Trick or ResNet Treat
β’
3
Ross Wightman
rwightman
AI & ML interests
Computer vision, transfer learning, semi/self supervised learning, robotics.
Recent Activity
New activity
5 days ago
timm/ViT-B-16-SigLIP-i18n-256:Are the languages that are supported documented anywhere?
updated
a model
7 days ago
timm/mobilenetv4_conv_medium.e180_ad_r384_in12k
Reacted to
jeffboudier's
post
with π€
10 days ago
New - add your bluesky account to your HF profile:
https://huggingface.co/settings/profile
Is the grass greener, the sky bluer? Will try and figure it out at https://bsky.app/profile/jeffboudier.bsky.social
By the way, HF people starter pack https://bsky.app/starter-pack/huggingface.bsky.social/3laz5x7naiz22
Articles
Organizations
rwightman's activity
reacted to
jeffboudier's
post with π€
10 days ago
Post
926
New - add your bluesky account to your HF profile:
https://huggingface.co/settings/profile
Is the grass greener, the sky bluer? Will try and figure it out at https://bsky.app/profile/jeffboudier.bsky.social
By the way, HF people starter pack https://bsky.app/starter-pack/huggingface.bsky.social/3laz5x7naiz22
https://huggingface.co/settings/profile
Is the grass greener, the sky bluer? Will try and figure it out at https://bsky.app/profile/jeffboudier.bsky.social
By the way, HF people starter pack https://bsky.app/starter-pack/huggingface.bsky.social/3laz5x7naiz22
reacted to
merve's
post with π₯
10 days ago
Post
2513
What a week! A recap for everything you missed βοΈ
merve/nov-22-releases-673fbbcfc1c97c4f411def07
Multimodal β¨
> Mistral AI
released Pixtral 124B, a gigantic open vision language model
> Llava-CoT (formerly known as Llava-o1) was released, a multimodal reproduction of o1 model by PKU
> OpenGVLab released MMPR: a new multimodal reasoning dataset
> Jina has released Jina-CLIP-v2 0.98B multilingual multimodal embeddings
> Apple released new SotA vision encoders AIMv2
LLMs π¦
> AllenAI dropped a huge release of models, datasets and scripts for TΓΌlu, a family of models based on Llama 3.1 aligned with SFT, DPO and a new technique they have developed called RLVR
> Jina has released embeddings-v3: new multilingual embeddings with longer context
> Hugging Face released SmolTalk: synthetic dataset used to align SmolLM2 using supervised fine-tuning
> Microsoft released orca-agentinstruct-1M-v1: a gigantic instruction dataset of 1M synthetic instruction pairs
Image Generation πΌοΈ
> Black Forest Labs released Flux 1. tools: four new models for different image modifications and two LoRAs to do image conditioning and better steer generations
Lastly Hugging Face released a new library Observers: a lightweight SDK for monitoring interactions with AI APIs and easily store and browse them π
$ pip install observers
merve/nov-22-releases-673fbbcfc1c97c4f411def07
Multimodal β¨
> Mistral AI
released Pixtral 124B, a gigantic open vision language model
> Llava-CoT (formerly known as Llava-o1) was released, a multimodal reproduction of o1 model by PKU
> OpenGVLab released MMPR: a new multimodal reasoning dataset
> Jina has released Jina-CLIP-v2 0.98B multilingual multimodal embeddings
> Apple released new SotA vision encoders AIMv2
LLMs π¦
> AllenAI dropped a huge release of models, datasets and scripts for TΓΌlu, a family of models based on Llama 3.1 aligned with SFT, DPO and a new technique they have developed called RLVR
> Jina has released embeddings-v3: new multilingual embeddings with longer context
> Hugging Face released SmolTalk: synthetic dataset used to align SmolLM2 using supervised fine-tuning
> Microsoft released orca-agentinstruct-1M-v1: a gigantic instruction dataset of 1M synthetic instruction pairs
Image Generation πΌοΈ
> Black Forest Labs released Flux 1. tools: four new models for different image modifications and two LoRAs to do image conditioning and better steer generations
Lastly Hugging Face released a new library Observers: a lightweight SDK for monitoring interactions with AI APIs and easily store and browse them π
$ pip install observers
- 3 replies
reacted to
BrigitteTousi's
post with π
10 days ago
Post
856
I'm biased but I think HF Posts is the #1 social platform for the AI community! π€ That being said, most of us are already on X and now also joining Bluesky.
