ML Engineer - Post-Training & Evaluation for Visual Gen

Nunchux AI; San Francisco (CA); Hybrid; USD 180,000 - 250,000; Full time

Benefits offered by this job

Health insurance

401(k)

Equity

Job summary

Nunchux AI is hiring a Machine Learning Engineer in San Francisco to build post-training pipelines and evaluation systems for visual generative models. You will curate data, develop benchmarks, and validate recipes to improve efficiency and quality for production deployments.

The role emphasizes translating research into robust production workflows, collaborating with CMU/ MIT–affiliated researchers, and advancing post-training methods on multi-GPU systems.

Qualifications

• Hands-on experience with image or video generation models.
• Depth in post-training methods or evaluation of visual generative models.
• Strong Python and PyTorch skills with maintainable code for production workflows.
• Experience running post-training workloads across multiple GPUs with FSDP, DeepSpeed, or similar tools.
• Ability to design clean experiments, interpret results, and translate data into practical recommendations.

Responsibilities

• Develop post-training recipes for image and video generation models.
• Build data pipelines to curate and version training and benchmark data.
• Create benchmarks and automated judges; run human preference studies to measure quality and efficiency.
• Benchmark and release models; measure against baselines and document evidence for decisions.
• Bridge research into production by integrating methods into the team’s pipelines.

Skills

Visual generation

Post-training/eval depth

Python & PyTorch

Distributed training

Experiment design

Tools

Python

PyTorch

Job description

Nunchux AI is hiring a Machine Learning Engineer in San Francisco to build post-training pipelines and evaluation systems for visual generative models. You will curate data, develop benchmarks, and validate recipes to improve efficiency and quality for production deployments.

The role emphasizes translating research into robust production workflows, collaborating with CMU/ MIT–affiliated researchers, and advancing post-training methods on multi-GPU systems.