github.com/1aal/kubeblocks@v0.0.0-20231107070852-e1c03e598921/deploy/llm/templates/clusterversion.yaml (about)

     1  apiVersion: apps.kubeblocks.io/v1alpha1
     2  kind: ClusterVersion
     3  metadata:
     4    name: vllm-baichuan-13b
     5    labels:
     6      {{- include "llm.labels" . | nindent 4 }}
     7  spec:
     8    clusterDefinitionRef: vllm
     9    componentVersions:
    10      - componentDefRef: vllm
    11        versionsContext:
    12          containers:
    13            - name: vllm
    14              image:  {{ .Values.image.registry | default "docker.io" }}/{{ .Values.image.repository}}:{{ default .Chart.AppVersion .Values.image.tag }}
    15              env:
    16                - name: MODEL_NAME
    17                  value: baichuan-inc/Baichuan-13B-Chat
    18                - name: EXTRA_ARGS
    19                  value: "--trust-remote-code"
    20  ---
    21  apiVersion: apps.kubeblocks.io/v1alpha1
    22  kind: ClusterVersion
    23  metadata:
    24    name: vllm-baichuan2-13b
    25    annotations:
    26      kubeblocks.io/is-default-cluster-version: "true"
    27    labels:
    28      {{- include "llm.labels" . | nindent 4 }}
    29  spec:
    30    clusterDefinitionRef: vllm
    31    componentVersions:
    32      - componentDefRef: vllm
    33        versionsContext:
    34          containers:
    35            - name: vllm
    36              image:  {{ .Values.image.registry | default "docker.io" }}/{{ .Values.image.repository}}:{{ default .Chart.AppVersion .Values.image.tag }}
    37              env:
    38                - name: MODEL_NAME
    39                  value: baichuan-inc/Baichuan2-13B-Chat
    40                - name: EXTRA_ARGS
    41                  value: "--trust-remote-code"
    42  ---
    43  apiVersion: apps.kubeblocks.io/v1alpha1
    44  kind: ClusterVersion
    45  metadata:
    46    name: vllm-baichuan2-13b-cn
    47    labels:
    48      {{- include "llm.labels" . | nindent 4 }}
    49  spec:
    50    clusterDefinitionRef: vllm
    51    componentVersions:
    52      - componentDefRef: vllm
    53        versionsContext:
    54          containers:
    55            - name: vllm
    56              image:  {{ .Values.image.registry | default "docker.io" }}/{{ .Values.image.repository}}:{{ default .Chart.AppVersion .Values.image.tag }}
    57              env:
    58                - name: MODEL_NAME
    59                  value: ../Baichuan2-13B-Chat
    60                - name: EXTRA_ARGS
    61                  value: "--trust-remote-code"
    62                - name: CLONE_MODEL_SCRIPT
    63                  value: "git lfs install; git clone https://www.modelscope.cn/baichuan-inc/Baichuan2-13B-Chat.git"
    64  ---
    65  apiVersion: apps.kubeblocks.io/v1alpha1
    66  kind: ClusterVersion
    67  metadata:
    68    name: vllm-baichuan-7b
    69    labels:
    70      {{- include "llm.labels" . | nindent 4 }}
    71  spec:
    72    clusterDefinitionRef: vllm
    73    componentVersions:
    74      - componentDefRef: vllm
    75        versionsContext:
    76          containers:
    77            - name: vllm
    78              image:  {{ .Values.image.registry | default "docker.io" }}/{{ .Values.image.repository}}:{{ default .Chart.AppVersion .Values.image.tag }}
    79              env:
    80                - name: MODEL_NAME
    81                  value: baichuan-inc/Baichuan-7B
    82                - name: EXTRA_ARGS
    83                  value: "--trust-remote-code"
    84  ---
    85  apiVersion: apps.kubeblocks.io/v1alpha1
    86  kind: ClusterVersion
    87  metadata:
    88    name: vllm-baichuan2-7b
    89    labels:
    90      {{- include "llm.labels" . | nindent 4 }}
    91  spec:
    92    clusterDefinitionRef: vllm
    93    componentVersions:
    94      - componentDefRef: vllm
    95        versionsContext:
    96          containers:
    97            - name: vllm
    98              image:  {{ .Values.image.registry | default "docker.io" }}/{{ .Values.image.repository}}:{{ default .Chart.AppVersion .Values.image.tag }}
    99              env:
   100                - name: MODEL_NAME
   101                  value: baichuan-inc/Baichuan2-7B-Chat
   102                - name: EXTRA_ARGS
   103                  value: "--trust-remote-code"
   104  ---
