What the posting requires
- Required
- Megatron-LM
- GPU
- Verl
- GRPO
- NCCL
- Communication/compute overlap
- NVIDIA Nsight
- Python
- C++
- Preferred
- Distributed training
- Mixed-precision
- Gradient checkpointing
- GPU kernels
- Training throughput bottlenecks
- Custom kernels
These tags are our reading of the posting. They can miss something; the original posting is the reference.
