mirror of
https://github.com/hanzo-ml/kserve.git
synced 2026-10-30 20:53:49 +00:00
f956207415
Signed-off-by: Vivek Karunai Kiri Ragavan <[email protected]> Signed-off-by: Pierangelo Di Pilato <[email protected]> Co-authored-by: Pierangelo Di Pilato <[email protected]>
33 lines
947 B
YAML
33 lines
947 B
YAML
apiVersion: serving.kserve.io/v1alpha1
|
|
kind: LLMInferenceService
|
|
metadata:
|
|
name: qwen2-7b-instruct-amd
|
|
spec:
|
|
model:
|
|
uri: hf://Qwen/Qwen2.5-7B-Instruct
|
|
name: Qwen/Qwen2.5-7B-Instruct
|
|
replicas: 2
|
|
# Same shared label as the NVIDIA instance so that the InferencePool
|
|
# selector matches these pods too.
|
|
labels:
|
|
app.kubernetes.io/name: qwen2-7b-instruct-nvidia
|
|
app.kubernetes.io/part-of: llminferenceservice
|
|
kserve.io/component: workload
|
|
|
|
# No router — no scheduler, no route, no gateway.
|
|
# Traffic reaches these pods through the shared InferencePool
|
|
# managed by the NVIDIA LLMInferenceService's scheduler/EPP.
|
|
template:
|
|
containers:
|
|
- name: main
|
|
image: ghcr.io/llm-d/llm-d-rocm:v0.7.0
|
|
resources:
|
|
limits:
|
|
cpu: '4'
|
|
memory: 32Gi
|
|
amd.com/gpu: "1"
|
|
requests:
|
|
cpu: '2'
|
|
memory: 16Gi
|
|
amd.com/gpu: "1"
|