fix: move test_huggingface.py to granite4.1; and small rag intrinsic …

docs/dev/intrinsics_and_adapters.md

@@ -13,7 +13,7 @@ These changes are specified by the Adapter that corresponds to a given Intrinsic
 
 ## Parts of an Intrinsic
 Intrinsics specify:
-- an adapter name (ie requirement_check)
+- an adapter name (ie requirement-check)
 - types of adapters suitable to be used (ie alora)
 - any kwargs necessary (ie a requirement like "make sure the last user message is...")
 

docs/dev/requirement_aLoRA_rerouting.md

@@ -14,7 +14,7 @@ The actual rule is slightly more complicated.
 
 ## The Actual Rule
 
-If a `Requirement` is validated using a backend that could either use a `requirement_check` aLoRA or perform an LLMaJ prompt on the underlying model, then the aLoRA is used for validation, even if the `backend.generate_from_context` method is called instead of the `backend._generate_from_intrinsic` method.
+If a `Requirement` is validated using a backend that could either use a `requirement-check` aLoRA or perform an LLMaJ prompt on the underlying model, then the aLoRA is used for validation, even if the `backend.generate_from_context` method is called instead of the `backend._generate_from_intrinsic` method.
 
 There are three exceptions to this rule:
 1. `Backend.default_to_constraint_checking_alora` is set to `False` (this parameter defaults to `True`).
@@ -39,8 +39,8 @@ from mellea.backends.adapters import IntrinsicAdapter
 m = start_session(
     "huggingface.LocalHFBackend:ibm-granite/granite-4.0-micro")
 
-# By default, the AloraRequirement uses a IntrinsicAdapter with "requirement_check".
-m.backend.add_adapter(IntrinsicAdapter("ibm-granite/rag-intrinsics-lib", "requirement_check", base_model_name="granite-4.0-micro"))
+# By default, the AloraRequirement uses a IntrinsicAdapter with "requirement-check".
+m.backend.add_adapter(IntrinsicAdapter("ibm-granite/rag-intrinsics-lib", "requirement-check", base_model_name="granite-4.0-micro"))
 
 m.instruct(
     "Corporate wants you to find the difference between these two strings:\n\naaa\naba")

docs/docs/advanced/intrinsics.md

@@ -226,7 +226,7 @@ backend = LocalHFBackend(model_id="ibm-granite/granite-4.0-micro")
 
 # Register an adapter by task name
 req_adapter = CustomIntrinsicAdapter(
-    "requirement_check",
+    "requirement-check",
     base_model_name=backend.base_model_name,
 )
 backend.add_adapter(req_adapter)
@@ -237,7 +237,7 @@ ctx = ctx.add(Message("assistant", "Yes! What can I help with?"))
 
 out, _ = mfuncs.act(
     Intrinsic(
-        "requirement_check",
+        "requirement-check",
         intrinsic_kwargs={"requirement": "The assistant is helpful."},
     ),
     ctx,
@@ -249,4 +249,4 @@ print(out)  # {"requirement_likelihood": 1.0}
 The `Intrinsic` component loads aLoRA adapters (falling back to LoRA) by task name.
 For OpenAI backends with Granite Switch, adapters are loaded from the model's
 HuggingFace repository configuration instead of the intrinsic catalog.
-Output format is task-specific — `requirement_check` returns a likelihood score.
+Output format is task-specific — `requirement-check` returns a likelihood score.

docs/examples/intrinsics/README.md

@@ -47,7 +47,7 @@ import mellea.stdlib.functional as mfuncs
 
 out, new_ctx = mfuncs.act(
     Intrinsic(
-        "requirement_check",
+        "requirement-check",
         intrinsic_kwargs={"requirement": "The assistant is helpful."}),
     ctx,
     backend

mellea/backends/huggingface.py

@@ -404,7 +404,7 @@ async def _generate_from_context(
             if isinstance(action, Requirement):
                 # See docs/dev/requirement_aLoRA_rerouting.md
                 reroute_to_alora = self.default_to_constraint_checking_alora
-                adapter_name = "requirement_check"
+                adapter_name = "requirement-check"
 
                 if isinstance(action, ALoraRequirement):
                     reroute_to_alora = True
@@ -416,7 +416,7 @@ async def _generate_from_context(
                     )
                     alora_action = ALoraRequirement(action.description, adapter_name)
 
-                # Check if a requirement_check (or AloraRequirement specified) adapter
+                # Check if a requirement-check (or AloraRequirement specified) adapter
                 # exists.
                 alora_req_adapter = get_adapter_for_intrinsic(
                     adapter_name, [AdapterType.ALORA], self._added_adapters

mellea/backends/openai.py

@@ -481,7 +481,7 @@ async def _generate_from_context(
             # Requirements can be automatically rerouted to a requirement adapter.
             if isinstance(action, Requirement):
                 reroute_to_alora = self.default_to_constraint_checking_alora
-                adapter_name = "requirement_check"
+                adapter_name = "requirement-check"
 
                 if isinstance(action, ALoraRequirement):
                     reroute_to_alora = True

mellea/stdlib/components/intrinsic/rag.py

@@ -198,7 +198,7 @@ def flag_hallucinated_content(
     documents: collections.abc.Iterable[str | Document],
     context: ChatContext,
     backend: AdapterMixin,
-) -> float:
+) -> list[dict]:
     """Flag potentially-hallucinated sentences in an agent's response.
 
