> Markdown version of [/jobs/ext/3743678-machine-learning-engineer-model-quantization-tesla-si](https://www.wearedevelopers.com/jobs/ext/3743678-machine-learning-engineer-model-quantization-tesla-si). Every page supports `.md` or `Accept: text/markdown`. Links point to the HTML versions so they work for humans too. Agent guide: [/agents.md](https://www.wearedevelopers.com/agents.md).

---

# Machine Learning Engineer, Model Quantization, Tesla SI

- **Company:** Tesla Motors
- **Location:** Palo Alto, CA, United States
- **Experience:** Experienced
- **Salary:** $124,000.0
- **Contract:** Permanent contract
- **Skills:** Artificial Intelligence, Computer Vision, C++ (Programming Language), Nvidia CUDA, Extract Transform Load (ETL), Dynamic Random-Access Memory, Python (Programming Language), Machine Learning, Software Engineering, Pytorch, Deep Learning, Discretization, Information Technology, Low Latency, Transformer Architectures, Model Quantization
- **Published:** October 15, 2026
- **Apply:** https://jobs.localjobnetwork.com/apply/add/87229367/1

## About the Role

* Degree or equivalent experience in Computer Science, Machine Learning, Robotics, Computer Vision, or related quantitative field
 * 2+ years of hands-on experience training, optimizing, and deploying large-scale quantized deep learning models
 * Strong technical understanding of the challenges inherent to quantizing large transformer architectures, including mitigating massive activation outliers, KV cache quantization, and maintaining the numerical stability of attention mechanisms at low precision
 * Deep expertise in the theory and low-level implementation of modern quantization algorithms (e.g., GPTQ, AWQ, SmoothQuant, OmniQuant)
 * Experience with low-level numerics and emerging data formats (e.g., FP8, INT4, W4A8, W8A8, micro-scaling/MX formats) and their trade-offs regarding latency, memory bandwidth, and model fidelity
 * Rigorous understanding of computer architecture and the roofline model. Familiarity with how to optimize for memory hierarchies, minimize SRAM/DRAM data movement, and efficiently map quantized GEMMs and memory-bound operators to custom silicon
 * Proficiency in writing custom CUDA/Triton kernels, implementing custom autograd functions (e.g., Straight-Through Estimators), and manipulating PyTorch computational graphs (e.g., FX tracing, torch.compile)
 * Strong software engineering skills - clean, production-grade Python/C++ code that ships reliably at scale
 * Proven ability to turn cutting-edge research into robust, real-world systems that improve safety, capability, efficiency, or digital productivity
 * Passion for Tesla's mission and excitement about deploying AI that moves both the physical and digital worlds forward

## Description

If you want your models to solve real physics, causality, long-horizon planning, dexterous manipulation, and autonomous digital task execution, this is the highest-leverage AI role on the planet. Let's build the future together.
What You'll Do

 * Architect and scale quantization pipelines (both Post-Training Quantization and Quantization-Aware Training) for massive multi-modal foundation models that fuse vision, prediction, and decision-making. You will optimize inference latency, memory bandwidth utilization, and power consumption for self-driving cars, Optimus robots, and digital agents operating at enterprise scale
 * Innovate quantization-aware-training recipes and algorithms that tackle complex optimization challenges inherent to low-precision training
 * Push the limits of low-precision AI: Research and implement advanced low-bitweight post-training quantization techniques to address hard algorithmic problems such as activation outlier mitigation, KV cache compression, and optimal layer-wise bit-allocation while strictly maintaining model accuracy
 * Collaborate closely with AI compiler, inference engine, and silicon teams to ensure models are architected to maximally utilize underlying hardware capabilities by co-designing quantization-friendly architectures, hardware-aware sparsity patterns, and mixed low-precision kernels
 * Collaborate across perception, planning, robotics, digital agents, and infrastructure teams to move models from research to fleet-wide, robot-wide, and enterprise-wide deployment