The ModelAdapter "qwen-code-lora" is invalid: spec.replicas: Unsupported value: 2: supported values: "1"

[XiaozanZhang]

🐛 Describe the bug

After using version 0.5.0, the resource YAML for the modelAdapter is as follows, and an error occurs during startup.：The ModelAdapter "qwen-code-lora" is invalid: spec.replicas: Unsupported value: 2: supported values: "1"

Steps to Reproduce

1. deploy version 0.5.0
2. create base model success
3. create lora fail

apiVersion: model.aibrix.ai/v1alpha1
kind: ModelAdapter
metadata:
  name: qwen-code-lora
  namespace: default
  labels:
    model.aibrix.ai/name: "qwen-code-lora"
    model.aibrix.ai/port: "8000"
spec:
  # Base model that this LoRA adapter extends
  baseModel: qwen-7b-instruct-asdfg

  # Pod selector to find pods that can host this adapter
  # Pods must have both labels: model.aibrix.ai/name and adapter.model.aibrix.ai/enabled=true
  podSelector:
    matchLabels:
      model.aibrix.ai/name: qwen-7b-instruct-asdfg
      adapter.model.aibrix.ai/enabled: "true"

  # URL for the LoRA adapter artifact
  # Supported formats: huggingface://, s3://, or absolute local path
  artifactURL: /modelscope/lora/HiklinkAgent/flight_all_0909_seed_100_format_prompt_shuffle2.5/checkpoint-1000

  # Optional: Number of replicas for the adapter (default: 1)
  # The controller will load the adapter on this many pods
  # Uncomment to enable high availability across multiple pods
  replicas: 2

  # Optional: Scheduler to use for pod selection (default: "default")
  # Available schedulers: "default", "least-adapters"
  schedulerName: least-adapters

error log:

The ModelAdapter "qwen-code-lora" is invalid: spec.replicas: Unsupported value: 2: supported values: "1"

Expected behavior

modelAdapter run success

Environment

aibrix version 0.5.0 vllm 0.10.1

[googs1025]

see: https://github.com/vllm-project/aibrix/pull/1670

the api has changed cc @Jeffwan

[XiaozanZhang]

ok i got it, but i find another quetion when base model add aibrix-runtime, aibrix/runtime:v0.5.0

apiVersion: model.aibrix.ai/v1alpha1 kind: ModelAdapter metadata: name: qwen-code-lora namespace: default labels: model.aibrix.ai/name: "qwen-code-lora" model.aibrix.ai/port: "8000" spec: baseModel: qwen-7b-instruct-asdfg

podSelector: matchLabels: model.aibrix.ai/name: qwen-7b-instruct-asdfg adapter.model.aibrix.ai/enabled: "true"

artifactURL: /modelscope/lora/HiklinkAgent/flight_all_0909_seed_100_format_prompt_shuffle2.5/checkpoint-1000 schedulerName: least-adapters

Error info: Warning MaxRetriesExceeded 55s (x13 over 97s) model-adapter-controller Max retries exceeded for pod qwen-7b-instruct-asdfg-5994cc4756-lnl59: load error: failed to load LoRA adapter: {"error":"Failed to load adapter: Unsupported URL scheme: . Supported schemes: s3, gcs, huggingface, http, https"}

why not support /modelscope/lora/HiklinkAgent/flight_all_0909_seed_100_format_prompt_shuffle2.5/checkpoint-1000?