Description
We develop and implement advanced methods for multimodal generation. The following modalities are used as input – images/text/sound/video, output u– video/sound. Focus – developing new architectures, training large models (tens/hundreds of billions of parameters), optimizing and reducing inference cost.
Responsibilities
- research and development of new methods for multimodal generation (research of existing architectures and development of new ones).
- research and development of methods for image to video, first-last frame to video, video continuation generation. Optimization of the training and inference pipeline – dataset collection, model architecture development, training, evaluation, optimization of inference speed and generation cost.
- research and development of methods for text(image) to video+audio generation. Optimization of the training and inference pipeline – dataset collection, model architecture development, training, evaluation, optimization of inference speed and generation cost.
- collaborative work with open-source and product teams to ensure stable and fast model inference in different environments and on different platforms – GigaChat/ComfyUI/diffusers etc.
- technical description of developed solutions – articles, documentation, technical reports.
Requirements
- expert level in Python, PyTorch.
- experience in developing audio/video generation models.
- deep understanding of training/distributed training methods.
- understanding of architectures of modern LLMs and Diffusion models.
- experience working with diffusion models.
Bonus: Experience in classical audio/video processing tasks. Digital signal processing/compression/noise reduction.
Conditions
- comfortable modern office near Kutuzovskaya metro station
- hybrid work format
- annual salary review, quarterly and annual bonus
- corporate gym and recreation areas
- more than 400 educational programs from SberUniversity for professional and career development
- adaptation program and manager assistance at the start
- extended VHI, preferential insurance for family and corporate pension program
- mortgage up to 7% more favorable for every employee
- free SberPrime+ subscription, discounts on products from partner companies
- reward for recommending friends to the Sber team
