AI Compilation Intern

Kalray; Montbonnot-Saint-Martin; Hybrid; EUR 6,700 - 13,000; Full time

Benefits offered by this job

Hybrid work

Meal vouchers

Team building activities

Job summary

Kalray invites a proactive intern to join the AI & Compute team in Montbonnot-Saint-Martin to accelerate LLM inference on a RISC-V CPU stack. You will work on an open-source MLIR/LLVM-based compiler and explore auto-tuning integration to optimize code generation.

The 6‑month internship offers hybrid work with meals, team building and other activities; compensation depends on your study level. You will develop a PoC to connect an auto-tuner to the compiler and assess gains against standard

Qualifications

• Interest in performance optimization on hardware.

Responsibilities

• Become familiar with our compiler on CPU and study its transformation passes.
• Study open-source auto-tuning solutions and select an appropriate approach.
• Develop a PoC connecting an auto-tuner to our compiler.
• Evaluate the gains compared to default configurations and the cost of the search.

Skills

Interest in performance optimization

Job description

Within the AI & Compute team, you will contribute to the development of a software stack designed to accelerate the inference of Large Language Models (LLMs) on RISC-V CPUs.

As part of this stack, we have an open-source compiler based on MLIR/LLVM that enables neural network models to be compiled from various representations (e.g. PyTorch) into optimized machine code. During this compilation process, various heuristics are used, notably to select the transformations to be performed and define their parameters.

Auto-tuning consists in automatically exploring different compilation configurations and measuring the performance of the generated code in order to identify those best suited to a given workload and hardware. Our compiler provides mechanisms to control certain code generation decisions, but we have not yet explored integration with auto-tuning frameworks (such as ytopt or OpenTuner).

As such, your main responsibilities will be the following:

• Become familiar with our compiler on CPU and study its transformation passes.
• Study open-source auto-tuning solutions and select an appropriate approach.
• Develop a PoC connecting an auto-tuner to our compiler.
• Evaluate the gains compared to default configurations and the cost of the search.

During your internship with us, you will join a dynamic team with diverse and varied skills, ranging from microkernel optimization to the deployment of LLMs on distributed machines.

REQUIRED AND DEVELOPABLE SKILLS

Required:

• Interest in performance optimization on hardware;

To be developed:

• Proficiency in compilation techniques and intermediate representations;
• Auto-tuning techniques;
• Program optimization on CPUs.

PROFILE

• You are curious, ambitious and passionate about high-tech and applications resulting from new technologies;
• You enjoy working as part of a team and taking initiatives;
• You are enthusiastic about being part of a human adventure and working in an agile environment
• You are comfortable in English, particularly in a technical context.

Are you passionate about high-tech and cutting-edge technologies? Do you want to be part of a human adventure and develop in an agile, dynamic and collaborative environment? This internship is for you.

CONTRACT

• Contract type: Internship agreement
• Duration: 6 months
• Compensation: Depending on level of study
• Collective agreement: Metallurgy
• Benefits: Hybrid work, meal vouchers, team building and other activities such as laser tag, escape games, sailing, skiing, go-karting, karaoke and much more!
• Location: Montbonnot(38)

ABOUT KALRAY

Founded in 2008 as a spin-off from CEA, Kalray is a leading European semiconductor company, backed by major industrial investors such as Safran, NXP Semiconductors and Bpifrance. With more than 30 patent families and more than €100 million invested in R&D, our technology is paving the way for a new era in data processing.

Thanks to their patented “manycore” architecture, Kalray’s MPPA® intelligent processors (known as DPU or xPU) are capable of managing multiple data streams in parallel – and without bottlenecks – enabling applications that process massive amounts of data to become smarter, more efficient and more energy-efficient, complementing traditional approaches (CPU and GPU).

Kalray’s offering includes hardware and software acceleration solutions, as well as a co-development – or design-ondemand – offering for processors and acceleration solutions optimized for specific uses in high-growth sectors such as data centers, AI gigafactories, 5G networks, Industry 4.0, space and defense, and many others.

Above all, Kalray is a team of passionate, enthusiastic and committed people who place excellence and collaboration at the heart of their work. Whether you are an intern or an apprentice, we will do everything we can to help you fully express your talent, develop your potential and contribute to the company’s success.

Our offices are located near Grenoble, in the French Alps, and we also have additional offices near Nice, in Sophia Antipolis, as well as in Romania.

We’re always looking for people with unique backgrounds and skill sets. If you don’t see a role that works for you, but you believe you have something to offer, drop us a line and tell us your story.