to train custom AI models with the trainer API π
* integrate your AI models with HF using the mixin classes π₯
happy reading everyone π€
πlink: https://huggingface.co/blog/not-lain/trainer-api-and-mixin-classes
I came back with another one this time π€
in this blog you will learn π :
* How to train custom AI models with the trainer API π
* integrate your AI models with HF using the mixin classes π₯
happy reading everyone π€
πlink: https://huggingface.co/blog/not-lain/trainer-api-and-mixin-classes
-
- 2 replies
I see you're using the image slider component too, nice work β€οΈβ€οΈ
Nice work @abhishek β¨
posted
an
update
19 days ago
Post
1731
π just reached 3K+ readers on this blog post about RAG using only HFπ€ related tools in just a little over 1 week from publishing.
πthe most interesting thing about it is that you can use the FAISS index in the datasets library to retrieve your most similar documents.
πhttps://huggingface.co/blog/not-lain/rag-chatbot-using-llama3
Happy reading everyone β¨
πthe most interesting thing about it is that you can use the FAISS index in the datasets library to retrieve your most similar documents.
πhttps://huggingface.co/blog/not-lain/rag-chatbot-using-llama3
Happy reading everyone β¨
replied to
VictorSanh's
post
19 days ago
Did anyone say HF?
literally on π₯
Nice work π₯³π₯³
replied to
HeshamHaroon's
post
27 days ago
nope it's not a bug, the posts feature is now open for everyone
posted
an
update
28 days ago
Post
2244
π₯ **transformers** 4.40.0 is out π₯
βοΈ you can now push your custom pipelines easily to π€, allowing people to easily use your model in a more friendly, unified way.
π€ I already updated my blog to match the new feature https://huggingface.co/blog/not-lain/custom-architectures-with-huggingface.
πA list of some repos that have custom pipelines :
* briaai/RMBG-1.4
* p1atdev/siglip-tagger-test-3
* sgugger/test-dynamic-pipeline
βοΈ you can now push your custom pipelines easily to π€, allowing people to easily use your model in a more friendly, unified way.
π€ I already updated my blog to match the new feature https://huggingface.co/blog/not-lain/custom-architectures-with-huggingface.
πA list of some repos that have custom pipelines :
* briaai/RMBG-1.4
* p1atdev/siglip-tagger-test-3
* sgugger/test-dynamic-pipeline
x3 download speed in HfFileSystem = π₯π
I also really loved using the PyTorchModelHubMixin class π€π€
try adding labels parameter to the forward function in my version of of the RMBG-1.4 model in here https://huggingface.co/not-lain/CustomCodeForRMBG/blob/main/briarmbg.py#L392, this should allow the model to be finetuned easily using the trainer API.
find out more about the labels parameter in this documentation :
https://huggingface.co/docs/transformers/custom_models#writing-a-custom-model
it should look somethind like this :
any open contributions are welcome.
Made a little contribution and it is now easier than ever to use the model via the transformers library
all you have to do is use the following code to access the model :
checkout more details in this pull request : https://huggingface.co/briaai/RMBG-1.4/discussions/9
or like make it a badge in their profiles if they exceed 100 contributions, could be a nice motive for people to make them engage more in the hub
maybe π§ββοΈπ§ββοΈ for open sorcerers out there
leaving this for future readers, huggingface has a really nice documentation about RAG which i highly recommend reading if you're interested about using them. It incorporates clear examples and neatly formatted code :
https://huggingface.co/docs/transformers/model_doc/rag
we're actually an anime/meme group and i made this community for AI + anime related stuff \(οΏ£οΈΆοΏ£*\))
@Tonic
we just reached 300000 members yerterday on FB π₯³π₯³