Prime Intellect
Research Engineer - Distributed Training
The whole loop for training a model on your own workload and keeping it improving — compute (on-demand GPUs and clusters), large-scale reinforcement learning…
San FranciscoFull-time
Checked 22 Nov
What you’d do
- Build and speed up the distributed training systems behind pre-training and large RL runs.
- Improve efficiency across compute, memory, networking and scheduling.
- Write low-level optimisations: kernels, communication paths and runtime improvements.
- Help shape the RL training stack, including asynchronous rollouts.
- Contribute to open-source training libraries.
What they look for
- Strong systems engineering for large-scale training or inference.
- Deep PyTorch and distributed training (FSDP, DeepSpeed, Megatron and similar).
- Has worked with data, tensor and pipeline parallelism.
- Knows GPU architecture and how to profile and debug performance.
Helpful: CUDA or Triton kernels, ML compilers, RL rollout systems, multi-node clusters.
Where: San Francisco, or remote. Pay: $150K to $350K including equity.
Know the company before you applyFull company page →
- funding, as the company states it
- Series A
Prime Intellect hires through MenFem. How, and what that means for our review, is on the company page.