Looking for us on Bsky? We started a team list here: https://bsky.app/starter-pack/did:plc:yyfrnpcutxghwc6eac4xplwp/3lbem54cnxp26
Looking for us on Bsky? We started a team list here: https://bsky.app/starter-pack/did:plc:yyfrnpcutxghwc6eac4xplwp/3lbem54cnxp26
posted
an
update
10 days ago
Post
1148
I'm currently on a push to expand the scope of image based datasets on the Hub. There's certainly a lot already, but for anyone who's looked closely, there's not a whole lot of standardization. I am to fix that, datasets under the https://huggingface.co/timm and https://huggingface.co/pixparse orgs will serve as canonical examples for various task / modality combinations and be useable without fuss in libraries like
I just uploaded the first multi-label dataset that I'll support with
Next up object detection & segmentation! I've got an annotation spec sorted out, a lot of datasets ready to rip, and yeah that means
timm, OpenCLIP, and hopefully more.I just uploaded the first multi-label dataset that I'll support with
timm scripts soon:
timm/plant-pathology-2021 Next up object detection & segmentation! I've got an annotation spec sorted out, a lot of datasets ready to rip, and yeah that means
timm support for object detection, eventually segmentation, is finally under development :O
posted
an
update
12 days ago
Post
1046
Want to validate some hparams or figure out what
I had this sitting on my drive and forgot where I pulled it together from. It's 100 classes of imagenet, 50k train and 10k val images (from ImageNet-1k train set), and 5k test images (from ImageNet-1k val set). 7.4GB instead of > 100GB for the full ImageNet-1k. This ver is not reduced resolution like some other 'mini' versions. Super easy to use with timm train/val scripts, checkout the dataset card.
I often check fine-tuning with even smaller datasets like:
* timm/resisc45
* timm/oxford-iiit-pet
But those are a bit small to train any modest size model w/o starting from pretrained weights.
timm model to use before commiting to download or training with a large dataset? Try mini-imagenet:
timm/mini-imagenet I had this sitting on my drive and forgot where I pulled it together from. It's 100 classes of imagenet, 50k train and 10k val images (from ImageNet-1k train set), and 5k test images (from ImageNet-1k val set). 7.4GB instead of > 100GB for the full ImageNet-1k. This ver is not reduced resolution like some other 'mini' versions. Super easy to use with timm train/val scripts, checkout the dataset card.
I often check fine-tuning with even smaller datasets like:
* timm/resisc45
* timm/oxford-iiit-pet
But those are a bit small to train any modest size model w/o starting from pretrained weights.
reacted to
dvilasuero's
post with π
13 days ago
Post
1048
@Jesse-marqo
and the Marqo team are killing it on the Hub: top embedding models and datasets!
Here's how to start using their new evaluation dataset for curation and labelling:
1. Deploy Argilla on Spaces: https://huggingface.co/new-space?template=argilla%2Fargilla-template-space
2. Load Marqo/amazon-products-eval with the UI wizard.
3. Start curating!
Here's how to start using their new evaluation dataset for curation and labelling:
1. Deploy Argilla on Spaces: https://huggingface.co/new-space?template=argilla%2Fargilla-template-space
2. Load Marqo/amazon-products-eval with the UI wizard.
3. Start curating!
reacted to
sayakpaul's
post with π
14 days ago
Post
2446
It's been a while we shipped native quantization support in
We currently support
This post is just a reminder of what's possible:
1. Loading a model with a quantization config
2. Saving a model with quantization config
3. Loading a pre-quantized model
4.
