Merge pull request #3 from neuro-inc/eng-393

ENG-393: Use dedicated app, pgvector settings

.neuro/live.yaml

@@ -71,7 +71,7 @@ jobs:
   vllm:
     image: vllm/vllm-openai:v0.6.1.post2
     name: vllm
-    preset: gpu-medium
+    preset: a100x1
     detach: true
     http_port: "8000"
     volumes:
@@ -82,8 +82,6 @@ jobs:
       --model meta-llama/Meta-Llama-3.1-8B-Instruct
       --tokenizer meta-llama/Meta-Llama-3.1-8B-Instruct
       --dtype=half
-      --max-model-len=50000
-      --tensor-parallel-size=2
     # cmd: >
     #   --model meta-llama/Meta-Llama-3.1-8B-Instruct
     #   --tokenizer meta-llama/Meta-Llama-3.1-8B-Instruct
@@ -102,7 +100,7 @@ jobs:
     image: ollama/ollama:0.1.35
     volumes:
       - ${{ volumes.ollama_models.ref_rw }}
-    preset: gpu-small
+    preset: a100x1
     detach: true
     env:
       MODEL: "nomic-embed-text"
@@ -119,5 +117,5 @@ jobs:
       POSTGRES_DB: postgres
       PGDATA: ${{ volumes.pgdata.mount }}/pgdata
     volumes:
-       - ${{ volumes.pgdata.ref_rw }}
-#      - ${{ volumes.pgdata_onprem.ref_rw }}
+      - ${{ volumes.pgdata.ref_rw }}
+      # - ${{ volumes.pgdata_onprem.ref_rw }}

settings/settings-app.yaml

@@ -0,0 +1,33 @@
+server:
+  env_name: ${APP_ENV:prod}
+  port: ${PORT:8080}
+
+llm:
+  mode: openailike
+  tokenizer: ${VLLM_TOKENIZER:-}
+  max_new_tokens: ${VLLM_MAX_NEW_TOKENS:5000}
+  context_window: ${VLLM_CONTEXT_WINDOW:4096}
+  temperature: ${VLLM_TEMPERATURE:0.1}
+
+openai:
+  api_base: ${VLLM_API_BASE:https://api.openai.com/v1/}
+  api_key: ${VLLM_API_KEY:EMPTY}
+  model: ${VLLM_MODEL:-}
+  embedding_api_base: ${EMBEDDING_API_BASE:https://api.openai.com/v1/}
+  embedding_api_key: ${EMBEDDING_API_KEY:EMPTY}
+  embeddings_model: ${EMBEDDING_MODEL:text-embedding-ada-002}
+
+embedding:
+  mode: ${EMBEDDING_PROVIDER:openai}
+  ingest_mode: ${EMBEDDING_MODE:parallel}
+  embed_dim: ${EMBEDDING_DIM:768}
+
+ollama:
+  # Note: if you change embedding model, you'll need to use a dedicated DB for ingext storage
+  embedding_model: ${OLLAMA_EMBEDDING_MODEL:nomic-embed-text}
+  embedding_api_base: ${OLLAMA_API_BASE:http://localhost:11434}
+  request_timeout: 300.0
+
+ui:
+  enabled: true
+  path: /

settings/settings-pgvector.yaml

@@ -0,0 +1,13 @@
+nodestore:
+  database: postgres
+
+vectorstore:
+  database: postgres
+
+postgres:
+  host: ${POSTGRES_HOST:localhost}
+  port: ${POSTGRES_PORT:5432}
+  database: ${POSTGRES_DB:postgres}
+  user: ${POSTGRES_USER:postgres}
+  password: ${POSTGRES_PASSWORD:postgres}
+  schema_name: ${POSTGRES_SHEMA:postgres}