[chatqna-core] Rename config variable name and create validator class

Signed-off-by: Yeoh, Hoong Tee <[email protected]>

Author: hteeyeoh

sample-applications/chat-question-and-answer-core/app/chain.py

@@ -17,18 +17,18 @@
 # If RUN_TEST is set to "True", the model download and conversion steps are skipped.
 # This flag is set in the conftest.py file before running the tests.
 if os.getenv("RUN_TEST", "").lower() != "true":
-    if config.MODEL_BACKEND == "openvino":
-        backend_module = importlib.import_module("app.openvino_backend")
-        backend_instance = backend_module.OpenVINOBackend()
+    if config.MODEL_RUNTIME == "openvino":
+        runtime_module = importlib.import_module("app.openvino_backend")
+        runtime_instance = runtime_module.OpenVINOBackend()
 
-    elif config.MODEL_BACKEND == "ollama":
-        backend_module = importlib.import_module("app.ollama_backend")
-        backend_instance = backend_module.OllamaBackend()
+    elif config.MODEL_RUNTIME == "ollama":
+        runtime_module = importlib.import_module("app.ollama_backend")
+        runtime_instance = runtime_module.OllamaBackend()
 
     else:
-        raise ValueError(f"Unsupported model backend: {config.MODEL_BACKEND}")
+        raise ValueError(f"Unsupported model runtime: {config.MODEL_RUNTIME}")
 
-    embedding, llm, reranker = backend_instance.init_models()
+    embedding, llm, reranker = runtime_instance.init_models()
 
     template = config.PROMPT_TEMPLATE
 
@@ -62,6 +62,7 @@ def get_retriever():
     """
 
     enable_rerank = config._ENABLE_RERANK
+    logger.info(f"Reranker enabled: {enable_rerank}")
     search_method = config._SEARCH_METHOD
     fetch_k = config._FETCH_K
 

sample-applications/chat-question-and-answer-core/app/config.py

@@ -3,13 +3,14 @@
 from typing import Union
 from os.path import dirname, abspath
 from .prompt import get_prompt_template
+from .runtime_validators import OpenVINOValidator, OllamaValidator
 import os
 import yaml
 
 class Settings(BaseSettings):
     """
     Settings class for configuring the Chatqna-Core application.
-    This class manages application settings, including model backend selection,
+    This class manages application settings, including model backend runtime selection,
     model IDs, device configurations, prompt templates, and various internal paths.
     It loads configuration from a YAML file, validates backend-specific requirements,
     and ensures prompt templates contain required placeholders.
@@ -20,7 +21,7 @@ class Settings(BaseSettings):
         SUPPORTED_FORMATS (set): Supported file formats for input documents.
         DEBUG (bool): Debug mode flag.
         HF_ACCESS_TOKEN (str): Hugging Face access token.
-        MODEL_BACKEND (str): Backend to use for models ('openvino' or 'ollama').
+        MODEL_RUNTIME (str): Backend runtime to use for models ('openvino' or 'ollama').
         EMBEDDING_MODEL_ID (str): Identifier for the embedding model.
         RERANKER_MODEL_ID (str): Identifier for the reranker model.
         LLM_MODEL_ID (str): Identifier for the large language model.
@@ -56,7 +57,7 @@ class Settings(BaseSettings):
     DEBUG: bool = False
 
     HF_ACCESS_TOKEN: str = ""
-    MODEL_BACKEND: str = ""
+    MODEL_RUNTIME: str = ""
     EMBEDDING_MODEL_ID: str = ""
     RERANKER_MODEL_ID: str = ""
     LLM_MODEL_ID: str = ""
@@ -98,58 +99,23 @@ def __init__(self, **kwargs):
                 if hasattr(self, key):
                     setattr(self, key, value)
 
-        self._validate_backend_settings()
+        self._validate_runtime_settings()
         self._check_and_validate_prompt_template()
 
+    def _validate_runtime_settings(self):
+        validators = {
+            "openvino": OpenVINOValidator,
+            "ollama": OllamaValidator,
+        }
 
