# Open-source model

Qwen2.5-1M is an open-source AI language model designed for long sequence tasks, with support for context lengths of up to 1 million tokens. Through innovative training methods and technical optimizations, it significantly enhances the performance and efficiency of long sequence processing. This model excels in long context tasks while maintaining strong performance in short text scenarios, making it an excellent open-source alternative among existing long context models. It is suitable for applications involving extensive text data, such as document analysis and information retrieval, providing developers with robust language processing capabilities.

LongWriter is a long text generation model developed by a team from Tsinghua University. Based on large-scale language models (LLMs), it can generate text contents exceeding 10,000 words. This model is particularly suitable for scenarios that require the generation of long, coherent texts, such as writing assistance and content creation. LongWriter has improved the quality and consistency of the generated text through fine-tuning and optimization while maintaining the model's efficiency and scalability.

PIXART LCM is a text-to-image synthesis framework that integrates the latent consistency model (LCM) and ControlNet into the advanced PIXART-α model. PIXART LCM is renowned for its ability to generate high-quality 1024px resolution images through an efficient training process. Integrating LCM in PIXART-δ significantly accelerates inference speed, allowing for the generation of high-quality images in just 2-4 steps. Notably, PIXART-δ achieves the milestone of generating 1024x1024 pixel images within 0.5 seconds, a 7-fold improvement over PIXART-α. Furthermore, PIXART-δ is meticulously designed for efficient training on a 32GB V100GPU within a single day. With 8-bit inference capability, PIXART-δ can synthesize 1024px images under an 8GB GPU memory constraint, considerably enhancing its usability and accessibility. Additionally, the introduction of a ControlNet-like module enables fine-grained control over text-to-image diffusion models. We propose a novel ControlNet-Transformer architecture, specifically tailored for Transformers, achieving explicit controllability and high-quality image generation. As a leading open-source image generation model, PIXART-δ offers a promising alternative within the stable diffusion model family, significantly contributing to the field of text-to-image synthesis.