小米Machine Learning Engineer-（Experienced）

Responsibilities Collaborate with GPU sales team and SCE AIML TPM team to provide technical support for customers both at pre-sales and after-sales stage. Take ownership of problems and work to identify solutions. Design, deploy, and manage infrastructure components such as cloud resources, distributed computing systems, and data storage solutions to support AI/ML workflows. Collaborate with customers’ scientists and software/infrastructure engineers to understand infrastructure requirements for training, testing, and deploying machine learning models. Implement automation solutions for provisioning, configuring, and monitoring AI/ML infrastructure to streamline operations and enhance productivity. Optimize infrastructure performance by tuning parameters, optimizing resource utilization, and implementing caching and data pre-processing techniques. Troubleshoot infrastructure performance, scalability, and reliability issues and implement solutions to mitigate risks and minimize downtime. Stay updated on emerging technologies and best practices in AI/ML infrastructure and evaluate their potential impact on our systems and workflows. Document infrastructure designs, configurations, and procedures to facilitate knowledge sharing and ensure maintainability. Qualifications: Experience in scripting and automation using tools like Ansible, Terraform, and/or Kubernetes. Experience with containerization technologies (e.g., Docker, Kubernetes) and orchestration tools for managing distributed systems. Solid understanding of networking concepts, security principles, and best practices. Excellent problem-solving skills, with the ability to troubleshoot complex issues and drive resolution in a fast-paced environment. Strong communication and collaboration skills, with the ability to work effectively in cross-functional teams and convey technical concepts to non-technical stakeholders. Strong documentation skills with experience documenting infrastructure designs, configurations, procedures, and troubleshooting steps to facilitate knowledge sharing, ensure maintainability, and enhance team collaboration. Strong Linux skills with hands-on experience in Oracle Linux/RHEL/CentOS, Ubuntu, and Debian distributions, including system administration, package management, shell scripting, and performance optimization.

LLM Application in Customer Service: Design, develop, and iterate on prompts for various LLM applications within the customer service center, including conversational AI, content generation, and summarization. Experiment with prompt formats, styles, and techniques to optimize LLM performance and output quality. Analyze LLM-generated responses, identify biases or limitations, and implement mitigation strategies to enhance customer satisfaction and service performance. Operational Excellence: Identify and propose innovative projects that leverage LLMs to solve complex problems or explore new capabilities in customer service. Conduct research, experiment with cutting-edge techniques, and develop prototypes for potential products or services. Collaborate with data scientists, machine learning engineers, and other stakeholders to integrate prompts and research findings into the broader AI/ML pipeline. Performance Analysis: Analyze the content of customer interactions to generate self-service rate analysis reports, explaining why customers transfer to agents and what can be done to improve these indicators. Generate regular chat-bot performance reports and continuously improve the algorithm model accuracy, customer satisfaction, and service performance of the chat-bot. Team Collaboration: Work closely with relevant function teams to deliver an integrated workflow of demand negotiations, standard creations, and project implementations. Share knowledge and best practices with the team, mentor junior AI operation peer, and contribute to a collaborative learning environment.

Key Responsibilities 1. Design and build batch/real-time data warehouses to support overseas e-commerce growth 2. Develop efficient ETL pipelines to optimize data processing performance and ensure data quality/stability 3. Build unified data middleware layer to reduce business data development costs and improve service reusability 4. Collaborate with business teams to identify core metrics and data requirements, delivering actionable data solutions 5. Discover data insights through collaboration with business owner 6. Participate in AI-driven efficiency enhancement initiatives, collaborating on machine learning algorithm development, feature engineering, and data processing workflows

