Responsibilities
1. Responsible for using big model technology to solve multimedia quality problems under the multimedia quality direction of Douyin products, including but not limited to designing audio and video quality evaluation models based on MLLM, understanding/evaluation of AIGC generated content (image and video generation, audio generation, etc.), user experience big model, audio and video negative case mining algorithm, multimodal content understanding, etc. 2. Responsible for the optimization of the deployment of deep learning models, and be able to cooperate with the engineering team to complete the online development and application of the model 3. Based on business models and algorithm technologies, explore product experience issues, and work closely with the R&D team to promote business implementation and improve product user experience 4. Follow the team to participate in school-enterprise horizontal projects, academic forums and conferences, data science competitions, etc., and assist the team in building academic/industrial influence.
Qualifications
1. Bachelor degree or above in computer science or artificial intelligence related majors, with experience in algorithm research and implementation related to deep learning, and practical experience in LLM development and application 2. Familiar with mainstream model architectures such as CNN/Transformer/Vit/BLIP/BERT, and proficient in machine learning frameworks such as Pytorch, Tensorflow, Caffe, and Keras 3. Have certain algorithm experience in multimodal large models (SFT/PE/RLHF/RAG), including training, fine-tuning and evaluation of MLLM, have practical experience in large model deployment, master mainstream large model deployment frameworks such as vLLM and TRT, and be familiar with the basic solutions for large model reasoning optimization 4. Familiar with multimedia generation algorithms such as images and audio, such as VAE, DIFFUSION and other basic architectures, and understand the evaluation methods of generation models. Bonus points: 1. Have engineering experience in algorithm engineering, model distillation, model deployment, model operator tuning, SDK design, etc., and have strong engineering capabilities 2. Priority will be given to those who have published articles in top machine learning and computer vision conferences such as CVPR, ICCV, ECCV, ICML, NeurlIPS, or have achieved excellent results in well-known data science competitions such as CVPR NTIRE, Kaggle, CCF, Tianchi, etc.
