Goals:

• We need to provide CPU-only builds for all CPU architectures of PyTorch and vllm to be able to go into a RHAIIS image.

Requirements:

• Torch 2.9.0
  • x86_64
  • aarch64
  • Power
  • Z
• vLLM (same latest version as CUDA, ROCm)
  • x86_64 (AVX2, priority)
  • aarch64 (optional)
  • Power (optional)
  • Z (optional)

Done - Acceptance Criteria:

• vLLM is provided for x86_64 CPU with AVX2. AVX512 support will be added in the future
• vLLM support for aarch64, Power, and Z are stretch goals.
• Component teams can install vllm and torch into the RHAIIS image using AIPCC base images without hardware accelerator support. (Implies work in the RHAIIS wheel pipeline repo)
• Partners can access the same builds using our public package indexes. (Implies work in the RHAI wheel pipeline repo)