     Intrinsic function that checks whether the sentences in an agent's response to a

mellea/stdlib/requirements/requirement.py

@@ -35,14 +35,21 @@ def requirement_check_to_bool(x: CBlock | str) -> bool:
     output = str(x)
     req_dict: dict[str, Any] = json.loads(output)
 
-    likelihood = req_dict.get("requirement_likelihood", None)
-    if likelihood is None:
+    likelihood = req_dict.get("requirement_check", None)
+    if not isinstance(likelihood, dict):
         MelleaLogger.get_logger().warning(
-            f"could not get value from alora requirement output; looking for `requirement_likelihood` in {req_dict}"
+            f"could not get value from alora requirement output; looking for `requirement_check` in {req_dict}"
         )
         return False
 
-    if likelihood > 0.5:
+    score = likelihood.get("score", None)
+    if score is None:
+        MelleaLogger.get_logger().warning(
+            f"could not get value from alora requirement output; looking for `score` in {req_dict}"
+        )
+        return False
+
+    if score > 0.5:
         return True
 
     return False
@@ -57,7 +64,7 @@ class ALoraRequirement(Requirement, Intrinsic):
     Args:
         description (str): Human-readable requirement description.
         intrinsic_name (str | None): Name of the ALoRA intrinsic to use.
-            Defaults to ``"requirement_check"``.
+            Defaults to ``"requirement-check"``.
 
     Attributes:
         use_aloras (bool): Always ``True``; this class always attempts to use
@@ -73,7 +80,7 @@ def __init__(self, description: str, intrinsic_name: str | None = None):
         self.use_aloras: bool = True
 
         if intrinsic_name is None:
-            intrinsic_name = "requirement_check"
+            intrinsic_name = "requirement-check"
 
         # Initialize the other side of the inheritance tree
         Intrinsic.__init__(

test/backends/test_huggingface.py

@@ -54,17 +54,15 @@ def backend():
     """Shared HuggingFace backend for all tests in this module.
 
     Uses Granite 3.3-8b for aLoRA adapter compatibility.
-    The "requirement_check" intrinsic only has adapters for Granite 3.3 models.
+    The "requirement-check" intrinsic only has adapters for Granite 3.3 models.
     Granite 4 adapters are not yet available.
     Other intrinsics are not affected by this issue.
     """
     backend = LocalHFBackend(
-        model_id="ibm-granite/granite-3.3-8b-instruct",
-        formatter=TemplateFormatter(model_id="ibm-granite/granite-4.0-tiny-preview"),
-        cache=SimpleLRUCache(5),
+        model_id=model_ids.IBM_GRANITE_4_1_3B, cache=SimpleLRUCache(5)
     )
     backend.add_adapter(
-        IntrinsicAdapter("requirement_check", base_model_name=backend.base_model_name)
+        IntrinsicAdapter("requirement-check", base_model_name=backend.base_model_name)
     )
     backend.add_adapter(
         IntrinsicAdapter("answerability", base_model_name=backend.base_model_name)
@@ -88,7 +86,7 @@ def session(backend):
 def test_adapters(backend) -> None:
     assert len(backend._added_adapters.items()) > 0
 
-    expected_qualified_name = "requirement_check_alora"
+    expected_qualified_name = "requirement-check_alora"
     adapter = backend._added_adapters[expected_qualified_name]
     backend.load_adapter(adapter.qualified_name)
     assert adapter.qualified_name in backend._loaded_adapters
@@ -125,7 +123,7 @@ def test_constraint_lora_with_requirement(session, backend) -> None:
     assert len(validation_outputs) == 1
     val_result = validation_outputs[0]
     assert isinstance(val_result, ValidationResult)
-    assert "requirement_likelihood" in str(val_result.reason)
+    assert "requirement_check" in str(val_result.reason)
 
 
 @pytest.mark.qualitative
@@ -158,7 +156,7 @@ def test_constraint_lora_override_does_not_override_alora(session, backend) -> N
     assert len(validation_outputs) == 1
     val_result = validation_outputs[0]
     assert isinstance(val_result, ValidationResult)
-    assert "requirement_likelihood" in str(val_result.reason)
+    assert "requirement_check" in str(val_result.reason)
 
