Skip to content

update pd disaggregation templates and example #133

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Closed
elieserr wants to merge 4 commits into from
Closed

update pd disaggregation templates and example #133

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
4 commits
Select commit Hold shift + click to select a range
d76a008
update pd dissagregation example to use inference scheduler 0.3.1 and…
elieserr Oct 15, 2025
0265cf6
fix linter error + include output-pd.yaml
elieserr Oct 15, 2025
1254d08
include generated files
elieserr Oct 20, 2025
7d78630
use proper image
elieserr Oct 20, 2025
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
18 changes: 10 additions & 8 deletions charts/llm-d-modelservice/templates/epp-deployment.yaml
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -21,20 +21,22 @@ spec:
imagePullPolicy: Always
image: {{ required "routing.epp.image must be specified" .Values.routing.epp.image }}
args:
- --poolName
- --pool-name
- {{ include "llm-d-modelservice.inferencePoolName" . }}
- --poolNamespace
- --pool-namespace
- {{ .Release.Namespace }}
- -v
- "{{ default 4 .Values.routing.epp.debugLevel }}"
- "--pool-group"
- "inference.networking.x-k8s.io"
- --v
- "4"
- --zap-encoder
- json
- --grpcPort
- "json"
- --grpc-port
- "9002"
- --grpcHealthPort
- --grpc-health-port
- "9003"
{{- if .Values.routing.epp.pluginsConfigFile }}
- "-configFile"
- "--config-file"
- "config/{{ .Values.routing.epp.pluginsConfigFile }}"
{{- end}}
{{- with .Values.routing.epp.env }}
Expand Down
8 changes: 8 additions & 0 deletions charts/llm-d-modelservice/templates/epp-plugin-configmap.yaml
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -83,7 +83,15 @@ data:
maxPrefixBlocksToMatch: 256
lruCapacityPerServer: 31250
- type: prefill-filter
parameters:
selector:
matchLabels:
llm-d.ai/role: prefill
- type: decode-filter
parameters:
selector:
matchLabels:
llm-d.ai/role: decode
- type: max-score-picker
- type: pd-profile-handler
parameters:
Expand Down
2 changes: 1 addition & 1 deletion charts/llm-d-modelservice/templates/epp-role.yaml
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -7,7 +7,7 @@ rules:
- apiGroups:
- inference.networking.x-k8s.io
resources:
- inferencemodels
- inferenceobjectives
- inferencepools
verbs:
- get
Expand Down
1 change: 0 additions & 1 deletion charts/llm-d-modelservice/templates/httproute.yaml
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -25,7 +25,6 @@ spec:
- group: {{ .Values.routing.inferencePool.apiGroup }}
kind: InferencePool
name: {{ include "llm-d-modelservice.inferencePoolName" . }}
port: {{ .Values.routing.servicePort }}
weight: 1
{{- if .Values.routing.httpRoute.timeouts}}
timeouts:
Expand Down
6 changes: 3 additions & 3 deletions charts/llm-d-modelservice/values.yaml
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -169,9 +169,9 @@ routing:

# Allow people to opt out of timeouts by unsetting the default value.
# They are set to 0s which in most situations defaults to the providers max timeout.
timeouts:
backendRequest: 0s
request: 0s
# timeouts:
# backendRequest: 0s
# request: 0s

# @schema
# additionalProperties: true
Expand Down
32 changes: 19 additions & 13 deletions examples/output-cpu.yaml
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -105,7 +105,15 @@ data:
maxPrefixBlocksToMatch: 256
lruCapacityPerServer: 31250
- type: prefill-filter
parameters:
selector:
matchLabels:
llm-d.ai/role: prefill
- type: decode-filter
parameters:
selector:
matchLabels:
llm-d.ai/role: decode
- type: max-score-picker
- type: pd-profile-handler
parameters:
Expand Down Expand Up @@ -134,7 +142,7 @@ rules:
- apiGroups:
- inference.networking.x-k8s.io
resources:
- inferencemodels
- inferenceobjectives
- inferencepools
verbs:
- get
Expand Down Expand Up @@ -308,19 +316,21 @@ spec:
imagePullPolicy: Always
image: ghcr.io/llm-d/llm-d-inference-scheduler:v0.2.1
args:
- --poolName
- --pool-name
- cpu-sim-llm-d-modelservice
- --poolNamespace
- --pool-namespace
- default
- -v
- "6"
- "--pool-group"
- "inference.networking.x-k8s.io"
- --v
- "4"
- --zap-encoder
- json
- --grpcPort
- "json"
- --grpc-port
- "9002"
- --grpcHealthPort
- --grpc-health-port
- "9003"
- "-configFile"
- "--config-file"
- "config/default-config.yaml"
ports:
- containerPort: 9002
Expand Down Expand Up @@ -448,11 +458,7 @@ spec:
- group: inference.networking.x-k8s.io
kind: InferencePool
name: cpu-sim-llm-d-modelservice
port: 8000
weight: 1
timeouts:
backendRequest: 0s
request: 0s
matches:
- headers:
- name: x-model-name
Expand Down
76 changes: 36 additions & 40 deletions examples/output-pd.yaml
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -105,7 +105,15 @@ data:
maxPrefixBlocksToMatch: 256
lruCapacityPerServer: 31250
- type: prefill-filter
parameters:
selector:
matchLabels:
llm-d.ai/role: prefill
- type: decode-filter
parameters:
selector:
matchLabels:
llm-d.ai/role: decode
- type: max-score-picker
- type: pd-profile-handler
parameters:
Expand Down Expand Up @@ -134,7 +142,7 @@ rules:
- apiGroups:
- inference.networking.x-k8s.io
resources:
- inferencemodels
- inferenceobjectives
- inferencepools
verbs:
- get
Expand Down Expand Up @@ -254,7 +262,7 @@ spec:

