fix(prompts): prevent IndexError when LLM provided via constructor with empty models config (#1334)

* fix(prompts): prevent IndexError when LLM provided via constructor with empty models config

- Add check in get_task_model to handle empty _models list gracefully
- Return None instead of throwing IndexError when no models match
- Add comprehensive test coverage for various model configuration scenarios

Fixes the issue where providing an LLM object directly to LLMRails constructor
would fail if the YAML config had an empty models list.

Author: Pouyanpi

nemoguardrails/llm/prompts.py

@@ -16,7 +16,7 @@
 """Prompts for the various steps in the interaction."""
 
 import os
-from typing import List, Union
+from typing import List, Optional, Union
 
 import yaml
 
@@ -129,7 +129,7 @@ def _get_prompt(
     raise ValueError(f"Could not find prompt for task {task_name} and model {model}")
 
 
-def get_task_model(config: RailsConfig, task: Union[str, Task]) -> Model:
+def get_task_model(config: RailsConfig, task: Union[str, Task]) -> Optional[Model]:
     """Return the model for the given task in the current config."""
     # Fetch current task parameters like name, models to use, and the prompting mode
     task_name = str(task.value) if isinstance(task, Task) else task
@@ -139,7 +139,8 @@ def get_task_model(config: RailsConfig, task: Union[str, Task]) -> Model:
         if not _models:
             _models = [model for model in config.models if model.type == "main"]
 
-        return _models[0]
+        if _models:
+            return _models[0]
 
     return None
 

tests/test_llm_task_manager.py

@@ -20,7 +20,7 @@
 
 from nemoguardrails import RailsConfig
 from nemoguardrails.llm.filters import conversation_to_events
-from nemoguardrails.llm.prompts import get_prompt
+from nemoguardrails.llm.prompts import get_prompt, get_task_model
 from nemoguardrails.llm.taskmanager import LLMTaskManager
 from nemoguardrails.llm.types import Task
 
@@ -457,3 +457,78 @@ def test_reasoning_traces_not_included_in_prompt_history():
         "Hi there!" in rendered_prompt
         or "I don't have access to real-time weather information." in rendered_prompt
     )
+
+
+def test_get_task_model_with_empty_models():
+    """Test that get_task_model returns None when models list is empty.
+
+    This tests the fix for the IndexError that occurred when the models list was empty.
+    """
+    config = RailsConfig.parse_object({"models": []})
+
+    result = get_task_model(config, "main")
+    assert result is None
+
+    result = get_task_model(config, Task.GENERAL)
+    assert result is None
+
+
+def test_get_task_model_with_no_matching_models():
+    """Test that get_task_model returns None when no models match the requested type."""
+    config = RailsConfig.parse_object(
+        {
+            "models": [
+                {
+                    "type": "embeddings",
+                    "engine": "openai",
+                    "model": "text-embedding-ada-002",
+                }
+            ]
+        }
+    )
+
+    result = get_task_model(config, "main")
+    assert result is None
+
+
+def test_get_task_model_with_main_model():
+    """Test that get_task_model returns the main model when present."""
+    config = RailsConfig.parse_object(
+        {
+            "models": [
+                {
+                    "type": "embeddings",
+                    "engine": "openai",
+                    "model": "text-embedding-ada-002",
+                },
+                {
+                    "type": "custom_task",
+                    "engine": "anthropic",
+                    "model": "claude-4.1-opus",
+                },
+                {
+                    "type": "fact_checking",
+                    "engine": "openai",
+                    "model": "gpt-4",
+                },
+                {"type": "main", "engine": "openai", "model": "gpt-3.5-turbo"},
+            ]
+        }
+    )
+
+    result = get_task_model(config, "main")
+    assert result is not None
+    assert result.type == "main"
+    assert result.engine == "openai"
+    assert result.model == "gpt-3.5-turbo"
+
+
+def test_get_task_model_fallback_to_main():
+    """Test that get_task_model falls back to main model when specific task model not found."""
+    config = RailsConfig.parse_object(
+        {"models": [{"type": "main", "engine": "openai", "model": "gpt-3.5-turbo"}]}
+    )
+
+    result = get_task_model(config, "some_other_task")
+    assert result is not None
+    assert result.type == "main"

tests/test_llmrails.py

@@ -856,6 +856,42 @@ async def test_other_models_honored(mock_init, llm_config_with_multiple_models):
     assert any(event.get("intent") == "express greeting" for event in new_events)
 
 
+@pytest.mark.asyncio
+async def test_llm_constructor_with_empty_models_config():
+    """Test that LLMRails can be initialized with constructor LLM when config has empty models list.
+
+    This tests the fix for the IndexError that occurred when providing an LLM via constructor
+    but having an empty models list in the config.
+    """
+    config = RailsConfig.parse_object(
+        {
+            "models": [],
+            "user_messages": {
+                "express greeting": ["Hello!"],
+            },
+            "flows": [
+                {
+                    "elements": [
+                        {"user": "express greeting"},
+                        {"bot": "express greeting"},
+                    ]
+                },
+            ],
+            "bot_messages": {
+                "express greeting": ["Hello! How are you?"],
+            },
+        }
+    )
+
+    injected_llm = FakeLLM(responses=["express greeting"])
+    llm_rails = LLMRails(config=config, llm=injected_llm)
+    assert llm_rails.llm == injected_llm
+
+    events = [{"type": "UtteranceUserActionFinished", "final_transcript": "Hello!"}]
+    new_events = await llm_rails.runtime.generate_events(events)
+    assert any(event.get("intent") == "express greeting" for event in new_events)
+
+
 @pytest.mark.asyncio
 @patch(
     "nemoguardrails.rails.llm.llmrails.init_llm_model",