Deep Learning System Engineer (Semiconductor)

Saaki Argus and Averil Consulting; Bengaluru Urban; Vor Ort; INR 4.000.000 - 7.000.000; Vollzeit

Zusammenfassung

Saaki Argus and Averil Consulting seeks an experienced Deep Learning System Engineer (Semiconductor) to optimise and run advanced AI models on a specialised many-core MIMD processor.

You will develop high-performance AI operations, benchmark kernels, and integrate with PyTorch, vLLM, and Hugging Face, driving model speed, accuracy, and efficiency in cutting-edge hardware environments.

Qualifikationen

• 8–15 years of experience in deep learning systems engineering.
• Hands-on experience developing and optimising custom AI kernels at hardware/low-level software interface.
• Strong programming skills in C/C++, Python, and CUDA.
• In-depth understanding of Transformers, CNNs, GNNs, and diffusion models.

Aufgaben

• Analyse and optimise AI models for execution on MIMD accelerator.
• Develop and optimise core AI operations (matrix multiply, convolution, attention, etc.).
• Implement and benchmark custom AI kernels with FP16, FP32, BF16 precision.
• Integrate optimized operations with PyTorch, vLLM, and Hugging Face Transformers.

Kenntnisse

C/C++

Python

CUDA

PyTorch internals

Mixed-precision computing

Performance profiling

Transformers

CNNs

Ausbildung

B.Tech/M.Tech

M.S.

Ph.D.

Tools

PyTorch

CUDA

cuDNN

Hugging Face Transformers

vLLM

Jobbeschreibung

• Role: Deep Learning System Engineer(Semiconductor)
• Exp: 8 to 15 Years

Role Overview

• We are looking for an experienced AI Engineer to optimise and run deep learning models on a specialised many-core MIMD processor. The role involves developing high-performance AI operations, improving model execution speed, and ensuring accuracy and efficiency without relying on conventional GPU tensor cores.
• Key Responsibilities
• Analyse and optimise AI models, including Transformers, CNNs, GNNs, recommendation systems, and diffusion models, for execution on the MIMD accelerator.
• Develop and optimise core AI operations such as matrix multiplication, convolution, attention, normalisation, activation, and reduction.
• Implement and benchmark custom AI kernels using FP16, FP32, and BF16 precision.
• Compare performance against established CUDA and cuDNN implementations on comparable hardware.

2. Framework Integration and Performance Validation

• Integrate optimised AI operations with frameworks such as PyTorch, vLLM, and Hugging Face Transformers.
• Validate model accuracy, numerical stability, and compatibility across different precision formats.
• Measure and optimise training and inference performance, including throughput, latency, memory utilisation, and power consumption.
• Collaborate with hardware and platform teams to identify performance bottlenecks and implement improvements.

3. Model Validation and Customer Support

• Validate and maintain a portfolio of AI models, including Llama, BERT, GPT-class models, ResNet, Stable Diffusion, and GNNs, on the company's processor.
• Develop performance benchmarks, technical documentation, and model deployment guidelines.
• Support early-access customers with model porting, performance optimisation, debugging, and troubleshooting.
• Required Qualifications
• B.Tech/M.Tech, M.S., or Ph.D. in Computer Science, Electrical Engineering, Machine Learning, or a related field.
• 8–15 years of experience in deep learning systems engineering, AI accelerator software, or ML infrastructure.
• Hands‑on experience developing and optimising custom AI kernels at the hardware or low-level software interface.
• Strong programming skills in C/C++, Python, and CUDA.
• In‑depth understanding of deep learning architectures, including Transformers, CNNs, GNNs, and diffusion models.
• Experience optimising matrix multiplication, attention, convolution, and other computationally intensive AI operations.
• Strong knowledge of PyTorch internals, operator dispatch, custom backends, and mixed-precision computing.
• Understanding of processor architecture, memory hierarchy, compilers, operating systems, and hardware‑software interactions.
• Experience with AI performance profiling, benchmarking, numerical stability, and quantization.