AI Back-end Engineer – PyTorch Internals
IBM · Bengaluru
| Company | IBM |
|---|---|
| Role | AI Back-end Engineer – PyTorch Internals |
| Location | Bengaluru |
| Work mode | Hybrid |
| Experience | 3–6 years |
| Job type | Full-time |
About this job
IBM is hiring an AI Back-end Engineer in Bengaluru to work on PyTorch internals: the compiler and runtime pieces that make AI models train and run fast.
You will design, code, debug and test PyTorch components such as TorchInductor, TorchDynamo and AOTAutograd, fix low-level runtime issues (autograd, dispatcher, kernels, graph breaks) and work with distributed training (FSDP). The posting asks for 5 to 10 years on Unix/Linux products in Python, C or C++.
Eligibility
- 5 to 10 years on Unix/Linux products in Python, C or C++
- Hands-on with PyTorch internals (TorchInductor, TorchDynamo, AOTAutograd)
- Debugging low-level runtime issues
- PyTorch FSDP experience
Skills required
What you will do
- Extend PyTorch internals and back-end architecture
- Debug autograd, dispatcher and kernel issues
- Optimise model training performance
- Work in an agile team with DevOps practices
How to apply
- Open the official job page using the button below and read the full description
- Tailor your resume to the skills listed here and apply directly on the company's careers site
Before the interview, prepare:
- How autograd works internally
- torch.compile: TorchDynamo and TorchInductor
- FSDP vs DDP for distributed training
- Debugging a graph break or performance drop
Get jobs like this every day on WhatsApp. Free, no spam, only verified openings.
Join channelStay safe: DailyITJobs is not a recruitment agency and is not connected to IBM. Apply only on the official website. Never pay anyone for a job, interview or offer letter.