Software Engineer - Model Performance
Baseten150,000 - 250,000 USD/year
- Full Time
- Mid-level (3 to 4 years), Senior (5 to 8 years)
$155,000 – $250,000Compensation
Mid-level (3 to 4 years), Senior (5 to 8 years)Experience Level
Full TimeJob Type
UnknownVisa
AI & Machine Learning, HardwareIndustries
Requirements
Candidates must possess strong programming skills in Python and have experience with machine learning frameworks such as PyTorch, TensorFlow, or JAX. Hands-on experience with model optimization, quantization, and inference acceleration is required, along with a deep understanding of Transformer architectures and distributed inference techniques. Knowledge of quantization methods and memory-efficient inference techniques is essential, as well as a solid grasp of software engineering best practices including CI/CD and containerization technologies like Docker and Kubernetes.
Responsibilities
The Machine Learning Engineer will design, build, and optimize machine learning deployment pipelines for large-scale models. They will implement and enhance model inference frameworks and develop automated workflows for model development, experimentation, and deployment. Collaboration with research, architecture, and engineering teams to improve model performance and efficiency is expected. The role also involves working with distributed computing frameworks to optimize model parallelism and deployment, implementing scalable KV caching and memory-efficient inference techniques, and monitoring and optimizing infrastructure performance across various levels of custom hardware.
Skills
Python
PyTorch
TensorFlow
JAX
Model Optimization
Quantization
Inference Acceleration
Transformer Architectures
Attention Mechanisms
Distributed Inference
Tensor Parallel
Pipeline Parallel
Sequence Parallel
KV Caching
Ray
d-Matrix
AI compute platform for datacenters
About d-Matrix
d-Matrix focuses on improving the efficiency of AI computing for large datacenter customers. Its main product is the digital in-memory compute (DIMC) engine, which combines computing capabilities directly within programmable memory. This design helps reduce power consumption and enhances data processing speed while ensuring accuracy. d-Matrix differentiates itself from competitors by offering a modular and scalable approach, utilizing low-power chiplets that can be tailored for different applications. The company's goal is to provide high-performance, energy-efficient AI inference solutions to large-scale datacenter operators.
Key Metrics
Santa Clara, CaliforniaHeadquarters
2019Year Founded
$149.8MTotal Funding
SERIES_BCompany Stage
Enterprise Software, AI & Machine LearningIndustries
201-500Employees