   105  apiVersion: apps.kubeblocks.io/v1alpha1
   106  kind: ClusterVersion
   107  metadata:
   108    name: vllm-vicuna-13b
   109    labels:
   110      {{- include "llm.labels" . | nindent 4 }}
   111  spec:
   112    clusterDefinitionRef: vllm
   113    componentVersions:
   114      - componentDefRef: vllm
   115        versionsContext:
   116          containers:
   117            - name: vllm
   118              image:  {{ .Values.image.registry | default "docker.io" }}/{{ .Values.image.repository}}:{{ default .Chart.AppVersion .Values.image.tag }}
   119              env:
   120                - name: MODEL_NAME
   121                  value: lmsys/vicuna-13b-v1.3
   122                - name: EXTRA_ARGS
   123                  value: "--trust-remote-code"
   124  ---
   125  apiVersion: apps.kubeblocks.io/v1alpha1
   126  kind: ClusterVersion
   127  metadata:
   128    name: vllm-vicuna-7b
   129    labels:
   130      {{- include "llm.labels" . | nindent 4 }}
   131  spec:
   132    clusterDefinitionRef: vllm
   133    componentVersions:
   134      - componentDefRef: vllm
   135        versionsContext:
   136          containers:
   137            - name: vllm
   138              image:  {{ .Values.image.registry | default "docker.io" }}/{{ .Values.image.repository}}:{{ default .Chart.AppVersion .Values.image.tag }}
   139              env:
   140                - name: MODEL_NAME
   141                  value: lmsys/vicuna-7b-v1.5
   142                - name: EXTRA_ARGS
   143                  value: "--trust-remote-code"
   144  ---
   145  apiVersion: apps.kubeblocks.io/v1alpha1
   146  kind: ClusterVersion
   147  metadata:
   148    name: vllm-llama-70b
   149    labels:
   150      {{- include "llm.labels" . | nindent 4 }}
   151  spec:
   152    clusterDefinitionRef: vllm
   153    componentVersions:
   154      - componentDefRef: vllm
   155        versionsContext:
   156          containers:
   157            - name: vllm
   158              image:  {{ .Values.image.registry | default "docker.io" }}/{{ .Values.image.repository}}:{{ default .Chart.AppVersion .Values.image.tag }}
   159              env:
   160                - name: MODEL_NAME
   161                  value: upstage/SOLAR-0-70b-16bit
   162                - name: EXTRA_ARGS
   163                  value: "--trust-remote-code --swap-space 1"
   164  ---
   165  apiVersion: apps.kubeblocks.io/v1alpha1
   166  kind: ClusterVersion
   167  metadata:
   168    name: ggml-baichuan-7b-q4
   169    annotations:
   170      kubeblocks.io/is-default-cluster-version: "true"
   171    labels:
   172      {{- include "llm.labels" . | nindent 4 }}
   173  spec:
   174    clusterDefinitionRef: ggml
   175    componentVersions:
   176      - componentDefRef: ggml
   177        versionsContext:
   178          initContainers:
   179            - name: download
   180              image: infracreate-registry.cn-zhangjiakou.cr.aliyuncs.com/apecloud/baichuan-llama-7b:ggmlv3.q4_0
   181              command: ["sh", "-c", "cp /models/baichuan-llama-7b.ggmlv3.q4_0.bin /models-target/"]
   182              volumeMounts:
   183                - name: models
   184                  mountPath: /models-target
   185          containers:
   186            - name: ggml
   187              image: {{ .Values.imageDev.registry | default "docker.io" }}/{{ .Values.imageDev.repository}}:{{ default .Chart.AppVersion .Values.imageDev.tag }}
   188              env:
   189                - name: MODEL
   190                  value: /models/baichuan-llama-7b.ggmlv3.q4_0.bin
   191              volumeMounts:
   192                - name: models
   193                  mountPath: /models
   194  ---
   195  apiVersion: apps.kubeblocks.io/v1alpha1
   196  kind: ClusterVersion
   197  metadata:
   198    name: ggml-baichuan2-13b-q4
   199    labels:
   200      {{- include "llm.labels" . | nindent 4 }}
   201  spec:
   202    clusterDefinitionRef: ggml
   203    componentVersions:
   204      - componentDefRef: ggml
   205        versionsContext:
   206          initContainers:
   207            - name: download
   208              image: infracreate-registry.cn-zhangjiakou.cr.aliyuncs.com/apecloud/baichuan2-13b-gguf:ggml-model-q4
   209              command: ["sh", "-c", "cp /models/ggml-model-q4.gguf /models-target/"]
