Update LLM Inference recipes to latest vLLM

[amgowda-oci]

vLLM version being used here is 0.6.3 while today we are at 0.9.1. Can you please update the vllm version, so we get the latest and better performance. Better yet please migrate to vllm:latest upstream container.

[dkennetzoracle]

This is largely complete @amgowda-oci - we don't want to use latest because we don't currently have the ability to automatically test this (which we're working on). I have v0.9.1 in most recipes, because there were performance regressions with B200s in v0.9.2 (which I did test). I'll mark recipe updates still needed. If there are recipes in the portal that are displaying older versions, please assign @grantneumanoracle

• https://github.com/oracle-quickstart/oci-ai-blueprints/blob/main/docs/sample_blueprints/llama-stack/llama_stack_basic.json
• https://github.com/oracle-quickstart/oci-ai-blueprints/blob/main/docs/sample_blueprints/multi-node-inference/multinode_inference_BM_A10.json
• https://github.com/oracle-quickstart/oci-ai-blueprints/blob/main/docs/sample_blueprints/multi-node-inference/multinode_inference_VM_A10.json
• https://github.com/oracle-quickstart/oci-ai-blueprints/blob/main/docs/sample_blueprints/using_rdma_enabled_node_pools/rdma_distributed_inference.json