-    def _validate_backend_settings(self):
-        if self.MODEL_BACKEND:
-            self.MODEL_BACKEND = self.MODEL_BACKEND.lower()
-        else:
-            raise ValueError("MODEL_BACKEND must not be an empty string.")
-
-        if self.MODEL_BACKEND == "openvino":
-            self._ENABLE_RERANK = True
-
-            # Validate Huggingface token
-            if not self.HF_ACCESS_TOKEN:
-                raise ValueError("HF_ACCESS_TOKEN must not be an empty string for 'openvino' backend.")
-
-            # Validate required model IDs
-            for model_name in ["EMBEDDING_MODEL_ID", "RERANKER_MODEL_ID", "LLM_MODEL_ID"]:
-                model_id = getattr(self, model_name)
-                if not model_id:
-                    raise ValueError(f"{model_name} must not be an empty string for 'openvino' backend.")
-
-        elif self.MODEL_BACKEND == "ollama":
-            self._ENABLE_RERANK = False
-
-            # Validate that all devices are set to "CPU" as ollama currently only enabled for CPU
-            invalid_devices = [
-                attr for attr in ["EMBEDDING_DEVICE", "RERANKER_DEVICE", "LLM_DEVICE"]
-                if getattr(self, attr, "") != "CPU"
-            ]
-
-            if invalid_devices:
-                raise ValueError(
-                    f"When MODEL_BACKEND is 'ollama', the following devices must be set to 'CPU': {', '.join(invalid_devices)}"
-                )
-
-            # Handle RERANKER_MODEL_ID
-            if self.RERANKER_MODEL_ID:
-                print("WARNING - RERANKER_MODEL_ID is ignored when MODEL_BACKEND is 'ollama'. Setting it to empty.")
-                self.RERANKER_MODEL_ID = ""
-            else:
-                print("INFO - MODEL_BACKEND is 'ollama'. Reranker model is not supported.")
-
-            # Validate required model IDs (excluding reranker)
-            for model_name in ["EMBEDDING_MODEL_ID", "LLM_MODEL_ID"]:
-                model_id = getattr(self, model_name)
-                if not model_id:
-                    raise ValueError(f"{model_name} must not be an empty string for 'ollama' backend.")
+        runtime = self.MODEL_RUNTIME.lower()
+        validator_cls = validators.get(runtime)
 
-        else:
-            raise ValueError(f"Unsupported MODEL_BACKEND '{self.MODEL_BACKEND}'. Only 'openvino' and 'ollama' are supported.")
+        if not validator_cls:
+            raise ValueError(f"Unsupported model runtime: {self.MODEL_RUNTIME}. Supported runtimes are: {', '.join(validators.keys())}")
+
+        validator = validator_cls(self)
+        validator.validate()
 
     def _check_and_validate_prompt_template(self):
         if not self.PROMPT_TEMPLATE:

sample-applications/chat-question-and-answer-core/app/runtime_validators.py

@@ -0,0 +1,50 @@
+from typing import TYPE_CHECKING
+
+if TYPE_CHECKING:
+    # To avoid circular imports
+    from config import Settings
+
+class BackendValidator:
+    def __init__(self, settings: "Settings"):
+        self.settings = settings
+
+    def validate(self):
+        raise NotImplementedError("Subclasses must implement this method.")
+
+
+class OpenVINOValidator(BackendValidator):
+    def validate(self):
+        self.settings.MODEL_RUNTIME = "openvino"
+        if not self.settings.HF_ACCESS_TOKEN:
+            raise ValueError("HF_ACCESS_TOKEN must not be empty for 'openvino' backend.")
+
+        for model_name in ["EMBEDDING_MODEL_ID", "RERANKER_MODEL_ID", "LLM_MODEL_ID"]:
+            if not getattr(self.settings, model_name):
+                raise ValueError(f"{model_name} must not be empty for 'openvino' backend.")
+
+        self.settings._ENABLE_RERANK = True
+
+
+class OllamaValidator(BackendValidator):
+    def validate(self):
+        self.settings.MODEL_RUNTIME = "ollama"
+        invalid_devices = [
+            attr for attr in ["EMBEDDING_DEVICE", "RERANKER_DEVICE", "LLM_DEVICE"]
+            if getattr(self.settings, attr) != "CPU"
+        ]
+        if invalid_devices:
+            raise ValueError(
+                f"When MODEL_RUNTIME is 'ollama', the following devices must be set to 'CPU': {', '.join(invalid_devices)}"
+            )
+
+        if self.settings.RERANKER_MODEL_ID:
+            print("WARNING - RERANKER_MODEL_ID is ignored for 'ollama'. Setting it to empty.")
+            self.settings.RERANKER_MODEL_ID = ""
+        else:
+            print("INFO - Reranker model not supported for 'ollama'.")
+
+        for model_name in ["EMBEDDING_MODEL_ID", "LLM_MODEL_ID"]:
+            if not getattr(self.settings, model_name):
+                raise ValueError(f"{model_name} must not be empty for 'ollama' backend.")
+
+        self.settings._ENABLE_RERANK = False

sample-applications/chat-question-and-answer-core/app/server.py

@@ -47,11 +47,11 @@ class ChatRequest(BaseModel):
 
 
 # Conditionally include OpenVINO routes
-if config.MODEL_BACKEND == "openvino":
+if config.MODEL_RUNTIME == "openvino":
     from .openvino_routes import router as openvino_router
     app.include_router(openvino_router)
 
-elif config.MODEL_BACKEND == "ollama":
+elif config.MODEL_RUNTIME == "ollama":
     from .ollama_routes import router as ollama_router
     app.include_router(ollama_router)
 

sample-applications/chat-question-and-answer-core/chart/templates/_helpers.tpl

@@ -27,10 +27,10 @@ Define the name for nginx Chart.
 {{- end }}
 
 {{- define "chatqna-core.validateGpuSettings" -}}
-{{- $backend := .Values.global.MODEL_BACKEND | lower }}
+{{- $backend := .Values.global.MODEL_RUNTIME | lower }}
 {{- if eq $backend "ollama" }}
   {{- if or (eq .Values.global.EMBEDDING_DEVICE "GPU") (eq .Values.global.LLM_DEVICE "GPU") }}
-    {{- fail "MODEL_BACKEND is set to 'ollama', but EMBEDDING_DEVICE or LLM_DEVICE is set to 'GPU'. Ollama backend only supports CPU devices." }}
+    {{- fail "MODEL_RUNTIME is set to 'ollama', but EMBEDDING_DEVICE or LLM_DEVICE is set to 'GPU'. Ollama backend only supports CPU devices." }}
   {{- end }}
 
