mirror of
https://github.com/hanzo-ml/kserve.git
synced 2026-10-30 20:53:49 +00:00
d285abaad4
Signed-off-by: Andres Llausas <[email protected]> Co-authored-by: Claude Sonnet 4.6 <[email protected]>
62 lines
1.7 KiB
YAML
62 lines
1.7 KiB
YAML
apiVersion: serving.kserve.io/v1alpha1
|
|
kind: LLMInferenceService
|
|
metadata:
|
|
name: lora-multi-hf-test
|
|
namespace: kserve-ci-e2e-test
|
|
spec:
|
|
model:
|
|
uri: hf://facebook/opt-125m
|
|
name: facebook/opt-125m
|
|
lora:
|
|
adapters:
|
|
# Multiple HuggingFace LoRA adapters
|
|
# Using the same adapter with different names to test multi-adapter functionality
|
|
- name: adapter-1
|
|
uri: hf://edbeeching/opt-125m-lora
|
|
- name: adapter-2
|
|
uri: hf://edbeeching/opt-125m-lora
|
|
# Optional: override vLLM LoRA runtime settings.
|
|
# maxRank defaults to 64, maxAdapters and maxCpuAdapters default to the number of configured adapters.
|
|
maxRank: 64
|
|
maxAdapters: 2
|
|
maxCpuAdapters: 4
|
|
replicas: 1
|
|
router:
|
|
scheduler: {}
|
|
route: {}
|
|
gateway: {}
|
|
template:
|
|
containers:
|
|
- name: main
|
|
image: public.ecr.aws/q9t5s3a7/vllm-cpu-release-repo:v0.17.1
|
|
env:
|
|
- name: VLLM_LOGGING_LEVEL
|
|
value: DEBUG
|
|
- name: VLLM_CPU_KVCACHE_SPACE
|
|
value: "1"
|
|
resources:
|
|
limits:
|
|
cpu: '2'
|
|
memory: 7Gi
|
|
requests:
|
|
cpu: '200m'
|
|
memory: 2Gi
|
|
livenessProbe:
|
|
httpGet:
|
|
path: /health
|
|
port: 8000
|
|
scheme: HTTPS
|
|
initialDelaySeconds: 180
|
|
periodSeconds: 30
|
|
timeoutSeconds: 30
|
|
failureThreshold: 8
|
|
readinessProbe:
|
|
httpGet:
|
|
path: /health
|
|
port: 8000
|
|
scheme: HTTPS
|
|
initialDelaySeconds: 30
|
|
periodSeconds: 10
|
|
timeoutSeconds: 5
|
|
failureThreshold: 3
|