Aleph Alpha
Aleph Alpha
51 – 200 Darbuotojai
B2BDirbtinis intelektas
„Aleph Alpha“ yra dirbtinio intelekto tyrimų ir taikymo įmonė, kurianti didelės apimties modelius kalbai, vaizdų duomenims ir strateginiam samprotavimui. Vokietijos Heidelberge įsikūrusi įmonė daugiausia dėmesio skiria pažangių dirbtinio intelekto tyrimų pavertimui praktinėmis sistemomis ir kartu remia Europos skaitmeninį suverenitetą. Jos veikla tenkina verslo ir organizacijų naudojimo atvejus B2B technologijų sektoriuje.

Senior AI Researcher, Foundation Model Pre-Training — Aleph Alpha, Heidelberg

Lead architecture and large-scale pre-training work for Aleph Alpha’s European foundation models, shaping training methods across thousands of GPUs. Build and refine PyTorch training recipes with the Heidelberg-based hybrid team.

Aprašymas

  • Develop architectures and training methods for Aleph Alpha’s next generation of foundation models.
  • Lead core training recipe components, including optimizers, schedules, and initialization.
  • Create PyTorch architecture improvements that strengthen convergence, stability, and training efficiency.
  • Establish hyperparameter scaling laws and methods for scaling training runs.
  • Run small proxy experiments to forecast behavior across thousands of GPUs and reduce training risk.
  • Diagnose convergence problems, including loss spikes and divergence.
  • Troubleshoot distributed training failures such as communication bottlenecks, race conditions, and synchronization errors.
  • Work with Compute Performance, Data, Evaluation, and Post-Training teams on system and model co-design.
  • Turn mathematical analysis and experimental evidence into sound training decisions.
  • Deliver production-quality code that improves model quality, run reliability, and efficiency.

Reikalavimai

  • Proficiency in Python and strong familiarity with PyTorch training workflows.
  • A proven record in machine learning research and software engineering, shown through shipped models, influential open-source work, or published research.
  • Strong mathematical grounding and ability to reason formally about optimization, scaling behavior, and training dynamics.
  • In-depth knowledge of transformer training dynamics, optimization, and large distributed training runs.
  • Ability to design rigorous experiments and use their findings to guide robust training decisions.
  • Practical experience pre-training large models, such as models with 7B or more parameters, on substantial infrastructure, such as clusters of 100 or more GPUs.
  • Sound software engineering practices, including maintainable, well-tested code and reproducible experiments.
  • Ability to implement complex architectures efficiently and reliably, and debug model code, training behavior, and distributed systems.
  • Clear communication and effective cross-team collaboration.
  • Eligibility to work in Germany and availability for regular on-site collaboration in Heidelberg.
  • Preferred: experience training LLMs or multimodal models on large GPU clusters with distributed frameworks such as Megatron-LM, DeepSpeed, or torchtitan.
  • Preferred: familiarity with scaling laws, hyperparameter transfer, or forecasting large-scale training behavior from proxy runs.
  • Preferred: experience profiling distributed jobs and diagnosing training anomalies.
  • Preferred: exposure to sparse training methods such as Mixture-of-Experts.
  • Preferred: top-tier publications, influential open-source contributions, or significant shipped technical work.

Privalumai

  • 30 days of paid annual leave.
  • Access to fitness and wellness offerings through Wellhub.
  • Mental health support via nilo.health.
  • Substantially subsidized company pension plan.
  • Subsidized nationwide transportation ticket in Germany.
  • Budget for additional technical equipment.
  • Flexible working hours.
  • Hybrid work arrangement.
  • JobRad® bicycle leasing.

Susiję darbai

Ovoko \ RRR.LT

Senior Data Platform Engineer

Ovoko \ RRR.LT

Build the data infrastructure behind Ovoko’s used-car-parts marketplace, supporting analytics, AI, and product development. Work across Kubernetes, AWS, and Snowflake to keep data platforms reliable and scalable.

Atidaryti
Mouser Electronics

Customer Service Representative – Milan Hybrid

Mouser Electronics
1 001 – 5 000 Darbuotojai
KonsultavimasLogistikaRinkodara

Handle Mouser Electronics customer orders, quotes, returns and product inquiries while helping grow sales and maintain customer satisfaction. Join the Milan-area team in a hybrid role.

Atidaryti
Aztec Group

Senior Enterprise Data Architect

Aztec Group

Design Aztec Group’s Databricks Lakehouse and Azure data platforms, setting enterprise architecture, master data management, and governance standards. Build secure, AI-ready data foundations for outsourced services.

Atidaryti
Teleperformance

Customer Service Specialist, Aviva Travel Insurance

Teleperformance

Work remotely in the United Kingdom for Teleperformance, processing Aviva travel insurance policies and upgrades. Manage FCA-regulated complaints, medical queries, underwriting risks, and support for vulnerable customers.

Atidaryti