k

backend/Dockerfile

@@ -68,7 +68,7 @@ RUN apt-get update && \
     rm -f /usr/local/lib/python3.11/site-packages/tornado/test/test.key
 
 # Pre-downloading models for setups with limited egress
-RUN python -c "from transformers import AutoTokenizer; AutoTokenizer.from_pretrained('intfloat/e5-base-v2')"
+RUN python -c "from transformers import AutoTokenizer; AutoTokenizer.from_pretrained('nomic-ai/nomic-embed-text-v1.5')"
 
 # Pre-downloading NLTK for setups with limited egress
 RUN python -c "import nltk; \

backend/Dockerfile.model_server

@@ -18,6 +18,8 @@ RUN apt-get remove -y --allow-remove-essential perl-base && \
     apt-get autoremove -y
 
 # Pre-downloading models for setups with limited egress
+ENV HF_HOME=/hf_model_cache
+WORKDIR /hf_model_cache
 RUN python -c "from transformers import AutoModel, AutoTokenizer, TFDistilBertForSequenceClassification; \
 from huggingface_hub import snapshot_download; \
 AutoTokenizer.from_pretrained('danswer/intent-model'); \
@@ -27,6 +29,10 @@ snapshot_download('danswer/intent-model'); \
 snapshot_download('intfloat/e5-base-v2'); \
 snapshot_download('mixedbread-ai/mxbai-rerank-xsmall-v1')"
 
+RUN python -c "from sentence_transformers import SentenceTransformer; \
+model = SentenceTransformer('nomic-ai/nomic-embed-text-v1.5', trust_remote_code=True); \
+model.save('nomic-ai/nomic-embed-text-v1.5')"
+
 WORKDIR /app
 
 # Utils used by model server

backend/danswer/configs/model_configs.py

@@ -12,10 +12,9 @@ import os
 # The useable models configured as below must be SentenceTransformer compatible
 # NOTE: DO NOT CHANGE SET THESE UNLESS YOU KNOW WHAT YOU ARE DOING
 # IDEALLY, YOU SHOULD CHANGE EMBEDDING MODELS VIA THE UI
-DEFAULT_DOCUMENT_ENCODER_MODEL = "intfloat/e5-base-v2"
+DEFAULT_DOCUMENT_ENCODER_MODEL = "nomic-ai/nomic-embed-text-v1"
-DOCUMENT_ENCODER_MODEL = (
+DOCUMENT_ENCODER_MODEL = DEFAULT_DOCUMENT_ENCODER_MODEL
-    os.environ.get("DOCUMENT_ENCODER_MODEL") or DEFAULT_DOCUMENT_ENCODER_MODEL
+
-)
 # If the below is changed, Vespa deployment must also be changed
 DOC_EMBEDDING_DIM = int(os.environ.get("DOC_EMBEDDING_DIM") or 768)
 # Model should be chosen with 512 context size, ideally don't change this

backend/model_server/encoders.py

@@ -82,10 +82,10 @@ def embed_text(
     max_context_length: int,
     normalize_embeddings: bool,
 ) -> list[list[float]]:
-    model = get_embedding_model(
+    model = SentenceTransformer(
-        model_name=model_name, max_context_length=max_context_length
+        "nomic-ai/nomic-embed-text-v1.5", trust_remote_code=True
     )
-    embeddings = model.encode(texts, normalize_embeddings=normalize_embeddings)
+    embeddings = model.encode(texts)
 
     if not isinstance(embeddings, list):
         embeddings = embeddings.tolist()

backend/requirements/model_server.txt

@@ -1,3 +1,4 @@
+einops==0.8.0
 fastapi==0.109.2
 h5py==3.9.0
 pydantic==1.10.13

deployment/docker_compose/docker-compose.dev.yml

@@ -101,6 +101,8 @@ services:
       options:
         max-size: "50m"
         max-file: "6"
+    networks:
+      - internal
 
 
   background:
@@ -203,6 +205,8 @@ services:
       options:
         max-size: "50m"
         max-file: "6"
+    networks:
+      - internal
 
 
   web_server:
@@ -231,6 +235,8 @@ services:
 
       # Enterprise Edition only
       - ENABLE_PAID_ENTERPRISE_EDITION_FEATURES=${ENABLE_PAID_ENTERPRISE_EDITION_FEATURES:-false}
+    networks:
+      - internal
 
 
   inference_model_server:
@@ -258,6 +264,8 @@ services:
       options:
         max-size: "50m"
         max-file: "6"
+    networks:
+      - internal
 
 
   indexing_model_server:
@@ -286,6 +294,9 @@ services:
       options:
         max-size: "50m"
         max-file: "6"
+    networks:
+      - internal
+
 
   relational_db:
     image: postgres:15.2-alpine
@@ -297,6 +308,8 @@ services:
       - "5432:5432"
     volumes:
       - db_volume:/var/lib/postgresql/data
+    networks:
+      - internal
 
 
   # This container name cannot have an underscore in it due to Vespa expectations of the URL
@@ -313,6 +326,8 @@ services:
       options:
         max-size: "50m"
         max-file: "6"
+    networks:
+      - internal
 
 
   nginx:
@@ -343,6 +358,9 @@ services:
     command: > 
       /bin/sh -c "dos2unix /etc/nginx/conf.d/run-nginx.sh 
       && /etc/nginx/conf.d/run-nginx.sh app.conf.template.dev"
+    networks:
+      - internal
+      - local
         
 
 volumes:
@@ -350,3 +368,10 @@ volumes:
   vespa_volume:
   # Created by the container itself
   model_cache_huggingface:
+
+
+networks:
+  internal:
+    internal: true
+  local:
+    internal: false