Package
nemo-rl-cuda-13.0-vllm
Component
vllm
Latest update
6.5
CVSS V3
Build, ship, and run secure software with minimal, hardened container images — rebuilt from source daily and guarded under our industry-leading remediation SLA.
Start for freeStatus
Impact
This package ships vLLM 0.20.0, the release NeMo RL 0.7.0 pins and patches against. The fix for this issue first appears in vLLM 0.32.0.
Upgrading within NeMo RL 0.7.0 is not possible: vLLM 0.22 and later move flashinfer to a version Megatron Core does not share, and NeMo RL carries patches written against vLLM 0.20's internals. Rebuilding does not help either — the vulnerable code is vLLM's Python source, which is identical in any rebuild of the 0.20.0 release.
This will be resolved when NVIDIA publishes a NeMo RL release built against vLLM 0.32.0 or later.
Status