diff --git a/.dockerignore b/.dockerignore
new file mode 100644
index 0000000..0a68b83
--- /dev/null
+++ b/.dockerignore
@@ -0,0 +1,4 @@
+/models
+/.venv
+/.idea
+/.vscode
\ No newline at end of file
diff --git a/Dockerfile b/Dockerfile
new file mode 100644
index 0000000..f8e4b74
--- /dev/null
+++ b/Dockerfile
@@ -0,0 +1,14 @@
+# docker build . --platform linux/amd64 --push  -t leafdev.top/ecosystem/zero-shot-classification:v0.0.1
+FROM python:3.12.7
+
+#
+WORKDIR /app
+
+COPY requirements.txt /app
+RUN pip install --no-cache-dir --upgrade -r requirements.txt
+
+#
+COPY . /app
+
+#
+CMD ["uvicorn", "main:app", "--proxy-headers", "--host", "0.0.0.0", "--port", "80"]
\ No newline at end of file
diff --git a/classification.py b/classification.py
index a4b8eee..8b585c5 100644
--- a/classification.py
+++ b/classification.py
@@ -4,13 +4,13 @@ from pydantic import BaseModel
 from transformers import pipeline
 
 classifier = pipeline("zero-shot-classification",
-                      model="morit/chinese_xlm_xnli")
+                      model="MoritzLaurer/mDeBERTa-v3-base-mnli-xnli")
 
 
 # 返回一个结构化的内容
 class ClassifyResult(BaseModel):
     sequence: str
-    labels: list
+    rank: list
     scores: list
     prediction: str
 
@@ -18,14 +18,14 @@ class ClassifyResult(BaseModel):
 def classify(text: str, labels: list):
     output = classifier(text, labels)
 
-    pprint(output)
+    # pprint(output)
 
     # 根据 score，寻找最高的 label
     prediction_rank = output['scores'].index(max(output['scores']))
 
     return ClassifyResult(
         sequence=text,
-        labels=output['labels'],
+        rank=output['labels'],
         scores=output['scores'],
         prediction=output['labels'][prediction_rank]
     )
diff --git a/main.py b/main.py
index cc9f306..91c4bf8 100644
--- a/main.py
+++ b/main.py
@@ -1,5 +1,3 @@
-from typing import Union
-
 import uvicorn
 from fastapi import FastAPI
 from pydantic import BaseModel
@@ -7,23 +5,23 @@ import classification
 
 app = FastAPI()
 
+
 class TextClassificationRequest(BaseModel):
     text: str
     labels: list[str]
 
+
 class TextClassificationResponse(BaseModel):
     prediction: str
-    labels: list[str]
-
+    ranks: list[str]
 
 
 @app.post("/classify")
 def classify(req: TextClassificationRequest) -> TextClassificationResponse:
     result = classification.classify(req.text, req.labels)
 
-    return TextClassificationResponse(prediction=result.labels[0], labels=result.labels)
-
+    return TextClassificationResponse(prediction=result.prediction, ranks=result.rank)
 
 
 if __name__ == "__main__":
-    uvicorn.run(app, host="0.0.0.0", port=8000)
\ No newline at end of file
+    uvicorn.run(app, host="0.0.0.0", port=8000)
diff --git a/manifest.yaml b/manifest.yaml
new file mode 100644
index 0000000..ae9540d
--- /dev/null
+++ b/manifest.yaml
@@ -0,0 +1,90 @@
+apiVersion: v1
+kind: PersistentVolumeClaim
+metadata:
+  name: text-classification-pvc
+  namespace: ecosystem
+spec:
+  accessModes:
+    - ReadWriteMany
+  resources:
+    requests:
+      storage: 2Gi
+
+---
+apiVersion: apps/v1
+kind: Deployment
+metadata:
+  name: text-classification
+  namespace: ecosystem
+spec:
+  selector:
+    matchLabels:
+      app: text-classification
+      tier: backend
+  template:
+    metadata:
+      labels:
+        app: text-classification
+        tier: backend
+    spec:
+      volumes:
+        - name: text-classification-models
+          persistentVolumeClaim:
+            claimName: text-classification-pvc
+      containers:
+        - name: text-classification
+          image: leafdev.top/ecosystem/zero-shot-classification:v0.0.1
+          env:
+            - name: HF_DATASETS_CACHE
+              value: "/app/models"
+            - name: HF_HOME
+              value: "/app/models"
+            - name: HUGGINGFACE_HUB_CACHE
+              value: "/app/models"
+            - name: TRANSFORMERS_CACHE
+              value: "/app/models"
+#            - name: CUDA_VISIBLE_DEVICES
+#              value: "1"
+          ports:
+            - containerPort: 80
+              protocol: TCP
+              name: http
+          #          resources:
+          #            requests:
+          #              cpu: 1000m
+          #              memory: 1024Mi
+          volumeMounts:
+            - mountPath: /app/models
+              name: text-classification-models
+---
+apiVersion: v1
+kind: Service
+metadata:
+  name: text-classification
+  namespace: ecosystem
+spec:
+  selector:
+    app: text-classification
+    tier: backend
+  type: ClusterIP
+  ports:
+    - port: 80
+      targetPort: 80
+      protocol: TCP
+      name: http
+---
+apiVersion: gateway.networking.k8s.io/v1
+kind: HTTPRoute
+metadata:
+  name: zero-shot-classification-http
+  namespace: ecosystem
+spec:
+  hostnames:
+    - text-classification-api-testing.leaflow.cn
+  parentRefs:
+    - name: http-gw
+      namespace: networking
+  rules:
+    - backendRefs:
+        - name: text-classification
+          port: 80
\ No newline at end of file