5. Training and loading LoRAs into quantized checkpoints
Docs:
https://huggingface.co/docs/diffusers/main/en/quantization/bitsandbytes
diffusers π§¨We currently support
bistandbytes as the official backend but using others like torchao is already very simple. This post is just a reminder of what's possible:
1. Loading a model with a quantization config
2. Saving a model with quantization config
3. Loading a pre-quantized model
4.
enable_model_cpu_offload()5. Training and loading LoRAs into quantized checkpoints
Docs:
https://huggingface.co/docs/diffusers/main/en/quantization/bitsandbytes
- 1 reply
posted
an
update
14 days ago
Post
1570
New MobileNetV4 weights were uploaded a few days ago -- more ImageNet-12k training at 384x384 for the speedy 'Conv Medium' models.
There are 3 weight variants here for those who like to tinker. On my hold-out eval they are ordered as below, not that different, but the Adopt 180 epochs closer to AdamW 250 than to AdamW 180.
* AdamW for 250 epochs - timm/mobilenetv4_conv_medium.e250_r384_in12k
* Adopt for 180 epochs - timm/mobilenetv4_conv_medium.e180_ad_r384_in12k
* AdamW for 180 epochs - timm/mobilenetv4_conv_medium.e180_r384_in12k
This was by request as a user reported impressive results using the 'Conv Large' ImagNet-12k pretrains as object detection backbones. ImageNet-1k fine-tunes are pending, the weights do behave differently with the 180 vs 250 epochs and the Adopt vs AdamW optimizer.
There are 3 weight variants here for those who like to tinker. On my hold-out eval they are ordered as below, not that different, but the Adopt 180 epochs closer to AdamW 250 than to AdamW 180.
* AdamW for 250 epochs - timm/mobilenetv4_conv_medium.e250_r384_in12k
* Adopt for 180 epochs - timm/mobilenetv4_conv_medium.e180_ad_r384_in12k
* AdamW for 180 epochs - timm/mobilenetv4_conv_medium.e180_r384_in12k
This was by request as a user reported impressive results using the 'Conv Large' ImagNet-12k pretrains as object detection backbones. ImageNet-1k fine-tunes are pending, the weights do behave differently with the 180 vs 250 epochs and the Adopt vs AdamW optimizer.
reacted to
merve's
post with π
about 1 month ago
Post
3939
I ported the hottest new shape-optimized SigLIP π₯ https://huggingface.co/merve/siglip-so400m-patch16-256-i18n
if you don't want to wait for the next transformers release install transformers from my PR https://github.com/huggingface/transformers/pull/32938 and initialize SigLIP from there
if you don't want to wait for the next transformers release install transformers from my PR https://github.com/huggingface/transformers/pull/32938 and initialize SigLIP from there
posted
an
update
about 2 months ago
Post
639
A new
Featured in the release are:
* The MambaOut model, a cheeky arch inspired by SSM but without the SSM part, a ConvNeXt with gating.
* Several timm trained MambaOut variations with arch tweaks and ImageNet-12k pretrain to verify scaling, supplement ported weights.
* The smallest MobileNetV4, a 0.5x width scaled Conv-Small.
* Two impressive MobileNetV3 Large models outperforming all previous, using MNV4 Small recipe.
* 'Zepto,' a new compact ConvNeXt variant even smaller than the previous Atto, 2.2M params, RMSNorm, and solid results for its size.
* Newly ported SigLIP SO400M/16 ViT multi-lingual weights, the largest i18n weights, prevous was B/16.
* Two ImageNet-1k fine-tuned SigLIP SO400M models at 378x378
* InternViT 300M weight port. A really solid ViT encoder distilled from OpenGVLab 6B VL model encoder.
* An assortment of very small, sub 1M param pretrained test models to improve library unit tests and serve low-resource applications.
timm release (1.0.11) is out now. A also wrote an article on one of the included models: https://huggingface.co/blog/rwightman/mambaoutFeatured in the release are:
* The MambaOut model, a cheeky arch inspired by SSM but without the SSM part, a ConvNeXt with gating.
* Several timm trained MambaOut variations with arch tweaks and ImageNet-12k pretrain to verify scaling, supplement ported weights.
* The smallest MobileNetV4, a 0.5x width scaled Conv-Small.
* Two impressive MobileNetV3 Large models outperforming all previous, using MNV4 Small recipe.
