LG AI Research

LLM Inference Engineer

LG AI Research
Apply
1 day ago
Seoul, Korea, SouthMid Level

Responsibilities

  • Design and participate in improvements for large language model inference performance.
  • Develop serving engines that reliably serve EXAONE models across diverse frameworks.
  • Implement requested features, bug fixes, and performance optimizations through official open-source channels for vLLM, SGLang, Ollama, and similar projects.
  • Design and improve chat_template usage so LLMs operate correctly across diverse languages and environments.
  • Design NPU/GPU-based serving strategies and research and implement model inference acceleration algorithms.

Requirements

  • Experience researching and implementing algorithms to improve model inference performance and implementing inference backends.
  • Experience designing and developing NPU/GPU-based model serving strategies.
  • Experience contributing to open-source LLM serving frameworks such as vLLM, SGLang, or Ollama.
  • Preferred: understanding and experience across the machine-learning model lifecycle, including development, training, serving, and deployment.
  • Preferred: strong understanding of AI and machine-learning technologies, including Python, C++, TensorFlow, PyTorch, ONNX, CUDA, and NPU.
  • Preferred: deep understanding of transformer-based models, attention mechanisms, and LLM architecture and operation.
  • Preferred: understanding of LLM decoding strategies and inference techniques such as Greedy Search, Beam Search, and Top-k Sampling.
  • Preferred: experience designing and applying chat_template for diverse environments.

Tech Stack

LG AI Research

About LG AI Research

201-500 employees

LG AI Research is the AI research arm of LG Group, building foundation models and enterprise AI services such as EXAONE and Chat EXAONE. It develops and operates data and API platforms and computing infrastructure to help LG affiliates and industry partners deploy AI for real-world problems. Founded in 2020 and headquartered in Seoul, it focuses on multimodal models and applications across manufacturing, consumer electronics, and other sectors.

Contact me