fix: collection of small fixes (#238)

* fix: add details to litellm init for model_id

* fix: litellm ollama test options

* fix: remove LLMaJ template; uses Requirement template

Author: jakelorocco

mellea/backends/litellm.py

@@ -54,12 +54,12 @@ def __init__(
         base_url: str | None = "http://localhost:11434",
         model_options: dict | None = None,
     ):
-        """Initialize and OpenAI compatible backend. For any additional kwargs that you need to pass the the client, pass them as a part of **kwargs.
+        """Initialize an OpenAI compatible backend using the [LiteLLM Python SDK](https://docs.litellm.ai/docs/#litellm-python-sdk).
 
         Note: If getting `Unclosed client session`, set `export DISABLE_AIOHTTP_TRANSPORT=True` in your environment. See: https://github.com/BerriAI/litellm/issues/13251.
 
         Args:
-            model_id : The LiteLLM model identifier. Make sure that all necessary credentials are in OS environment variables.
+            model_id : The LiteLLM model identifier; in most cases requires some combination of `<provider>/<model_creator>/<model_name>`. Make sure that all necessary credentials are in OS environment variables.
             formatter: A custom formatter based on backend.If None, defaults to TemplateFormatter
             base_url : Base url for LLM API. Defaults to None.
             model_options : Generation options to pass to the LLM. Defaults to None.

mellea/templates/prompts/default/LLMaJRequirement.jinja2

@@ -109,11 +109,6 @@ def test_litellm_ollama_instruct_options(session):
         ModelOption.SEED: 123,
         ModelOption.TEMPERATURE: 0.5,
         ModelOption.MAX_NEW_TOKENS: 100,
-        # Ollama thinking controls currently broken on Granite; see
-        # https://github.com/ollama/ollama/issues/10983
-        # TODO: Re-enable when this upstream bug gets fixed.
-        # ModelOption.THINKING: True,
-        # "reasoning_effort": True,
         "homer_simpson": "option should be kicked out",
     }
 

test/backends/test_litellm_ollama.py

@@ -1,7 +1,7 @@
 import asyncio
 import pytest
 from mellea.stdlib.base import ChatContext, ModelOutputThunk
-from mellea.stdlib.requirement import Requirement, simple_validate
+from mellea.stdlib.requirement import LLMaJRequirement, Requirement, simple_validate
 from mellea.stdlib.session import start_session
 
 ctx = ChatContext()
@@ -19,6 +19,17 @@ async def test_llmaj_validation_req_output_field():
     )
 
 
+async def test_llmaj_requirement_uses_requirement_template():
+    m = start_session(ctx=ctx)
+    req = LLMaJRequirement("Must output test.")
+    assert req._output is None
+
+    _ = await req.validate(m.backend, ctx=ctx)
+    assert req._output is None, (
+        "requirement's output shouldn't be updated during/after validation"
+    )
+
+
 def test_simple_validate_bool():
     validation_func = simple_validate(lambda x: False, reason="static reason")
     val_result = validation_func(ctx)