demonstrate using selfhosted ollama as Eval Model. Arize-ai#2280

packages/phoenix-evals/src/phoenix/evals/models/litellm.py

@@ -65,7 +65,7 @@ def _init_environment(self) -> None:
             from litellm import validate_environment
 
             self._litellm = litellm
-            env_info = validate_environment(self._litellm.utils.get_llm_provider(self.model))
+            env_info = validate_environment(self.model)
 
             if not env_info["keys_in_environment"]:
                 raise RuntimeError(

packages/phoenix-evals/tests/phoenix/evals/models/test_ollama.py

@@ -0,0 +1,43 @@
+import os
+import unittest
+from unittest import mock
+
+from models import LiteLLMModel
+
+@mock.patch.dict(os.environ, {"OLLAMA_API_BASE": "just to make litellm.validate_environment happy"}, clear=True)
+@mock.patch("litellm.llms.ollama.get_ollama_response")
+def test_selfhosted_ollama_via_model_kwargs(get_ollama_response):
+    ollama_response = unittest.mock.MagicMock()
+    ollama_response["choices"][0]["message"]["content"] = "barely understand Python mocks"
+    ollama_response.choices[0].message.content = "42 per tail"
+
+    get_ollama_response.return_value = ollama_response
+
+    lllmm = LiteLLMModel(model="ollama/monstral",
+                         model_kwargs=dict(
+                             base_url="http://hosted.olla.ma:11434"))
+    result = lllmm("How much is the fish?")
+
+    assert result == "42 per tail"
+    call_args = get_ollama_response.call_args[0]
+    assert call_args[0] == "http://hosted.olla.ma:11434"
+    assert call_args[1] == "monstral"
+    assert "How much is the fish?" in call_args[2]
+
+@mock.patch.dict(os.environ, {"OLLAMA_API_BASE": "http://hosted.olla.ma:11434"}, clear=True)
+@mock.patch("litellm.llms.ollama.get_ollama_response")
+def test_selfhosted_ollama_via_env(get_ollama_response):
+    ollama_response = unittest.mock.MagicMock()
+    ollama_response["choices"][0]["message"]["content"] = "barely understand Python mocks"
+    ollama_response.choices[0].message.content = "42 per tail"
+
+    get_ollama_response.return_value = ollama_response
+
+    lllmm = LiteLLMModel(model="ollama/monstral")
+    result = lllmm("How much is the fish?")
+
+    assert result == "42 per tail"
+    call_args = get_ollama_response.call_args[0]
+    assert call_args[0] == "http://hosted.olla.ma:11434"
+    assert call_args[1] == "monstral"
+    assert "How much is the fish?" in call_args[2]