Category: VLMs

In this article, we fine-tune the DeepSeek-OCR 2 model using Unsloth for Hindi language OCR. We create a simple Gradio application to run inference and check the diff between the original and the inference result. ...

In this article, we discuss the DeepSeek-OCR 2 paper. We start from the DeepEncoder V2, the architecture, and finally discuss the code from Hugging Face. ...

In this article, we cover inference using DeepSeek-OCR 2. We create a simple pipeline where we can provide either to path to a PDF or an image for processing. We also create a Gradio application for for better experience with DeepSeek-OCR 2. ...

In this article, are grounding the Qwen3-VL object detection capabilities with SAM2 segmentation. The pipeline uses Qwen3-VL to detect objects via natural language whose coordinates are then fed to the SAM2 model for segmentation. ...

In this article, we are fine-tuning the Qwen3-VL 2B model for sketch and image to HTML. After fine-tuning, we will be able to feed an image of a website to the model and get the HTML code for it. ...