Qwen2-7B is a 7-billion-parameter model that delivers high performance and accuracy. It efficiently runs on mid-range GPUs and serves as a foundation for developing specialized solutions across various domains.
Qwen2-1.5B is a lightweight, balanced model with 1.5 billion parameters, designed for basic tasks on local machines and small servers. It delivers solid performance in text generation, summarization, and translation while maintaining moderate resource requirements.
Qwen2-0.5B is an ultra-compact model with 0.5 billion parameters and a 32K context window, optimized for deployment on mobile devices and IoT systems. It is suitable for building simple applications and text autocompletion systems.
An incredibly popular multilingual model in the community, trained on 15 trillion tokens, with 8 billion parameters and a context window of 128K. The model is adapted to solve a wide range of tasks, supports function calling, and is ideally suited for building intelligent dialogue systems, software assistants, and agent applications.
A fine-tuned version of Llama-3-8B-Instruct, trained using a high-quality synthetic dataset generated by the GPT-4o model and optimized for Russian language processing. This model combines a compact size (8B parameters) with strong results in Russian, surpassing GPT-3.5-turbo in this aspect while maintaining English performance on par with the base version, making it a powerful and balanced solution for Russian-language dialog systems.
Phi-3.5-mini is a compact and highly efficient language model capable of running on mobile and edge devices, delivering generation quality comparable to that of larger models. Thanks to optimized training on high-quality data and multilingual support, it is ideal for chatbots, educational applications, and tasks with limited computational resources.
A legendary open-source model, released in April 2024, which became one of the starting points for the rapid development of the open-source AI ecosystem. The 8-billion-parameter version combines compactness with high performance, utilizing Grouped-Query Attention (GQA) techniques. Combined with training on 15+ trillion tokens, this ensures fast, high-quality text and code generation and makes it accessible for local deployment.
This is diffusion text-to-image model designed to generate high-aesthetic images of 1024x1024 pixels, including portraits and landscapes. It is the successor to Playground v2 and demonstrates superiority over models like SDXL, PixArt-α, DALL-E 3, and Midjourney 5.2 in user studies focused on aesthetic quality.
This is an image generation model based on text prompts (text-to-image), built on the architecture of Stable Diffusion XL (SDXL). It represents a fine-tuned version of the base model stabilityai/stable-diffusion-xl-base-1.0.
SDXL-Turbo is a distilled version of SDXL 1.0, which allows sampling large-scale foundational image diffusion models in 1 to 4 steps at high image quality.
Kandinsky-3 is a diffusion model for text-to-image generation, developed based on previous versions of the Kandinsky2-x family. It has been improved through expanded data volume, including information related to Russian culture, enabling the generation of images reflecting this theme. The model also demonstrates enhanced text understanding and improved visual quality due to larger text encoder and Diffusion U-Net model sizes.
This is a diffusion model developed by Stability AI for generating short video clips from a static image (image-to-video). The model creates videos up to 4 seconds long (25 frames at 576×1024 resolution), using the input image as a conditional frame.
The most popular automatic speech recognition (ASR) model from OpenAI, trained on 5 million hours of labeled audio data. It supports multilingual transcription and translation into English, demonstrating accuracy and robustness across various acoustic conditions without the need for additional fine-tuning.
Stable Diffusion XL-base-1.0 is the base text-to-image generation model, improved upon previous versions of Stable Diffusion models. It is designed to generate images at 1024x1024 pixels. It is also not recommended to choose an image size smaller than 512x512 pixels.
Refiner Model specializes in the final stages of noise reduction and enhances the visual accuracy of images generated by the base model.
Kandinsky 2.2 is a free Russian model for image generation developed by Sber AI. It operates based on a diffusion model: initially adding noise to images it was trained on, then restoring them through a reverse diffusion process to create new unique images.
The Stable Diffusion v2-1 is a diffusion-based text-to-image generation model. It is fine-tuned from the Stable Diffusion v2 checkpoint. Generates high-resolution images (up to 768x768) and supports text-guided generation.
Stable Diffusion x4 upscaler - is a diffusion model for 4x image resolution upscaling based on text prompts. It takes as input a low-resolution image and a text prompt, along with the noise_level parameter.
Stable Diffusion v1.5 is a diffusion model for image generation based on textual prompts. The model was initialized with the weights of the previous version, Stable Diffusion v1.2, and subsequently fine-tuned. It supports image generation at 512×512 pixels and image modification via textual prompts.