Research Engineer - Inference (Fully Remote)

Prime Intellect, Inc.; United States; À distance; USD 180 000 - 240 000; Plein temps

Avantages offerts par ce poste

Equity incentives

Visa sponsorship

Relocation assistance

Flexible remote or in-person in SF

Résumé du poste

Prime Intellect in San Francisco is seeking a developer to build and optimize GPU inference systems powering RL research, scaling toward 1–3T parameter models on NVL72 clusters. The role aims to improve rollout throughput and enable frequent model updates.

You will extend vLLM integration for new model architectures, optimize multi-node MoE inference and transfer, and reduce trainer-to-inference weight update overhead.

Qualifications

• Experience building or optimizing GPU inference systems with understanding of inference-engine internals.
• Strong Python skills and experience with PyTorch or comparable ML frameworks.
• Ability to profile end-to-end workloads and turn measurements into verified performance improvements.

Responsabilités

• Extend prime-rl’s vLLM integration for new model architectures with small-scale validation before larger runs.
• Optimize multi-node MoE inference, expert parallelism, and prefill/decode separation.
• Reduce trainer-to-inference weight-update overhead through sharding-aware transfer and communication overlap.
• Collaborate with researchers to implement deployment strategies.

Connaissances

GPU inference systems

Python

PyTorch

Performance profiling

Description du poste

• Build the inference systems that power Prime Intellect’s RL research, scaling toward 1–3T parameter models on NVL72 clusters.
• Improve rollout throughput and make frequent model updates practical.

Key Responsibilities:

• Extend prime-rl’s vLLM integration for new model architectures with small-scale validation before larger runs.
• Optimize multi-node MoE inference, expert parallelism, and prefill/decode separation.
• Reduce trainer-to-inference weight-update overhead through sharding-aware transfer and communication overlap.

Requirements:

• Experience building or optimizing GPU inference systems with understanding of inference-engine internals.
• Strong Python skills and experience with PyTorch or comparable ML frameworks.
• Ability to profile end-to-end workloads and turn measurements into verified performance improvements.
• Competitive cash compensation with equity incentives.
• Flexible remote or in-person work in San Francisco, plus visa sponsorship and relocation assistance.
• Quarterly off-sites, hackathons, conferences, and learning opportunities.

Prime Intellect

Prime Intellect is building the open superintelligence stack, unifying compute, environments, and deployment for frontier-scale AI post-training. Backed by $150M in funding, the team is mission-driven and hard-working, focused on accelerating open AI research.