Sahaj.blogs

Day page

5 Aug 2025

Paper summary

Paper summary: Qwen-Image Technical Report

Qwen-Image is an image-generation foundation model focused on rendering complex text (including Chinese) inside images and on consistent image editing, using a text-rendering curriculum and dual encoding of the input image.

  • A curriculum moves from images without text, to simple text, up to paragraph-level text, to teach native text rendering.
  • For editing, the input image goes to both Qwen2.5-VL (meaning) and a VAE (pixel detail), balancing semantic consistency and visual fidelity.

Read the post →

AI updates

gpt-oss: OpenAI releases open-weight models under Apache 2.0

OpenAI released gpt-oss-120b and gpt-oss-20b, mixture-of-experts reasoning models quantised with MXFP4 and licensed Apache 2.0; the large one fits on one H100 and the small one in 16GB.

  • Both are MoE models stored in a 4-bit MXFP4 scheme, so the 120B model fits on a single H100 GPU.
  • The release is under Apache 2.0, with a minimal usage policy.

Read the post →