The champion candidates would beryllium apical 1% astatine aggregate parts of the conclusion stack.
work connected PD disaggregation research
Morph builds the conclusion infrastructure down the fastest unfastened models. Our stack spans kernels, exemplary serving, routing, autoscaling, and capacity. We are hiring a capacity technologist to make the full strategy faster, cheaper, and much reliable.
What you’ll do
- Find the spread betwixt theoretical hardware capacity and accumulation performance
- Trace latency and throughput regressions from the API furniture down to individual kernels
- Optimize batching, scheduling, routing, quantization, and distributed execution
- Build benchmarks and observability that make bottlenecks obvious
- Validate that each optimization preserves exemplary value and correctness
- Stack-rank opportunities and vessel the highest-impact fixes yourself
You mightiness beryllium a fresh if you
- Have optimized analyzable accumulation systems
- Understand GPU performance, representation bandwidth, collectives, and conclusion serving
- Are beardown successful Python and comfortable navigating unfamiliar codebases
- Can move profiling information into clear engineering decisions
- Care astir tokens per second, tokens per dollar, and correctness equally
You will activity straight pinch the founders connected problems that find really efficiently frontier-scale models tin beryllium served. Small team, tremendous compute, contiguous accumulation impact.
English (US) · 
Indonesian (ID) ·