feat: make the model names configurable

Author: IgnacioHeredia

README.md

@@ -27,6 +27,9 @@ You should add an `.env` file in the root folder with the following contents:
 OPENAI_API_KEY=<your-llm-api-key>  # token for the AI4EOSC LLM
 GRADIO_PASSWORD=<password-for-gradio-ui>  # password for Gradio UI (user: admin)
 IS_DEV=True  # shows the DEV UI with the finegrained configuration
+LLM_ENDPOINT="https://vllm.cloud.ai4eosc.eu"
+LLM_EMBEDDINGS="AI4EOSC/Qwen/Qwen3-Embedding-4B"
+LLM_CHAT="AI4EOSC/mistralai/Mistral-Small-3.1-24B-Instruct-2503"
 ```
 
 ## Usage

docker/docker-compose.yml

@@ -4,7 +4,11 @@ services:
         image: "ignacioheredia/dome-copilot:latest"
         restart: always
         environment:
-            - OPENAI_API_KEY=*****************************
-            - GRADIO_PASSWORD=****************************
+            OPENAI_API_KEY: "*****************************"
+            GRADIO_PASSWORD: "****************************"
+            IS_DEV: False
+            LLM_ENDPOINT: "https://vllm.cloud.ai4eosc.eu"
+            LLM_EMBEDDINGS: "AI4EOSC/Qwen/Qwen3-Embedding-4B"
+            LLM_CHAT: "AI4EOSC/mistralai/Mistral-Small-3.1-24B-Instruct-2503"
         ports:
             - 5000:7860

dome_copilot/conf.py

@@ -2,17 +2,25 @@
 from pathlib import Path
 import yaml
 
-from jinja2 import Template
 from dotenv import load_dotenv
+from jinja2 import Template
+from openai import OpenAI
 
 
 main_path = Path(__file__).parents[1].absolute()
 load_dotenv(main_path / ".env")
 
-# Load envars
+# Load optional envars
+IS_DEV = os.environ.get("IS_DEV", "false").lower() == "true"
+LLM_ENDPOINT = os.environ.get("LLM_ENDPOINT", "https://vllm.cloud.ai4eosc.eu")
+LLM_EMBEDDINGS = os.environ.get("LLM_EMBEDDINGS", "AI4EOSC/Qwen/Qwen3-Embedding-4B")
+LLM_CHAT = os.environ.get(
+    "LLM_CHAT", "AI4EOSC/mistralai/Mistral-Small-3.1-24B-Instruct-2503"
+)
+
+# Load secrets
 OPENAI_API_KEY = os.environ.get("OPENAI_API_KEY")
 GRADIO_PASSWORD = os.environ.get("GRADIO_PASSWORD")
-IS_DEV = os.environ.get("IS_DEV", "false").lower() == "true"
 
 if not all([OPENAI_API_KEY, GRADIO_PASSWORD]):
     raise Exception("Please provide an OPENAI_API_KEY and GRADIO_PASSWORD.")
@@ -31,12 +39,21 @@ def load_jinja(path):
 pre_prompt = load_jinja(templates_dir / "prompt-preprocessing.jinja")
 post_prompt = load_jinja(templates_dir / "prompt-postprocessing.jinja")
 
-# DOME questionnaire
-
 
+# DOME questionnaire
 def load_yaml(path):
     with open(path) as stream:
         return yaml.safe_load(stream)
 
 
 dome = load_yaml(templates_dir / "dome-questionnaire.yml")
+
+# Check that the models exist, to catch typos
+openai_client = OpenAI(api_key=OPENAI_API_KEY, base_url=LLM_ENDPOINT)
+models = [model.id for model in openai_client.models.list().data]
+if LLM_EMBEDDINGS not in models:
+    raise Exception(
+        f"Invalid embeddings model name:\n'{LLM_EMBEDDINGS}' not in {models}"
+    )
+if LLM_CHAT not in models:
+    raise Exception(f"Invalid chat model name:\n'{LLM_CHAT}' not in {models}")

dome_copilot/inference.py

@@ -24,14 +24,14 @@
 
 # Initialize LLM and embedding model
 Settings.embed_model = OpenAILikeEmbedding(
-    api_base="https://vllm.cloud.ai4eosc.eu",
+    api_base=conf.LLM_ENDPOINT,
     api_key=conf.OPENAI_API_KEY,
-    model_name="AI4EOSC/Qwen3-Embedding",
+    model_name=conf.LLM_EMBEDDINGS,
 )
 Settings.llm = OpenAILike(
-    api_base="https://vllm.cloud.ai4eosc.eu",
+    api_base=conf.LLM_ENDPOINT,
     api_key=conf.OPENAI_API_KEY,
-    model="AI4EOSC/Small",
+    model=conf.LLM_CHAT,
     context_window=25000,
     is_chat_model=True,
     is_function_calling_model=False,
@@ -266,14 +266,11 @@ def postprocessing(annotations: dict, post_prompt: str = None):
     post_prompt = post_prompt.render()
 
     # Initialize the OpenAI client with a longer timeout
-    client = OpenAI(
-        api_key=conf.OPENAI_API_KEY,
-        base_url="https://vllm.cloud.ai4eosc.eu",
-    )
+    client = OpenAI(api_key=conf.OPENAI_API_KEY, base_url=conf.LLM_ENDPOINT)
 
     # Call the LLM (override timeout per request as well)
     response = client.chat.completions.create(
-        model="AI4EOSC/Small",
+        model=conf.LLM_CHAT,
         messages=[
             {"role": "system", "content": post_prompt},
             {"role": "user", "content": json.dumps(annotations)},

requirements.txt

@@ -2,6 +2,7 @@ python-dotenv>=1.2.2,<2.0
 llama-index==0.14.4
 llama-index-llms-openai-like==0.5.3
 llama-index-embeddings-openai-like==0.2.2
+openai>=1.109.1, <2.0
 gradio>5.49.1,<6.0
 docling>2.63.0,<3.0
 llama-index-readers-docling>=0.4.2