* 'Zepto,' a new compact ConvNeXt variant even smaller than the previous Atto, 2.2M params, RMSNorm, and solid results for its size.
* Newly ported SigLIP SO400M/16 ViT multi-lingual weights, the largest i18n weights, prevous was B/16.
* Two ImageNet-1k fine-tuned SigLIP SO400M models at 378x378
* InternViT 300M weight port. A really solid ViT encoder distilled from OpenGVLab 6B VL model encoder.
* An assortment of very small, sub 1M param pretrained test models to improve library unit tests and serve low-resource applications.
reacted to
merve's
post with π₯
about 2 months ago
Post
3749
Meta AI vision has been cooking
@facebook
They shipped multiple models and demos for their papers at @ECCV π€
Here's a compilation of my top picks:
- Sapiens is family of foundation models for human-centric depth estimation, segmentation and more, all models have open weights and demos π
All models have their demos and even torchscript checkpoints!
A collection of models and demos: facebook/sapiens-66d22047daa6402d565cb2fc
- VFusion3D is state-of-the-art consistent 3D generation model from images
Model: facebook/vfusion3d
Demo: facebook/VFusion3D
- CoTracker is the state-of-the-art point (pixel) tracking model
Demo: facebook/cotracker
Model: facebook/cotracker
They shipped multiple models and demos for their papers at @ECCV π€
Here's a compilation of my top picks:
- Sapiens is family of foundation models for human-centric depth estimation, segmentation and more, all models have open weights and demos π
All models have their demos and even torchscript checkpoints!
A collection of models and demos: facebook/sapiens-66d22047daa6402d565cb2fc
- VFusion3D is state-of-the-art consistent 3D generation model from images
Model: facebook/vfusion3d
Demo: facebook/VFusion3D
- CoTracker is the state-of-the-art point (pixel) tracking model
Demo: facebook/cotracker
Model: facebook/cotracker
reacted to
Wauplin's
post with π₯
2 months ago
Post
2738
What a great milestone to celebrate! The huggingface_hub library is slowly becoming a cornerstone of the Python ML ecosystem when it comes to interacting with the
@huggingface
Hub. It wouldn't be there without the hundreds of community contributions and feedback! No matter if you are loading a model, sharing a dataset, running remote inference or starting jobs on our infra, you are for sure using it! And this is only the beginning so give a star if you wanna follow the project π https://github.com/huggingface/huggingface_hub
- 1 reply
reacted to
abidlabs's
post with ππ₯
2 months ago
Post
4310
π Hi Gradio community,
I'm excited to share that Gradio 5 will launch in October with improvements across security, performance, SEO, design (see the screenshot for Gradio 4 vs. Gradio 5), and user experience, making Gradio a mature framework for web-based ML applications.
Gradio 5 is currently in beta, so if you'd like to try it out early, please refer to the instructions below:
---------- Installation -------------
Gradio 5 depends on Python 3.10 or higher, so if you are running Gradio locally, please ensure that you have Python 3.10 or higher, or download it here: https://www.python.org/downloads/
* Locally: If you are running gradio locally, simply install the release candidate with
* Spaces: If you would like to update an existing gradio Space to use Gradio 5, you can simply update the
In most cases, thatβs all you have to do to run Gradio 5.0. If you start your Gradio application, you should see your Gradio app running, with a fresh new UI.
-----------------------------
Fore more information, please see: https://github.com/gradio-app/gradio/issues/9463
I'm excited to share that Gradio 5 will launch in October with improvements across security, performance, SEO, design (see the screenshot for Gradio 4 vs. Gradio 5), and user experience, making Gradio a mature framework for web-based ML applications.
Gradio 5 is currently in beta, so if you'd like to try it out early, please refer to the instructions below:
---------- Installation -------------
Gradio 5 depends on Python 3.10 or higher, so if you are running Gradio locally, please ensure that you have Python 3.10 or higher, or download it here: https://www.python.org/downloads/
* Locally: If you are running gradio locally, simply install the release candidate with
pip install gradio --pre* Spaces: If you would like to update an existing gradio Space to use Gradio 5, you can simply update the
sdk_version to be 5.0.0b3 in the README.md file on Spaces.In most cases, thatβs all you have to do to run Gradio 5.0. If you start your Gradio application, you should see your Gradio app running, with a fresh new UI.
