Software Engineer, Inference Runtime

LM Studio; New York (NY); Hybrid; USD 140,000 - 210,000; Full time

Benefits offered by this job

Competitive salary and equity grants

Catered team lunch / expensed dinners

Job summary

LM Studio is seeking an Inference Runtime Software Engineer to advance our on-device and cloud inference stack. You will integrate new engines, bring up open-weight models, and optimize performance across CPU and GPU targets.

You will contribute to open-source projects and collaborate on new model architectures and multimodal capabilities, while improving latency and reliability in diverse runtimes. Office lunches and equity grants are part of the package.

Qualifications

• Significant experience building production ML systems, inference runtimes, or performance-sensitive infrastructure.
• Strong programming ability in Python and C++.
• Deep understanding of transformer architectures and the mechanics of model inference.
• Experience profiling CPU or GPU workloads and reasoning about compute, memory, synchronization, and data movement.
• Experience with PyTorch and inference systems such as llama.cpp, MLX, ExecuTorch, vLLM, SGLang, or TensorRT-LLM.
• Strong debugging instincts across model code, runtime internals, operating systems, and CPU or GPU execution.
• Takes personal responsibility for the correctness and performance of their work.

Responsibilities

• Maintain and push forward our inference stack on-device and in the cloud.
• Bring up new model architectures and multimodal models.
• Improve latency, throughput, memory use, and reliability across CPU, CUDA, Metal, Vulkan, and ROCm runtimes.
• Build runtime capabilities for model loading, batching, scheduling, caching, and distributed execution.
• Benchmark and diagnose correctness and performance problems across the inference stack.
• Contribute upstream to open-source projects such as llama.cpp and MLX.
• Competitive salary and equity grants.
• Catered team lunch / expensed dinners in the office.

Skills

ML systems

Python & C++

Transformer models

Profiling workloads

PyTorch & inference

Debugging runtimes

Ownership & responsibility

Tools

llama.cpp

MLX

ExecuTorch

vLLM

SGLang

TensorRT-LLM

CUDA

Metal

Vulkan

ROCm

Job description

LM Studio is used by millions of people around the world to run AI on their own computers, and now with Bionic - also in the cloud. Our values prioritize putting the human in the center, and creating tools that we want to use ourselves, and recommend to our friends and family.

As a team, we work with high technical intensity and personal responsibility. We are looking for curious, self-motivated, creative, and technically excellent teammates to join us and build the future of human-AI interactions in software.

The Role

We are looking for an Inference Runtime Software Engineer to push forward LM Studio's inference stack on-device and in the cloud. You will integrate new inference engines and runtime capabilities, bring up new open-weight models and modalities, and optimize model execution for a wide range of CPU and GPU targets. You will also contribute improvements to the open-source projects we build on.

Qualifications

• Significant experience building production ML systems, inference runtimes, or performance-sensitive infrastructure
• Strong programming ability in Python and C++
• Deep understanding of transformer architectures and the mechanics of model inference
• Experience profiling CPU or GPU workloads and reasoning about compute, memory, synchronization, and data movement
• Experience with PyTorch and inference systems such as llama.cpp, MLX, ExecuTorch, vLLM, SGLang, or TensorRT-LLM
• Strong debugging instincts across model code, runtime internals, operating systems, and CPU or GPU execution
• Takes personal responsibility for the correctness and performance of their work

Bonus Qualifications

• Past contributions to open-source inference runtime projects such as llama.cpp, MLX, ExecuTorch, vLLM, SGLang, or TensorRT-LLM

Responsibilities

• Maintain and push forward our inference stack on-device and in the cloud
• Bring up new model architectures and multimodal models
• Improve latency, throughput, memory use, and reliability across CPU, CUDA, Metal, Vulkan, and ROCm runtimes
• Build runtime capabilities for model loading, batching, scheduling, caching, and distributed execution
• Benchmark and diagnose correctness and performance problems across the inference stack
• Contribute upstream to open-source projects such as llama.cpp and MLX
• Competitive salary and equity grants
• Catered team lunch / expensed dinners in the office