feat(deploy): add DeepSeek R1 MI300 data-parallel inference preset

[bongwoobak]

우선 preset안에 폴더 만들어서 넣어놨는데 구조적으로 수정 할 부분이 필요하다면 커멘트 남겨주세요

[hhk7734]

moai-inference-preset에는 InferenceServiceTemplate만 들어갈거에요.
아래와 같은 구조로 변경해주세요

base/
  - workertemplate...
deepseek-r1/
  - deepseek-r1...

[hhk7734]

• end-to-end (co-located, prefill + decode)
  • deepseek-r1-mi300x-tp8
  • deepseek-r1-mi300x-dp8ep
• prefill only
  • deepseek-r1-prefill-mi300x-tp8
• decode only
  • deepseek-r1-decode-mi308x-tp8

[Copilot]

Pull request overview

This PR adds DeepSeek R1 inference presets for MI300X GPUs with data-parallel inference support, along with a collection of base vLLM data-parallel templates for the Moreh inference framework.

• Introduces DeepSeek R1 model configuration for both prefill and decode workers with MI300X-specific optimizations
• Adds reusable base templates for vLLM data-parallel deployments including core functionality, metadata labeling, offline HuggingFace hub support, and decode proxy configuration
• Organizes templates in a subdirectory structure (base/ and deepseek-r1/) for better maintainability

Reviewed changes

Copilot reviewed 7 out of 7 changed files in this pull request and generated 7 comments.

Show a summary per file

File	Description
deploy/helm/moai-inference-preset/templates/deepseek-r1/vllm-deepseek-r1-prefill-mi300x-dp8ep.helm.yaml	DeepSeek R1 prefill worker configuration with MI300X-specific environment variables and vLLM settings
deploy/helm/moai-inference-preset/templates/deepseek-r1/vllm-deepseek-r1-decode-mi300x-dp8ep.helm.yaml	DeepSeek R1 decode worker configuration with MI300X-specific environment variables and vLLM settings
deploy/helm/moai-inference-preset/templates/base/vllm-dp.helm.yaml	Base vLLM data-parallel template with complete pod specification including container setup, health checks, and volume configuration
deploy/helm/moai-inference-preset/templates/base/vllm-dp-prefill-meta.helm.yaml	Metadata template for labeling prefill workers
deploy/helm/moai-inference-preset/templates/base/vllm-dp-hf-hub-offline.helm.yaml	Configuration for offline HuggingFace Hub usage with persistent volume mount
deploy/helm/moai-inference-preset/templates/base/vllm-dp-decode-proxy.helm.yaml	Decode worker proxy configuration with initialization container and port offset settings
deploy/helm/moai-inference-preset/templates/base/vllm-dp-decode-meta.helm.yaml	Metadata template for labeling decode workers