-----------------------------
Fore more information, please see: https://github.com/gradio-app/gradio/issues/9463
- 2 replies
posted
an
update
2 months ago
Post
2496
A 'small' MobileNet-V4 update, I just pushed weights for the smallest model I've trained in the series, a 0.5 width multiplier version of the MobileNet-V4 Conv Small.
Now you may look at this and say hey, why is this impressive? 64.8% top-1 and 2.2M params? MobileNetV3-Small 0.75, and MobileNet-V2 0.5 are both fewer params (at ~2M) and over 65% top-1, what gives? Well this is where MobileNet-V4 differs from the previous versions of the model family, it trades off (gives up) a little parameter efficiency for some computational efficiency.
So, let's look at the speed. On a 4090 w/ torchcompile
* 98K img/sec - timm/mobilenetv4_conv_small_050.e3000_r224_in1k
* 58K img/sec - timm/mobilenetv3_small_075.lamb_in1k
* 37K img/sec - timm/mobilenetv2_050.lamb_in1k
And there you go, if you have a need for speed, MNV4 is the better option.
Now you may look at this and say hey, why is this impressive? 64.8% top-1 and 2.2M params? MobileNetV3-Small 0.75, and MobileNet-V2 0.5 are both fewer params (at ~2M) and over 65% top-1, what gives? Well this is where MobileNet-V4 differs from the previous versions of the model family, it trades off (gives up) a little parameter efficiency for some computational efficiency.
So, let's look at the speed. On a 4090 w/ torchcompile
* 98K img/sec - timm/mobilenetv4_conv_small_050.e3000_r224_in1k
* 58K img/sec - timm/mobilenetv3_small_075.lamb_in1k
* 37K img/sec - timm/mobilenetv2_050.lamb_in1k
And there you go, if you have a need for speed, MNV4 is the better option.
reacted to
cbensimon's
post with π
3 months ago
Post
4276
Hello everybody,
We've rolled out a major update to ZeroGPU! All the Spaces are now running on it.
Major improvements:
1. GPU cold starts about twice as fast!
2. RAM usage reduced by two-thirds, allowing more effective resource usage, meaning more GPUs for the community!
3. ZeroGPU initializations (coldstarts) can now be tracked and displayed (use
4. Improved compatibility and PyTorch integration, increasing ZeroGPU compatible spaces without requiring any modifications!
Feel free to answer in the post if you have any questions
π€ Best regards,
Charles
We've rolled out a major update to ZeroGPU! All the Spaces are now running on it.
Major improvements:
1. GPU cold starts about twice as fast!
2. RAM usage reduced by two-thirds, allowing more effective resource usage, meaning more GPUs for the community!
3. ZeroGPU initializations (coldstarts) can now be tracked and displayed (use
progress=gr.Progress(track_tqdm=True))4. Improved compatibility and PyTorch integration, increasing ZeroGPU compatible spaces without requiring any modifications!