 {{- if and (not .Values.gpu.enabled) (or (eq .Values.global.EMBEDDING_DEVICE "GPU") (eq .Values.global.RERANKER_DEVICE "GPU") (eq .Values.global.LLM_DEVICE "GPU")) }}

sample-applications/chat-question-and-answer-core/chart/templates/configmap.yaml

@@ -7,7 +7,7 @@ metadata:
 data:
   config.yaml: |
     model_settings:
-      MODEL_BACKEND: "{{ .Values.global.MODEL_BACKEND }}"
+      MODEL_RUNTIME: "{{ .Values.global.MODEL_RUNTIME }}"
       EMBEDDING_MODEL_ID: "{{ .Values.global.EMBEDDING_MODEL }}"
       LLM_MODEL_ID: "{{ .Values.global.LLM_MODEL }}"
       MAX_TOKENS: {{ .Values.global.MAX_TOKENS }}
@@ -16,15 +16,15 @@ data:
       {{ .Values.global.PROMPT_TEMPLATE | nindent 8 }}
       {{- end }}
 
-      {{- if and (ne (.Values.global.MODEL_BACKEND  | lower) "ollama") (.Values.global.RERANKER_MODEL) }}
+      {{- if and (ne (.Values.global.MODEL_RUNTIME  | lower) "ollama") (.Values.global.RERANKER_MODEL) }}
       RERANKER_MODEL_ID: "{{ .Values.global.RERANKER_MODEL }}"
       {{- end }}
 
-      {{- if eq (.Values.global.MODEL_BACKEND | lower) "ollama" }}
+      {{- if eq (.Values.global.MODEL_RUNTIME | lower) "ollama" }}
       KEEP_ALIVE: {{ .Values.global.KEEP_ALIVE }}
       {{- end }}
 
-    {{- if ne (.Values.global.MODEL_BACKEND | lower) "ollama" }}
+    {{- if ne (.Values.global.MODEL_RUNTIME | lower) "ollama" }}
     device_settings:
       EMBEDDING_DEVICE: "{{ .Values.global.EMBEDDING_DEVICE }}"
       LLM_DEVICE: "{{ .Values.global.LLM_DEVICE }}"

sample-applications/chat-question-and-answer-core/chart/templates/deployment.yaml

@@ -20,7 +20,7 @@ spec:
         fsGroup: 1000
       containers:
         - name: chatqna-core
-          {{- if eq (.Values.global.MODEL_BACKEND | lower) "ollama" }}
+          {{- if eq (.Values.global.MODEL_RUNTIME | lower) "ollama" }}
           image: "{{ .Values.image.registry }}chatqna:{{ .Values.image.tags.ollama }}"
           {{- else }}
           image: "{{ .Values.image.registry }}chatqna:{{ if .Values.gpu.enabled }}{{ .Values.image.tags.openvinoGPU }}{{ else }}{{ .Values.image.tags.openvinoCPU }}{{ end }}"
@@ -41,7 +41,7 @@ spec:
               value: "{{ .Values.global.https_proxy }}"
             - name: no_proxy
               value: "{{ .Values.global.no_proxy }},127.0.0.1"
-            {{- if eq (.Values.global.MODEL_BACKEND | lower) "openvino" }}
+            {{- if eq (.Values.global.MODEL_RUNTIME | lower) "openvino" }}
             - name: HF_ACCESS_TOKEN
               value: "{{ .Values.global.huggingface.apiToken }}"
             {{- end }}

sample-applications/chat-question-and-answer-core/chart/values-ollama.yaml

@@ -1,5 +1,5 @@
 global:
-  MODEL_BACKEND: "ollama"
+  MODEL_RUNTIME: "ollama"
   EMBEDDING_MODEL:
   LLM_MODEL:
   KEEP_ALIVE: -1

sample-applications/chat-question-and-answer-core/chart/values-openvino.yaml

@@ -1,7 +1,7 @@
 global:
   huggingface:
     apiToken:
-  MODEL_BACKEND: "openvino"
+  MODEL_RUNTIME: "openvino"
   EMBEDDING_MODEL:
   LLM_MODEL:
   RERANKER_MODEL:

sample-applications/chat-question-and-answer-core/model_config/ollama/llama3.1-config.yaml

@@ -1,5 +1,5 @@
 model_settings:
-  MODEL_BACKEND: "ollama"
+  MODEL_RUNTIME: "ollama"
   EMBEDDING_MODEL_ID: "mxbai-embed-large"
   LLM_MODEL_ID: "llama3.1"
   KEEP_ALIVE: -1