LLM提示工程師 / LLM Prompt Engineer
薪資範圍:50,000 ~ 70,000 TWD / month
LLM提示工程師 (LLMPrompt & Testing Engineer)
1. 設計與優化大型語言模型(LLM)的提示工程： 專精於連鎖思維提示(Chain of Thought Prompting)、指令提示(Instruction Prompt)、摘要(Summarization)、分類(Classification)、推理(Reasoning)、內容回吐(Regurgitation)等技術，提升模型的表現與準確性。
2. 開發與執行基準測試與數據分析： 設計並比較LLM性能基準，分析多語言輸出，使用定量與定性指標評估模型行為，並針對模型性能進行改進建議。
3. 測試迭代與優化管理： 主導測試計劃的制定與執行，負責測試數據集的生成、結果分析以及多輪迭代改進，確保模型測試全流程的穩定性與效率。
4. 與資料科學家合作創建新數據： 協同資料科學家生成、標註與清理新數據集，為模型測試與提示工程提供高質量的數據基礎。
5. 多語言能力： 精通中文與英文，熟悉日文者更佳，能根據多語言特性設計提示並分析模型在不同語言環境中的表現。
6. 資料處理與版本管理： 熟悉數據清理、處理、生成、版本控制，能使用Parquet、正則表達式(Regex)等工具優化數據處理流程，並管理數據版本歷史。
7. 提示工程與深度應用： 熟悉連鎖思維提示的設計與實施，精通模型行為推理與指令提示的優化，並具備實際Instruct Tuning經驗。
8. 程式開發與工具集成： 精通Python，熟悉使用版本控制工具(Git)與數據分析框架，能將提示工程與基準測試集成到工作流中。
9. 模型行為與特性分析： 熟練處理分類、摘要生成與推理能力測試，並基於模型表現分析提出改進策略。
10. 分析與改進： 通過基準測試與數據比較，持續改進提示設計與模型性能，並根據業務需求調整提示行為。
+++++主動積極，提出解決方案和改進建議，我們喜歡這樣的團隊成員！+++++
LLM Prompt Engineer Job Description
Job Title: Prompt & Testing Engineer
1. Design and optimize prompt engineering for LLMs: Expertise in Chain of Thought Prompting, Instruction Prompts, Summarization, Classification, Reasoning, and Regurgitation techniques to enhance model performance and accuracy.
2. Develop and execute benchmarks and data analysis: Design and compare LLM performance benchmarks, analyze multilingual outputs, and provide quantitative and qualitative evaluations to refine model behavior.
3. Testing iteration and optimization management: Lead test planning and execution, manage test dataset creation, result analysis, and iterative improvements to ensure stability and efficiency in the testing pipeline.
4. Collaborate with data scientists to create new datasets: Work closely with data scientists to generate, annotate, and clean new datasets, ensuring high-quality data for model testing and prompt engineering.
5. Multilingual proficiency: Fluent in Chinese and English, with Japanese as a plus, capable of designing and analyzing prompts tailored to diverse language characteristics and evaluating model performance in multilingual settings.
6. Data processing and version control: Skilled in data cleaning, processing, generation, and versioning using tools like Parquet and Regex to streamline workflows and manage data history effectively.
7. Prompt engineering applications: Proficient in designing and implementing advanced prompt strategies, including Chain of Thought and Instruction Prompts, with hands-on experience in Instruct Tuning.
8. Programming and workflow integration: Advanced Python expertise, with experience using version control tools (e.g., Git) and integrating benchmark testing and prompt workflows into seamless processes.
9. Model behavior and capability analysis: Adept at testing classification, summarization, and reasoning capabilities and leveraging insights to propose actionable strategies for performance improvement.
10. Analysis and improvement: Continuously refine prompt designs and model outputs through benchmarks and data comparisons, aligning improvements with project objectives and requirements.
+++++Be proactive; we love team members who suggest solutions and bring improvements to the table!+++++
語文條件
•英文(聽:精通、說:精通、讀:精通、寫:精通)
•日文(聽:精通、說:精通、讀:精通、寫:精通)
•中文(聽:精通、說:精通、讀:精通、寫:精通)
擅長工具
•UNIX
•Git
•ASP.NET
•C++
•Python
•Visual Studio .net
•Prompt
•Excel
•Data Architect
•Data Marts
•Hive
•SAS
•SPSS
工作技能
•系統架構規劃
•軟體程式設計
•結構化程式設計
•模組化系統設計
•Machine Learning
•資料庫系統管理維護
•資料庫程式設計
•資料庫軟體應用
其他條件
1.考慮英文+中文能力, 日文加分
2.公司使用A100、H100大量資料訓練資料處理。
3.對於開發個性化的AI解決方案，情感陪伴，包括陪伴型遊戲、互動角色和養成型遊戲。熱愛動漫、角色以及IP行業的候選人優先考慮。
We collaborate with large gaming corporations and anime IPs to develop personalized AI solutions, including companionship games, interactive characters, and nurturing-style games. Candidates with a passion for anime, characters, and the IP industry will be an integral part of our team. Our work focuses on integrating AI into the anime, manga, and gaming industries, creating unforgettable experiences for global users.
公司地址:
民生東路三段109號10樓D區其他:
**應徵者需接受公司測試**-2025-01-14