-
Spike
-
Resolution: Unresolved
-
Undefined
-
None
-
RHELAI 1.4 GA
-
None
-
False
-
-
False
-
-
Intel's vllm-fork v0.5.3 only supports the Ray distributed executor for multi-card inference. Having merged into upstream vLLM, the new 0.6.2 release may have relaxed this requirement, unblocking multi-card inference for RHEL AI.
Investigate the status of this dependency relaxation in upstream vLLM for Gaudi inference.