Job Description

软件工程师 - 大模型端侧优化和部署

北京、广州、上海

全职

职位描述

负责将大模型部署到自动驾驶车端芯片平台（NVIDIA Orin / Thor 等），并进行极致的性能、内存与能效优化。你将在算力、内存、功耗与实时性严格受限、且对安全与稳定性要求极高的车载环境下，打通从推理引擎定制、核心算子优化到车端运行时管理的全栈链路，推动大模型在自动驾驶（端到端决策、场景理解等）场景的规模化落地。
岗位职责
1. 适配、移植与深度优化车端推理引擎/运行时（TensorRT-LLM、TensorRT Edge-LLM、vLLM、SGLang 等），通过计算图优化、算子融合、高性能 CUDA 算子开发及 KV Cache、投机采样等优化，充分发挥 Orin / Thor 平台算力，满足实时性要求；
2. 建立 Benchmark 与 Profiling 体系，跟踪首字延迟（TTFT）、吞吐、内存峰值、冷启动、模型切换、cache 命中率、长稳与抖动等车端核心指标。支持场景理解、Function Calling、结构化输出、流式输出、多模态推理等核心场景的低延迟落地；

职位要求

1. 计算机、电子工程、人工智能、自动化等相关专业，本科及以上学历；
2. 具备 2 年以上 AI 推理部署、端侧AI、系统性能优化或模型工程化相关经验；
3. 精通 C/C++ 与 Python，具备扎实的计算机体系结构知识，熟悉 Linux 下的开发与调试；
4. 深入理解 Transformer / LLM / VLM 推理机制，熟悉 prefill、decode、KV Cache、attention、batching、streaming 等关键概念；
5. 精通至少一种推理框架（TensorRT-LLM / vLLM / SGLang / ONNX Runtime 等），具备引擎定制、算子开发或 graph 优化经验；
6. 精通大模型核心算子（Attention、FFN/MoE、LayerNorm/RMSNorm）的底层实现与性能优化方法；
加分项
1. 有 Qwen 等主流开源大模型的部署经验；
2. 有在 NVIDIA Orin / Thor 等车端平台部署大模型的经验（熟悉 TensorRT Edge-LLM 等工具链，了解 Orin 支持 FP16/INT8/INT4、Thor 支持 FP8/NVFP4 等平台特性差异）；
3. 有自动驾驶大模型量产落地经验，有端到端自动驾驶大模型或多模态大模型部署优化经验。

投递

About Pony.ai

Pony AI Inc. (“Pony.ai”) is a global leader in the large-scale commercialization of autonomous mobility.

Leveraging its vehicle-agnostic Virtual Driver technology, full-stack autonomous driving technology that seamlessly integrates its proprietary software, hardware, and services, Pony.ai is developing a commercially viable and sustainable business model that enables the mass production and deployment of vehicles across transportation use cases.

Founded in 2016, Pony.ai has expanded its presence across China, Europe, East Asia, the Middle East, and other regions, ensuring widespread accessibility to its advanced technology.

Pony.ai is among the first in China to obtain licenses to operate fully driverless vehicles in all four Tier-1 cities in China (Beijing, Guangzhou, Shanghai, Shenzhen) and has begun to offer public-facing, fare-charging robotaxi services without safety drivers in Beijing, Guangzhou and Shenzhen. Pony.ai operates a fleet consisting of over 250 robotaxis.

To date, Pony.ai has driven nearly 45 million autonomous testing and operation kilometers on open roads worldwide.

Industry

IT & Software

Company Size

501-1,000 employees

Headquarters

Fremont, California

Year Founded

2016

Website

pony.ai

Social Media