containers:
- name: vllm
image: ghcr.io/llm-d/llm-d:v0.2.0
image: ghcr.io/llm-d/llm-d-cuda:v0.3.0

command: ["vllm", "serve"]
args:
Expand All @@ -277,7 +285,7 @@ spec:
fieldRef:
fieldPath: status.podIP
- name: VLLM_NIXL_SIDE_CHANNEL_PORT
value: "5557"
value: "5600"
- name: VLLM_LOGGING_LEVEL
value: DEBUG
- name: DP_SIZE
Expand All @@ -292,14 +300,16 @@ spec:
protocol: TCP
- containerPort: 5557
protocol: TCP
- containerPort: 5600
protocol: TCP

resources:
limits:
cpu: "16"
cpu: "6"
memory: 16Gi
nvidia.com/gpu: "1"
requests:
cpu: "16"
cpu: "6"
memory: 16Gi
nvidia.com/gpu: "1"

Expand Down Expand Up @@ -328,22 +338,24 @@ spec:
containers:
- name: epp
imagePullPolicy: Always
image: ghcr.io/llm-d/llm-d-inference-scheduler:v0.2.1
image: ghcr.io/llm-d/llm-d-inference-scheduler:v0.3.1
args:
- --poolName
- --pool-name
- pd-llm-d-modelservice
- --poolNamespace
- --pool-namespace
- default
- -v
- "--pool-group"
- "inference.networking.x-k8s.io"
- --v
- "4"
- --zap-encoder
- json
- --grpcPort
- "json"
- --grpc-port
- "9002"
- --grpcHealthPort
- --grpc-health-port
- "9003"
- "-configFile"
- "config/default-config.yaml"
- "--config-file"
- "config/default-pd-config.yaml"
ports:
- containerPort: 9002
name: grpc
Expand Down Expand Up @@ -417,7 +429,7 @@ spec:

containers:
- name: vllm
image: ghcr.io/llm-d/llm-d:v0.2.0
image: ghcr.io/llm-d/llm-d-cuda:v0.3.0

command: ["vllm", "serve"]
args:
Expand All @@ -436,7 +448,7 @@ spec:
- name: UCX_TLS
value: cuda_ipc,cuda_copy,tcp
- name: VLLM_NIXL_SIDE_CHANNEL_PORT
value: "5557"
value: "5600"
- name: VLLM_NIXL_SIDE_CHANNEL_HOST
valueFrom:
fieldRef:
Expand All @@ -455,35 +467,23 @@ spec:
protocol: TCP
- containerPort: 5557
protocol: TCP
- containerPort: 5600
protocol: TCP

resources:
limits:
cpu: "16"
cpu: "6"
memory: 16Gi
nvidia.com/gpu: "1"
requests:
cpu: "16"
cpu: "6"
memory: 16Gi
nvidia.com/gpu: "1"

volumeMounts:
- name: model-storage
mountPath: /model-cache
---
# Source: llm-d-modelservice/templates/inferencemodel.yaml
apiVersion: inference.networking.x-k8s.io/v1alpha2
kind: InferenceModel
metadata:
labels:
llm-d.ai/inferenceServing: "true"
llm-d.ai/model: pd-llm-d-modelservice
name: pd-llm-d-modelservice
spec:
criticality: Critical
modelName: facebook/opt-125m
poolRef:
name: pd-llm-d-modelservice
---
# Source: llm-d-modelservice/templates/inferencepool.yaml
apiVersion: inference.networking.x-k8s.io/v1alpha2
kind: InferencePool
Expand Down Expand Up @@ -524,13 +524,9 @@ spec:
- group: inference.networking.x-k8s.io
kind: InferencePool
name: pd-llm-d-modelservice
port: 8000
weight: 1
timeouts:
backendRequest: 0s
request: 0s
matches:
- headers:
- name: x-model-name
type: Exact
value: 'facebook/opt-125m'
- headers: []
path:
type: PathPrefix
value: /
Loading
Loading