Feel free to answer in the post if you have any questions
π€ Best regards,
Charles
reacted to
merve's
post with π
3 months ago
Post
3815
If you have documents that do not only have text and you're doing retrieval or RAG (using OCR and LLMs), give it up and give ColPali and vision language models a try π€
Why? Documents consist of multiple modalities: layout, table, text, chart, images. Document processing pipelines often consist of multiple models and they're immensely brittle and slow. π₯²
How? ColPali is a ColBERT-like document retrieval model built on PaliGemma, it operates over image patches directly, and indexing takes far less time with more accuracy. You can use it for retrieval, and if you want to do retrieval augmented generation, find the closest document, and do not process it, give it directly to a VLM like Qwen2-VL (as image input) and give your text query. π€
This is much faster + you do not lose out on any information + much easier to maintain too! π₯³
Multimodal RAG merve/multimodal-rag-66d97602e781122aae0a5139 π¬
Document AI (made it way before, for folks who want structured input/output and can fine-tune a model) merve/awesome-document-ai-65ef1cdc2e97ef9cc85c898e π
Why? Documents consist of multiple modalities: layout, table, text, chart, images. Document processing pipelines often consist of multiple models and they're immensely brittle and slow. π₯²
How? ColPali is a ColBERT-like document retrieval model built on PaliGemma, it operates over image patches directly, and indexing takes far less time with more accuracy. You can use it for retrieval, and if you want to do retrieval augmented generation, find the closest document, and do not process it, give it directly to a VLM like Qwen2-VL (as image input) and give your text query. π€
This is much faster + you do not lose out on any information + much easier to maintain too! π₯³
Multimodal RAG merve/multimodal-rag-66d97602e781122aae0a5139 π¬
Document AI (made it way before, for folks who want structured input/output and can fine-tune a model) merve/awesome-document-ai-65ef1cdc2e97ef9cc85c898e π
- 2 replies
reacted to
maxiw's
post with π
3 months ago
Post
2417
The new Qwen-2 VL models seem to perform quite well in object detection. You can prompt them to respond with bounding boxes in a reference frame of 1k x 1k pixels and scale those boxes to the original image size.
You can try it out with my space maxiw/Qwen2-VL-Detection
You can try it out with my space maxiw/Qwen2-VL-Detection
- 4 replies
posted
an
update
3 months ago
Post
1282
The
Also worth pointing out, there are three rather newish 'test' models that you'll see at the top of any samples/sec comparison:
* test_vit ( timm/test_vit.r160_in1k)
* test_efficientnet ( timm/test_efficientnet.r160_in1k)
* test_byobnet ( timm/test_byobnet.r160_in1k, a mix of resnet, darknet, effnet/regnet like blocks)
They are < 0.5M params, insanely fast and originally intended for unit testing w/ real weights. They have awful ImageNet top-1, it's rare to have anyone bother to train a model this small on ImageNet (the classifier is roughly 30-70% of the param count!). However, they are FAST on very limited hadware and you can fine-tune them well on small data. Could be the model you're looking for?
timm leaderboard
timm/leaderboard has been updated with the ability to select different hardware benchmark sets: RTX4090, RTX3090, two different CPUs along with some NCHW / NHWC layout and torch.compile (dynamo) variations. Also worth pointing out, there are three rather newish 'test' models that you'll see at the top of any samples/sec comparison:
* test_vit ( timm/test_vit.r160_in1k)
* test_efficientnet ( timm/test_efficientnet.r160_in1k)
* test_byobnet ( timm/test_byobnet.r160_in1k, a mix of resnet, darknet, effnet/regnet like blocks)
They are < 0.5M params, insanely fast and originally intended for unit testing w/ real weights. They have awful ImageNet top-1, it's rare to have anyone bother to train a model this small on ImageNet (the classifier is roughly 30-70% of the param count!). However, they are FAST on very limited hadware and you can fine-tune them well on small data. Could be the model you're looking for?
reacted to
merve's
post with π₯
3 months ago
Post
5523
I have put together a notebook on Multimodal RAG, where we do not process the documents with hefty pipelines but natively use:
- vidore/colpali for retrieval π it doesn't need indexing with image-text pairs but just images!
- Qwen/Qwen2-VL-2B-Instruct for generation π¬ directly feed images as is to a vision language model with no processing to text!
I used ColPali implementation of the new π Byaldi library by @bclavie π€
https://github.com/answerdotai/byaldi
Link to notebook: https://github.com/merveenoyan/smol-vision/blob/main/ColPali_%2B_Qwen2_VL.ipynb
- vidore/colpali for retrieval π it doesn't need indexing with image-text pairs but just images!
- Qwen/Qwen2-VL-2B-Instruct for generation π¬ directly feed images as is to a vision language model with no processing to text!
I used ColPali implementation of the new π Byaldi library by @bclavie π€
https://github.com/answerdotai/byaldi
Link to notebook: https://github.com/merveenoyan/smol-vision/blob/main/ColPali_%2B_Qwen2_VL.ipynb