Team Introduction: TikTok Content Security Algorithm Research Team The International Content Safety Algorithm Research Team is dedicated to maintaining a safe and trustworthy environment for users of ByteDance's international products. We develop and iterate on machine learning models and information systems to identify risks earlier, respond to incidents faster, and monitor potential threats more effectively. The team also leads the development of foundational large models for products. In the R&D process, we tackle key challenges such as data compliance, model reasoning capability, and multilingual performance optimization. Our goal is to build secure, compliant, and high-performance models that empower various business scenarios across the platform, including content moderation, search, and recommendation. Research Project Background: In recent years, Large Language Models (LLMs) have achieved remarkable progress across various domains of natural language processing (NLP) and artificial intelligence. These models have demonstrated impressive capabilities in tasks such as language generation, question answering, and text translation. However, reasoning remains a key area for further improvement. Current approaches to enhancing reasoning abilities often rely on large amounts of Supervised Fine-Tuning (SFT) data. However, acquiring such high-quality SFT data is expensive and poses a significant barrier to scalable model development and deployment. To address this, OpenAI's o1 series of models have made progress by increasing the length of the Chain-of-Thought (CoT) reasoning process. While this technique has proven effective, how to efficiently scale this approach in practical testing remains an open question. Recent research has explored alternative methods such as Process-based Reward Model (PRM), Reinforcement Learning (RL), and Monte Carlo Tree Search (MCTS) to improve reasoning. However, these approaches still fall short of the general reasoning performance achieved by OpenAI's o1 series of models. Notably, the recent DeepSeek R1 paper suggests that pure RL methods can enable LLM to autonomously develop reasoning skills without relying on the expensive SFT data, revealing the substantial potential of RL in advancing LLM capabilities. 团队介绍： 国际化内容安全算法研究团队致力于为字节跳动国际化产品的用户维护安全可信赖环境，通过开发、迭代机器学习模型和信息系统以更早、更快发掘风险、监控风险、响应紧急事件，团队同时负责产品基座大模型的研发，我们在研发过程中需要解决数据合规、模型推理能力、多语种性能优化等方面的问题，从而为平台上的内容审核、搜索、推荐等多项业务提供安全合规，性能优越的基座模型。 课题介绍: 课题背景： 近年来，大规模语言模型（Large Language Models, LLM）在自然语言处理和人工智能的各个领域都取得了显著的进展。这些模型展示了强大的能力，例如在生成语言、回答问题、翻译文本等任务上表现优异。然而，LLM 的推理能力仍有很大的提升空间。在现有的研究中，通常依赖于大量的监督微调（Supervised Fine-Tuning, SFT）数据来增强模型的推理性能。然而，高质量 SFT 数据的获取成本高昂，这对模型的开发和应用带来了极大的限制。 为了提升推理能力，OpenAI 的 o1 系列模型通过增加思维链（Chain-of-Thought, CoT）的推理过程长度取得了一定的成功。这种方法虽然有效，但在实际测试时如何高效地进行扩展仍是一个开放的问题。一些研究尝试使用基于过程的奖励模型（Process-based Reward Model, PRM）、强化学习（Reinforcement Learning, RL）以及蒙特卡洛树搜索算法（Monte Carlo Tree Search, MCTS）等方法来解决推理问题，然而这些方法尚未能达到 OpenAI o1 系列模型的通用推理性能水平。最近deepseek r1在论文中提到通过纯强化学习的方法，可以使得 LLM 自主发展推理能力，而无需依赖昂贵的 SFT 数据。这一系列的工作都揭示着强化学习对LLM的巨大潜力。 课题挑战： 1、Reward模型的设计：在强化学习过程中，设计一个合适的reward模型是关键。Reward模型需要准确地反映推理过程的效果，并引导模型逐步提升其推理能力。这不仅要求对不同任务精准设定评估标准，还要确保reward模型能够在训练过程中动态调整，以适应模型性能的变化和提高。 2、稳定的训练过程：在缺乏高质量SFT数据的情况下，如何确保强化学习过程中的稳定训练是一个重大挑战。强化学习过程通常涉及大量的探索和试错，这可能导致训练不稳定甚至模型性能下降。需要开发具有鲁棒性的训练方法，以保证模型在训练过程中的稳定性和效果。 3、如何从数学和代码任务上拓展到自然语言任务上：现有的推理强化方法主要应用在数学和代码这些CoT数据量相对丰富的任务上。然而，自然语言任务的开放性和复杂性更高，如何将成功的RL策略从这些相对简单的任务拓展到自然语言处理任务上，要求对数据处理和RL方法进行深入的研究和创新，以实现跨任务的通用推理能力。 4、推理效率的提升：在保证推理性能的前提下，提升推理效率也是一个重要挑战。推理过程的效率直接影响到模型在实际应用中的可用性和经济性。可以考虑利用知识蒸馏技术，将复杂模型的知识传递给较小的模型，以减少计算资源消耗。另外，使用长思维链（Long Chain-of-Thought, Long-CoT）技术来改进短思维链（Short-CoT）模型，也是一种潜在的方法，以在保证推理质量的同时提升推理速度。