     # Ensure the ValidationResult has its thunk and context set. Ensure the context has
     # the correct actions / results in it.
@@ -366,7 +364,7 @@ async def test_generate_with_lock(backend) -> None:
     b._added_adapters = {}
     b._loaded_adapters = {}
     b.add_adapter(
-        IntrinsicAdapter("requirement_check", base_model_name=b.base_model_name)
+        IntrinsicAdapter("requirement-check", base_model_name=b.base_model_name)
     )
     b.add_adapter(IntrinsicAdapter("answerability", base_model_name=b.base_model_name))
 
@@ -397,7 +395,7 @@ def populate_mocked_dict(input_ids, *args, **kwargs):
     ctx = ChatContext().add(Message("user", "hello"))
     act = CBlock("hello")
     raw_act = CBlock("goodb")
-    req_intrinsic = Intrinsic("requirement_check", {"requirement": "did nothing"})
+    req_intrinsic = Intrinsic("requirement-check", {"requirement": "did nothing"})
     answerability_intrinsic = Intrinsic("answerability")
 
     def call_backend_generate():
@@ -462,7 +460,7 @@ async def test_generate_with_lock_does_not_block_when_awaiting_value(backend) ->
     # Set up the inputs.
     ctx = ChatContext().add(Message("user", "hello"))
     act = CBlock("hello")
-    req_intrinsic = Intrinsic("requirement_check", {"requirement": "did nothing"})
+    req_intrinsic = Intrinsic("requirement-check", {"requirement": "did nothing"})
     answerability_intrinsic = Intrinsic("answerability")
 
     # Create a few model output thunks:
@@ -516,7 +514,7 @@ async def test_generate_with_lock_does_not_block_when_awaiting_value(backend) ->
 @pytest.mark.qualitative
 async def test_streaming_error_with_intrinsics(backend) -> None:
     ctx = ChatContext().add(Message("user", "hello"))
-    req_intrinsic = Intrinsic("requirement_check", {"requirement": "did nothing"})
+    req_intrinsic = Intrinsic("requirement-check", {"requirement": "did nothing"})
 
     with pytest.raises(Exception, match="Intrinsics do not support streaming"):
         _, _ = await backend.generate_from_context(
@@ -540,7 +538,7 @@ async def test_error_during_generate_with_lock(backend) -> None:
     b._added_adapters = {}
     b._loaded_adapters = {}
     b.add_adapter(
-        IntrinsicAdapter("requirement_check", base_model_name=b.base_model_name)
+        IntrinsicAdapter("requirement-check", base_model_name=b.base_model_name)
     )
 
     regular_generate = b._model.generate
@@ -559,7 +557,7 @@ def generate_and_raise_exc(*args, **kwargs):
     # Set up the inputs.
     ctx = ChatContext().add(Message("user", "hello"))
     act = CBlock("hello")
-    req_intrinsic = Intrinsic("requirement_check", {"requirement": "did nothing"})
+    req_intrinsic = Intrinsic("requirement-check", {"requirement": "did nothing"})
 
     reg_mot, _ = await b.generate_from_context(act, ctx)
     req_mot, _ = await b.generate_from_context(req_intrinsic, ctx)
@@ -753,7 +751,7 @@ def get_temperature(location: str) -> int:
     decoded = backend._tokenizer.decode(input_tokens[0])
     print(decoded)
     assert "get_temperature" in decoded
-    assert "available_tools" in decoded, (
+    assert "access to the following tools" in decoded, (
         "expected string from system prompt with tool calls not found; if you changed the model, that might have caused this issue"
     )
 

test/stdlib/requirements/test_requirement.py

@@ -76,16 +76,16 @@ def test_simple_validate_invalid():
 
 
 def test_requirement_check_to_bool_above_threshold():
-    assert requirement_check_to_bool('{"requirement_likelihood": 0.8}') is True
+    assert requirement_check_to_bool('{"requirement_check": {"score": 0.8}}') is True
 
 
 def test_requirement_check_to_bool_below_threshold():
-    assert requirement_check_to_bool('{"requirement_likelihood": 0.3}') is False
+    assert requirement_check_to_bool('{"requirement_check": {"score":0.3}}') is False
 
 
 def test_requirement_check_to_bool_at_threshold():
     """0.5 is NOT > 0.5, so should return False."""
-    assert requirement_check_to_bool('{"requirement_likelihood": 0.5}') is False
+    assert requirement_check_to_bool('{"requirement_check": {"score": 0.5}}') is False
 
 
 def test_requirement_check_to_bool_missing_key():
@@ -163,7 +163,7 @@ def test_alora_requirement_default_intrinsic(mock_intrinsic_init):
     # Intrinsic.__init__ is unbound; mock receives self as first positional arg.
     mock_intrinsic_init.assert_called_once_with(
         r,
-        intrinsic_name="requirement_check",
+        intrinsic_name="requirement-check",
         intrinsic_kwargs={"requirement": "must be valid"},
     )