   210              volumeMounts:
   211                - name: models
   212                  mountPath: /models-target
   213          containers:
   214            - name: ggml
   215              image: {{ .Values.imageDev.registry | default "docker.io" }}/{{ .Values.imageDev.repository}}:{{ default .Chart.AppVersion .Values.imageDev.tagNew }}
   216              env:
   217                - name: MODEL
   218                  value: /models/ggml-model-q4.gguf
   219              volumeMounts:
   220                - name: models
   221                  mountPath: /models
   222  ---
   223  apiVersion: apps.kubeblocks.io/v1alpha1
   224  kind: ClusterVersion
   225  metadata:
   226    name: ggml-replit-code-3b-f16
   227    labels:
   228      {{- include "llm.labels" . | nindent 4 }}
   229  spec:
   230    clusterDefinitionRef: ggml
   231    componentVersions:
   232      - componentDefRef: ggml
   233        versionsContext:
   234          initContainers:
   235            - name: download
   236              image: infracreate-registry.cn-zhangjiakou.cr.aliyuncs.com/apecloud/replit-code-3b:ggml-model-f16
   237              command: ["sh", "-c", "cp /models/ggml-model-f16.gguf /models-target/"]
   238              volumeMounts:
   239                - name: models
   240                  mountPath: /models-target
   241          containers:
   242            - name: ggml
   243              image: {{ .Values.imageDev.registry | default "docker.io" }}/{{ .Values.imageDev.repository}}:{{ default .Chart.AppVersion .Values.imageDev.tagNew }}
   244              env:
   245                - name: MODEL
   246                  value: /models/ggml-model-f16.gguf
   247              volumeMounts:
   248                - name: models
   249                  mountPath: /models
   250  ---
   251  apiVersion: apps.kubeblocks.io/v1alpha1
   252  kind: ClusterVersion
   253  metadata:
   254    name: ggml-codeshell-7b-chat-q4
   255    labels:
   256      {{- include "llm.labels" . | nindent 4 }}
   257  spec:
   258    clusterDefinitionRef: ggml
   259    componentVersions:
   260      - componentDefRef: ggml
   261        versionsContext:
   262          initContainers:
   263            - name: download
   264              image: infracreate-registry.cn-zhangjiakou.cr.aliyuncs.com/apecloud/codeshell-7b-chat:codeshell-chat-q4_0
   265              command: ["sh", "-c", "cp /models/codeshell-chat-q4_0.gguf /models-target/"]
   266              volumeMounts:
   267                - name: models
   268                  mountPath: /models-target
   269          containers:
   270            - name: ggml
   271              image: {{ .Values.imageDev.registry | default "docker.io" }}/{{ .Values.imageDev.repository}}:{{ default .Chart.AppVersion .Values.imageDev.tagCodeShell }}
   272              env:
   273                - name: MODEL
   274                  value: /models/codeshell-chat-q4_0.gguf
   275              volumeMounts:
   276                - name: models
   277                  mountPath: /models
   278  ---
   279  apiVersion: apps.kubeblocks.io/v1alpha1
   280  kind: ClusterVersion
   281  metadata:
   282    name: ggml-baichuan2-7b-4q
   283    labels:
   284      {{- include "llm.labels" . | nindent 4 }}
   285  spec:
   286    clusterDefinitionRef: ggml
   287    componentVersions:
   288      - componentDefRef: ggml
   289        versionsContext:
   290          initContainers:
   291            - name: download
   292              image: {{ .Values.model.registry | default "docker.io" }}/{{ .Values.model.baichuan2_7b.ggml.repository}}:{{ default .Chart.AppVersion .Values.model.baichuan2_7b.ggml.version }}
   293              command: ["sh", "-c", "cp /models/ggml-model-q4.gguf /models-target/"]
   294              volumeMounts:
   295                - name: models
   296                  mountPath: /models-target
   297          containers:
   298            - name: ggml
   299              image: {{ .Values.imageDev.registry | default "docker.io" }}/{{ .Values.imageDev.repository}}:{{ default .Chart.AppVersion .Values.imageDev.tagNew }}
   300              env:
   301                  - name: MODEL
   302                    value: /models/ggml-model-q4.gguf
   303              volumeMounts:
   304                - name: models
   305                  mountPath: /models