feat: add hf token secret accessor

.github/workflows/ci-production.yml

@@ -4,6 +4,9 @@ on:
   push:
     branches:
       - production
+    path:
+      - "core-model-prediction/**"
+      - ".github/workflows/**"
 
 jobs:
   setup-build-publish-deploy:
@@ -26,7 +29,7 @@ jobs:
           gcloud auth configure-docker ${{ secrets.GCP_REPO_REGION }}-docker.pkg.dev --quiet
 
       - name: Build image
-        working-directory: ./
+        working-directory: ./core-model-prediction
         run: docker build . --file Dockerfile --tag ${{ secrets.GCP_REPO_REGION }}-docker.pkg.dev/${{ secrets.GCP_PROJECT_ID }}/interview-ai-detector/model-prediction
 
       - name: Push image
@@ -61,4 +64,5 @@ jobs:
             --model=${{ env.MODEL_ID }} \
             --display-name=interview-ai-detector-deployment \
             --machine-type="n1-standard-4" \
-            --accelerator=count=1,type=nvidia-tesla-t4
+            --accelerator=count=1,type=nvidia-tesla-t4 \
+            --verbosity="debug"

.gitignore

@@ -0,0 +1,3 @@
+__pycache__
+.env
+*.json

core-model-prediction/.gitignore

@@ -1 +0,0 @@
-__pycache__

core-model-prediction/gemma2b_dependencies.py

@@ -4,6 +4,7 @@ from torch.nn.functional import cosine_similarity
 from collections import Counter
 import numpy as np
 from device_manager import DeviceManager
+from google.cloud import secretmanager
 
 
 class Gemma2BDependencies:
@@ -12,14 +13,21 @@ class Gemma2BDependencies:
     def __new__(cls):
         if cls._instance is None:
             cls._instance = super(Gemma2BDependencies, cls).__new__(cls)
+            token = cls._instance.access_hf_token_secret()
             cls._instance.tokenizer = AutoTokenizer.from_pretrained(
-                "google/gemma-2b")
+                "google/gemma-2b", token=token)
             cls._instance.model = AutoModelForCausalLM.from_pretrained(
-                "google/gemma-2b")
+                "google/gemma-2b", token=token)
             cls._instance.device = DeviceManager()
             cls._instance.model.to(cls._instance.device)
         return cls._instance
 
+    def access_hf_token_secret(self):
+        client = secretmanager.SecretManagerServiceClient()
+        name = "projects/steady-climate-416810/secrets/HF_TOKEN/versions/1"
+        response = client.access_secret_version(request={"name": name})
+        return response.payload.data.decode('UTF-8')
+
     def calculate_perplexity(self, text: str):
         inputs = self.tokenizer(text, return_tensors="pt",
                                 truncation=True, max_length=1024)

core-model-prediction/prediction.py

@@ -4,8 +4,6 @@ from hypothesis import BaseModelHypothesis
 from random_forest_dependencies import RandomForestDependencies
 from random_forest_model import RandomForestModel
 from main_model import PredictMainModel
-import torch.nn as nn
-import torch
 import numpy as np
 from typing import List
 

core-model-prediction/requirements.txt

@@ -6,4 +6,5 @@ textstat
 scikit-learn==1.4.1.post1
 transformers
 fastapi
-uvicorn
+uvicorn
+google-cloud-secret-manager