[Feat] Support Additional Pre-deployed Inference Stacks

[wangchen615]

Description

Enhance fmperf to support various pre-deployed distributed inference stacks, enabling users to benchmark different production-grade deployment solutions.

Target Stacks

• Dynamo AIBrix
• vLLM Production Stack
• LLM-D
• Other distributed inference solutions

Implementation Requirements

1. Stack Configuration

   • Define stack specifications for each deployment type
   • Support for distributed deployment configurations
   • Handle different service discovery mechanisms

2. Deployment Management

   • Integration with existing deployment orchestration
   • Support for multi-node deployments
   • Handle different scaling configurations

3. Monitoring & Metrics

   • Handle logging of existing stacks

Expected Benefits

• Enable benchmarking of production-grade distributed deployments
• Support for more realistic deployment scenarios
• Better comparison between different deployment solutions

Related Components

• fmperf/Cluster.py
• fmperf/StackSpec.py
• Deployment configuration files