I

Institute of Foundation Models

Researches and develops foundation models

Inference Optimization Intern - Performance Modeling

Summer 2026Posted on 6/25/2026
No salary listed
Internship
Bachelor's
Sunnyvale, CA, USA
In PersonOn-site in Sunnyvale, California.
AI & Machine Learning

About the job

Requirements
  • Currently pursuing a degree in Computer Science, Computer Engineering, Electrical Engineering, Artificial Intelligence, High-Performance Computing, or a related quantitative discipline.
Responsibilities
  • Develop analytical performance models for GPU kernels and inference workloads.
  • Build and validate a simulator to estimate theoretical hardware performance limits.
  • Compare measured kernel performance against architectural peak throughput.
  • Identify performance bottlenecks in compute, memory, communication, and scheduling.
  • Analyze GPU execution using NVIDIA Nsight Systems and Nsight Compute.
  • Investigate PTX and SASS code generation to understand low-level execution behavior.
  • Collaborate with researchers and engineers to optimize inference kernels for transformer-based models.
  • Evaluate utilization of Tensor Cores, memory bandwidth, caches, and instruction pipelines.
  • Design profiling methodologies for Hopper and Blackwell architectures.
  • Document findings and provide actionable recommendations for performance improvements.
Desired Qualifications
  • Experience with CUDA programming and GPU kernel development.
  • Understanding of NVIDIA GPU architecture and memory hierarchy.
  • Familiarity with performance profiling tools such as Nsight Systems and Nsight Compute.
  • Knowledge of PTX, SASS, and low-level GPU execution.
  • Experience optimizing CUDA kernels for throughput and latency.
  • Understanding of roofline analysis, performance modeling, and hardware utilization metrics.
  • Experience with deep learning frameworks such as PyTorch or TensorFlow.
  • Strong programming skills in C++, CUDA, and Python.
  • Performance engineering mindset.
  • Strong analytical and debugging abilities.
  • Interest in AI systems, inference optimization, and hardware-software co-design.
  • Ability to work independently on research and engineering challenges.
  • Excellent written and verbal communication skills.

About the company

I

Institute of Foundation Models

View
Website

IFM builds and studies foundation models—large AI models designed to learn broadly from diverse data and be used across many tasks. Its work centers on academic research to create open, fast, and practical models that address real-world societal needs, rather than narrow applications. The models are developed by a global team across Abu Dhabi, Paris, and Silicon Valley, with a strong emphasis on openness and collaboration to advance AI science and accessibility. Unlike typical private labs that lock models behind paywalls, IFM aims to provide publicly accessible, efficient models that can be used by researchers and developers to solve real problems. The overarching goal is to push the science of foundation models forward while ensuring their benefits reach society at large.

Company Size

N/A

Company Stage

N/A

Total Funding

N/A

Headquarters

United Arab Emirates

Founded

N/A

Get referred to Institute of Foundation Models

See people who can refer or advise you

Simplify Jobs

Simplify's Take

What believers are saying

  • AWS, Cerebras, Nebius, Hugging Face, vLLM, and SGLang already distribute K2 Horizon.
  • October 2026 benchmark chatter praises K2 Horizon's small models and 375B enterprise flagship.
  • MBZUAI's October 2026 hiring surge supports rapid expansion across agents, world models, and infrastructure.

What critics are saying

  • Open-sourcing K2 Horizon hands competitors reproducible recipes, shrinking IFM's moat by 2027.
  • MBZUAI listed 45 IFM vacancies on October 2, 2026, signaling heavy execution strain.
  • If K2 Horizon misses commercial adoption, IFM remains a prestige lab dependent on state capital.

What makes Institute of Foundation Models unique

  • September 3, 2026 K2 Horizon shipped six fully open models with weights, code, and data.
  • IFM offers Apache 2.0 frontier models, unlike closed labs protecting training recipes and checkpoints.
  • MBZUAI-backed labs in Abu Dhabi, Silicon Valley, and Paris combine research depth with deployment access.

Help us improve and share your feedback! Did you find this helpful?

Benefits

  • Health Insurance
  • Dental Insurance
  • Vision Insurance
  • Paid Vacation
  • Paid Holidays
  • Parental Leave
  • Employee Assistance Program
  • Life Insurance
  • Disability Insurance
  • 401(k) Plan
  • Wellness Program
  • Flexible Work Hours

Company News

PR Newswire
Oct 15th, 2026
Jais 3 launches with six open Arabic models, up to 375B parameters, plus dialect-aware speech recognition

The Institute of Foundation Models (IFM), part of Mohamed bin Zayed University of Artificial Intelligence, and Inception42 launched Jais 3, a fleet of Arabic-native AI models, on October 15, 2026. It pairs six open language models, from 900 million to 375 billion parameters, with Jais Speech, a speech recognition model for Arabic and its dialects. The models are built on K2 Horizon, IFM's open-source fleet released in September. They handle Modern Standard Arabic and regional dialects, and can plan tasks, search the web, and produce Arabic documents and presentations. In the companies' tests on 24 Arabic benchmarks, the largest model matched the best open models on Arabic understanding and led on agentic tasks. Jais Speech recorded the lowest error rates among models compared. Jais 3 is available on Hugging Face. IFM is also releasing its benchmarks and code.

PR Newswire
Sep 3rd, 2026
Institute of Foundation Models launches K2 Horizon: largest fully open-source AI model fleet with weights, code, and training data

The Institute of Foundation Models has launched K2 Horizon, a fleet of six fully open-source AI models ranging from 0.9 billion to 375 billion parameters. The models include weights, code, training data, and methodologies, allowing researchers and developers to inspect and reproduce the work. The 0.9B, 3.7B, and 7B models set new benchmarks at their respective scales. The smallest model can run on watches and glasses, whilst the 3.7B and 7B models work on phones. The dense 32B model and sparse 36B model suit local hosting, and the 375B flagship model targets enterprise deployments. Launched by Mohamed bin Zayed University of Artificial Intelligence in May 2025, IFM now operates across Silicon Valley, Paris, and Abu Dhabi. The models are available through Hugging Face, vLLM, and SGLang, with API access through partners including AWS and Cerebras. All models use the Apache 2.0 licence.

INACTIVE