aipythonpytorchcomputer visionreinforcement learningtransformerdistributed trainingdeepspeedmegatron lmvision language models
Вакансия из Hirify Global, списка международных tech-компаний Для мэтча и отклика нужен Plus
Пожаловаться
86
Крутая вакансия
развернуть
Чётко описанная исследовательская роль в Company hidden предполагает работу с передовыми vision-language моделями, обучением на больших масштабах и reinforcement learning, но высокий научный порог и офисный формат в Сингапуре ограничивают доступность вакансии.
?Высокий научный порог
Передовой стек
Передовой AI
Понятная зона ответственности
Мэтч & Сопровод
Для мэтча с этой вакансией нужен Plus
Подключить Plus
Описание вакансии
Текст:
Hirify AI/Оригинал
TL;DR
Algorithm Researcher (Agentic Vision) (Vision-Language Models/AI): Developing post-training systems and visual agent capabilities for vision-language foundation models with an accent on reinforcement learning alignment, perception, spatial awareness, and hallucination reduction. Focus on building hallucination detection and data flywheel pipelines, optimizing large-scale distributed training, and improving multi-turn image-text reasoning and instruction following.
Location: Singapore-CapitaSky; onsite
Company
hirify.global's Technology Engineering Group supports technology platforms, operational infrastructure, R&D management, data centers, and infrastructure innovation across the company.
What you will do
Own full-stack algorithm R&D for post-training vision-language models using supervised fine-tuning and reinforcement learning alignment methods such as RLHF, PPO, and GRPO.
Optimize low-level visual perception, fine-grained understanding, spatial awareness, and visual agent capabilities for physical-world reasoning and execution.
Investigate multimodal model hallucinations, build detection pipelines, develop data-cleaning strategies, and research reinforcement learning-based hallucination removal.
Align models for real-world user scenarios, improving logical reasoning, instruction following, detail perception, and multi-turn image-text interaction.
Build online badcase monitoring and data flywheel feedback loops for continuous model refinement.
Requirements
Master's degree or above in Computer Science, Software Engineering, Artificial Intelligence, or a related field.
Deep understanding of Transformer architecture and large language model training principles, with extensive research and hands-on experience in at least one post-training area such as LLM alignment, RLHF, reward modeling, personalized LLMs, or memory/RAG mechanisms.
Practical experience with user profiling, recommendation systems integrated with large models, or long-context and long-term memory model optimization.
Strong algorithmic foundations and engineering implementation skills; proficiency in Python and familiarity with PyTorch and TensorFlow.
Hands-on distributed training experience and familiarity with large-scale training and inference frameworks including Megatron-LM, DeepSpeed, and vLLM.
Experience training or fine-tuning models with tens of billions to hundreds of billions of parameters.
Strong research ability demonstrated through publications at top-tier conferences such as NeurIPS, ICLR, ICML, ACL, or EMNLP, or significant contributions to high-impact open-source projects such as Hugging Face.
Technical passion, self-drive, complex problem-solving ability, teamwork, and communication skills.
Culture & Benefits
Full-time onsite work in Singapore.
Equal employment opportunity and an environment focused on supporting employees and diverse perspectives.
Work within a large technology engineering organization supporting infrastructure R&D, platforms, and data centers.
Будьте осторожны
Если работодатель просит войти в их систему, используя iCloud/Google, прислать код/пароль, запустить код/ПО, не делайте этого. Это мошенники. Обязательно жмите «Пожаловаться» или пишите в поддержку. Подробнее в гайде →