ML Engineer (Training Infra), Foundational Models

SarvamIndiranagar, BangalorePosted 25 days ago

Mid level
Full time
3+ years

A free account shows the apply link for every role and matches you to the ones that fit your resume.

About the role

Sarvam is building a sovereign AI platform for India, including foundational models and infrastructure. You will build and optimize distributed training infrastructure and ensure large-scale model training runs perform reliably.

What you'll do

Build and maintain the distributed training stack across GPU clusters
Design parallelism strategies for different model architectures and scales
Profile and optimize training throughput, kernels, communication, and data loading
Write and tune custom GPU kernels
Ensure long-running training jobs recover reliably and detect faults
Partner with researchers and data teams to improve training efficiency

Skills

Must have

Distributed training
Distributed systems
GPU architecture
CUDA
PyTorch
GPU profiling
Open-source contributions

Nice to have

Triton
Kernel optimization
Mixed precision
Quantization-aware training
Large-scale model training

Tools

Megatron-lm
Deepspeed
Fsdp
Nemo
Nsight
PyTorch profiler
Nccl
Vllm
Slurm
Kubernetes

Who it's for

Education: Engineering degree
Degree: computer science
Experience: 3+ years
Domain: ai infrastructure

About Sarvam

Sarvam has 28 open roles in Bangalore, at 732, Chinmaya Mission Hospital Road, Indiranagar Stage 1, Bengaluru, Karnataka 560038. See every Sarvam role.