Pavel Iakubovskii
@qubvelx
Followers
284
Following
548
Media
8
Statuses
101
ML Engineer @ 🤗 | Kaggle Competition Master | Creator of segmentation_models.pytorch
Lisbon, Portugal
Joined May 2020
RT @joao_gante: LET'S GO! Cursor using local 🤗 transformers models! You can now test ANY transformers-compatible LLM against your codebase.….
0
40
0
RT @NielsRogge: New model alert in Transformers: EoMT!. EoMT greatly simplifies the design of ViTs for image segmentation 🙌. Unlike Mask2Fo….
0
52
0
RT @LysandreJik: BOOOM! transformers now has a baked-in http server w/ OpenAI spec compatible API. Launch it with `transformers serve` and….
0
29
0
RT @NielsRogge: Another classic has made it into the Transformers library: LightGlue (ICCV '23)🔥. A deep neural network that learns to matc….
0
43
0
🔧 Easy Fine-Tuning Notebook .Colab notebook that shows you how to fine-tune V-JEPA 2 on your video clips to create a custom video classification model.
0
0
0
👋 Live Webcam Demo on HF Spaces . This is the super fun part, you can use your own webcam to see the model in action. Fire it up, do something in front of the camera, and watch as the V-JEPA 2 tries to guess what you're doing in real-time.
1
0
2
🎬 4 Video Classification Models . We've converted to 🤗 Transformers 4 pretrained video classification models. The star of the show is a model fine-tuned on the huge "Something-Something-v2" dataset.
1
0
1
V-JEPA 2 is an update of the first world model trained on video that achieves state-of-the-art visual understanding, enabling zero-shot robot control in new environments. Now, we've made it even easier to play with and build on. Here’s what’s new:.
1
0
1
💥 💥 💥 We've just rolled out an update for @AIatMeta's V-JEPA 2 📹 models on @huggingface Hub 🤗
1
5
25
RT @soumikRakshit96: ✨ CVPR 2025 highlight: A Distractor-Aware Memory for Visual Object Tracking with SAM2. the authors propose a new distr….
0
56
0
🔧 This is a small step forward, focused on `pipeline()`, but there's more work to do across the board. Improving typings throughout the library is definitely on our radar. We hope this change makes things a bit smoother for you and we'd love your help and feedback!.
0
0
1
Thanks to the magic of `overload` in Python, we can give you a much better experience: . ✅ Correctly inferred types for instantiated `pipeline()`, including matching docstrings and input signatures .✅ Smarter return types depending on your input (single vs. batch)
1
0
2
But behind the scenes, `pipeline()` can return 20+ different pipeline classes, each with unique arguments and return types. Until now, that made working with them in IDEs not ideal - you often had to rely on external docs or code examples to figure things out.
1
0
1
🚀 Better type annotations for 🤗 Transformers pipeline()!. `pipeline()` is a powerful and very flexible high-level API - just one line, and you're up and running with anything from object detection to text generation.
3
6
23
V-JEPA 2 - video embeddings model in HF transformers from day zero. See We will release video classification pretrained models soon!.
Our vision is for AI that uses world models to adapt in new and dynamic environments and efficiently learn new skills. We’re sharing V-JEPA 2, a new world model with state-of-the-art performance in visual understanding and prediction. V-JEPA 2 is a 1.2 billion-parameter model,
2
7
19
RT @LysandreJik: The Transformers library is undergoing it's largest pivot to date 🙌. It now cements its role as the central model definiti….
0
59
0
RT @ariG23498: Fine tune Gemma 3 for object detection. I am a big fan of PaliGemma (@giffmana et. al.). They showed us how image specific t….
0
14
0
RT @mervenoyann: VLMS 2025 UPDATE 🔥. We just shipped a blog on everything latest on vision language models, including.🤖 GUI agents, agentic….
0
123
0
RT @ariG23498: I have seen that Hugging Face really values ideas. Today @mervenoyann pointed out that while reading the Qwen 2.5 VL paper….
0
15
0
RT @NielsRogge: New model alert in @huggingface Transformers! 🔥. D-FINE (@iclr_conf '25 spotlight) was just added. SOTA real-time object de….
0
53
0