Testing

.github/workflows/docker-build-push-backend-container-on-tag.yml

@@ -31,14 +31,5 @@ jobs:
         push: true
         tags: |
           danswer/danswer-backend:${{ github.ref_name }}
-          danswer/danswer-backend:latest
         build-args: |
           DANSWER_VERSION=${{ github.ref_name }}
-
-    - name: Run Trivy vulnerability scanner
-      uses: aquasecurity/trivy-action@master
-      with:
-        # To run locally: trivy image --severity HIGH,CRITICAL danswer/danswer-backend
-        image-ref: docker.io/danswer/danswer-backend:${{ github.ref_name }}
-        severity: 'CRITICAL,HIGH'
-        trivyignores: ./backend/.trivyignore

.github/workflows/docker-build-push-model-server-container-on-tag.yml

@@ -31,12 +31,5 @@ jobs:
         push: true
         tags: |
           danswer/danswer-model-server:${{ github.ref_name }}
-          danswer/danswer-model-server:latest
         build-args: |
           DANSWER_VERSION=${{ github.ref_name }}
-
-    - name: Run Trivy vulnerability scanner
-      uses: aquasecurity/trivy-action@master
-      with:
-        image-ref: docker.io/danswer/danswer-model-server:${{ github.ref_name }}
-        severity: 'CRITICAL,HIGH'

.github/workflows/docker-build-push-web-container-on-tag.yml

@@ -31,12 +31,5 @@ jobs:
         push: true
         tags: |
           danswer/danswer-web-server:${{ github.ref_name }}
-          danswer/danswer-web-server:latest
         build-args: |
           DANSWER_VERSION=${{ github.ref_name }}
-
-    - name: Run Trivy vulnerability scanner
-      uses: aquasecurity/trivy-action@master
-      with:
-        image-ref: docker.io/danswer/danswer-web-server:${{ github.ref_name }}
-        severity: 'CRITICAL,HIGH'

backend/danswer/search/search_nlp_models.py

@@ -5,7 +5,9 @@ from typing import Optional
 from typing import TYPE_CHECKING
 
 import requests
+from huggingface_hub.constants import HF_HOME
 from transformers import logging as transformer_logging  # type:ignore
+from transformers.utils import TRANSFORMERS_CACHE
 
 from danswer.configs.model_configs import DOC_EMBEDDING_CONTEXT_SIZE
 from danswer.configs.model_configs import DOCUMENT_ENCODER_MODEL
@@ -58,6 +60,17 @@ def get_default_tokenizer(model_name: str | None = None) -> "AutoTokenizer":
             # This could be inaccurate
             model_name = DOCUMENT_ENCODER_MODEL
 
+        print(f"HF Home: {HF_HOME}")
+        print(f"Cache dir: {TRANSFORMERS_CACHE}")
+
+        from danswer.search.hf_test import hf_hub_download
+        from transformers.tokenization_utils_base import TOKENIZER_CONFIG_FILE
+
+        hf_file_path = hf_hub_download(
+            model_name, TOKENIZER_CONFIG_FILE, local_files_only=True
+        )
+        print("hf_file_path", hf_file_path)
+
         _TOKENIZER = (AutoTokenizer.from_pretrained(model_name), model_name)
 
         if hasattr(_TOKENIZER[0], "is_fast") and _TOKENIZER[0].is_fast:

backend/model_server/custom_models.py

@@ -3,8 +3,10 @@ from typing import Optional
 import numpy as np
 import tensorflow as tf  # type: ignore
 from fastapi import APIRouter
+from huggingface_hub.constants import HF_HOME
 from transformers import AutoTokenizer  # type: ignore
 from transformers import TFDistilBertForSequenceClassification
+from transformers.utils import TRANSFORMERS_CACHE
 
 from model_server.constants import MODEL_WARM_UP_STRING
 from model_server.utils import simple_log_function_time
@@ -24,6 +26,17 @@ _INTENT_MODEL: Optional[TFDistilBertForSequenceClassification] = None
 def get_intent_model_tokenizer(
     model_name: str = INTENT_MODEL_VERSION,
 ) -> "AutoTokenizer":
+    print(f"HF Home: {HF_HOME}")
+    print(f"Cache dir: {TRANSFORMERS_CACHE}")
+
+    from model_server.hf import hf_hub_download
+    from transformers.tokenization_utils_base import TOKENIZER_CONFIG_FILE
+
+    hf_file_path = hf_hub_download(
+        model_name, TOKENIZER_CONFIG_FILE, local_files_only=True
+    )
+    print("hf_file_path", hf_file_path)
+
     global _INTENT_TOKENIZER
     if _INTENT_TOKENIZER is None:
         _INTENT_TOKENIZER = AutoTokenizer.from_pretrained(model_name)
@@ -36,6 +49,8 @@ def get_local_intent_model(
 ) -> TFDistilBertForSequenceClassification:
     global _INTENT_MODEL
     if _INTENT_MODEL is None or max_context_length != _INTENT_MODEL.max_seq_length:
+        print(f"HF Home: {HF_HOME}")
+        print(f"Cache dir: {TRANSFORMERS_CACHE}")
         _INTENT_MODEL = TFDistilBertForSequenceClassification.from_pretrained(
             model_name
         )