ML Researcher - Scalable Image & Video Diffusion

Krea; San Francisco, Northern (CA, KY); Hybrid; USD 140,000 - 240,000; Full time

Benefits offered by this job

Health & wellness

Flexible PTO

401k with 4% company match

Meals in the office

Transit: Ubers covered

Visa sponsorship

Job summary

Krea in San Francisco is seeking an experienced Researcher with engineering skills to advance large-scale image and video model training. You will work on diffusion models, optimize distributed training pipelines, and experiment with architecture, data, and training strategies to push model quality and reliability.

The role requires hands-on experience with PyTorch, distributed training, and data pipelines.

Qualifications

• Proven track record in working with image or video models at scale (publications or open-source contributions a plus).
• Strong proficiency in PyTorch and understanding of its inner workings.
• Strong background in distributed training paradigms such as FSDP, CP, SP, USP, TP, and EP. Knowing how different parallelism strategies work together and their tradeoffs.
• Experience in profiling and debugging large distributed training. Being comfortable with analyzing traces to identify bottlenecks and look for improvements.
• Good knowledge of low precision training / inference in FP8, NVFP4, and MXFP8.
• Solid understanding of diffusion model training pipeline across pretraining, midtraining, preference optimization, and reinforcement learning.
• Keeping up with the developments in related fields such as LLM, VLM, representation learning, and robotics research.
• Being comfortable working in a goal-oriented research environment.
• Having good judgement around when one should explore different training strategies and when it's time to commit to a specific strategy to scale compute and data.
• Be comfortable getting your hands dirty with data and designing custom data pipelines to improve data quality.

Responsibilities

• Train diffusion models for image and video generation on large GPU clusters.
• Fully optimize and profile large distributed training runs across model architectures, kernels, data loading, memory constraints, and communication.
• Implement and improve various distributed training strategies including FSDP, CP, SP, TP, and EP.
• Continuously improve model quality and reliability through data, model architecture, training pipeline, structuring experiments, and eval design.
• Debug distributed training errors and implement fault tolerance solutions, identifying bad GPU, NVLink, Infiniband (IB) components as well as monitoring numerical errors and NCCL issues.
• Ablate different architecture, attention, optimizer, data, and algorithmic choices to reliably improve efficiency and performance of our models.

Skills

Image models

PyTorch

Distributed training

Profiling

FP8 training

Diffusion models

Experiment design

Tools

GPU clusters

Job description

Krea in San Francisco is seeking an experienced Researcher with engineering skills to advance large-scale image and video model training. You will work on diffusion models, optimize distributed training pipelines, and experiment with architecture, data, and training strategies to push model quality and reliability.

The role requires hands-on experience with PyTorch, distributed training, and data pipelines.