omnimancer-cli 0.3.2__tar.gz → 0.3.4__tar.gz

{omnimancer_cli-0.3.2 → omnimancer_cli-0.3.4}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: omnimancer-cli
-Version: 0.3.2
+Version: 0.3.4
 Summary: A unified command-line interface for multiple AI language models
 Author-email: Kellan Strong <kellan@jite.ai>
 License: MIT

{omnimancer_cli-0.3.2 → omnimancer_cli-0.3.4}/omnimancer/providers/bedrock.py

@@ -8,7 +8,7 @@ API key authentication, and region configuration.
 
 import json
 from datetime import datetime
-from typing import Any, Dict, List, Union
+from typing import Any, Dict, List, Optional, Union
 
 import httpx
 
@@ -70,10 +70,13 @@ class BedrockProvider(BaseProvider):
         # AWS-specific configuration
         self.aws_region = kwargs.get("aws_region", "us-east-1")
 
-        # Standard parameters
+        # Standard parameters. Bedrock fronts third-party models that reject
+        # temperature/topP in inferenceConfig outright, and nothing in
+        # Omnimancer lets a user pick these values, so an unset param stays
+        # None and is omitted rather than defaulted into every request.
         self.max_tokens = kwargs.get("max_tokens", 4096)
-        self.temperature = kwargs.get("temperature", 0.7)
-        self.top_p = kwargs.get("top_p", 1.0)
+        self.temperature = kwargs.get("temperature")
+        self.top_p = kwargs.get("top_p")
         self.top_k = kwargs.get("top_k", 250)
 
         # Build base URL for Bedrock API key authentication
@@ -258,7 +261,7 @@ class BedrockProvider(BaseProvider):
             # Test the specific model with a minimal request
             test_request = {
                 "messages": [{"role": "user", "content": [{"text": "Hi"}]}],
-                "inferenceConfig": {"maxTokens": 10, "temperature": 0.1},
+                "inferenceConfig": self._build_inference_config(max_tokens=10),
             }
 
             # If using ARN format, include the model ARN in the request
@@ -373,6 +376,23 @@ class BedrockProvider(BaseProvider):
                 "message": f"Failed to test model: {str(e)}",
             }
 
+    def _build_inference_config(
+        self, max_tokens: Optional[int] = None
+    ) -> Dict[str, Any]:
+        """Build inferenceConfig, omitting sampling params that are unset.
+
+        Converse raises a ValidationException for any inference field the
+        target model does not accept, so only configured values are sent.
+        """
+        inference: Dict[str, Any] = {
+            "maxTokens": max_tokens if max_tokens is not None else self.max_tokens
+        }
+        if self.temperature is not None:
+            inference["temperature"] = self.temperature
+        if self.top_p is not None:
+            inference["topP"] = self.top_p
+        return inference
+
     # Converse prompt caching is model-gated: sending a cachePoint to an
     # unsupported model is a hard ValidationException, so only inject for
     # families AWS documents as supported.
@@ -429,11 +449,7 @@ class BedrockProvider(BaseProvider):
 
         request_data = {
             "messages": messages,
-            "inferenceConfig": {
-                "maxTokens": self.max_tokens,
-                "temperature": self.temperature,
-                "topP": self.top_p,
-            },
+            "inferenceConfig": self._build_inference_config(),
         }
 
         # If using ARN format, include the model ARN in the request
@@ -470,11 +486,7 @@ class BedrockProvider(BaseProvider):
 
         request_data = {
             "messages": messages,
-            "inferenceConfig": {
-                "maxTokens": self.max_tokens,
-                "temperature": self.temperature,
-                "topP": self.top_p,
-            },
+            "inferenceConfig": self._build_inference_config(),
         }
 
         # Add tools in Bedrock Converse format

{omnimancer_cli-0.3.2 → omnimancer_cli-0.3.4}/omnimancer_cli.egg-info/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: omnimancer-cli
-Version: 0.3.2
+Version: 0.3.4
 Summary: A unified command-line interface for multiple AI language models
 Author-email: Kellan Strong <kellan@jite.ai>
 License: MIT

{omnimancer_cli-0.3.2 → omnimancer_cli-0.3.4}/pyproject.toml

@@ -4,7 +4,7 @@ build-backend = "setuptools.build_meta"
 
 [project]
 name = "omnimancer-cli"
-version = "0.3.2"
+version = "0.3.4"
 description = "A unified command-line interface for multiple AI language models"
 readme = "README.md"
 requires-python = ">=3.10"

{omnimancer_cli-0.3.2 → omnimancer_cli-0.3.4}/tests/test_bedrock_provider.py

@@ -218,12 +218,20 @@ class TestBedrockProviderInitialization:
         assert "eu-west-1" in provider.base_url
 
     def test_initialization_default_generation_params(self, bedrock_provider):
-        """Test default generation parameter values."""
+        """Unset sampling params stay None so the model's defaults apply."""
         assert bedrock_provider.max_tokens == 4096
-        assert bedrock_provider.temperature == 0.7
-        assert bedrock_provider.top_p == 1.0
+        assert bedrock_provider.temperature is None
+        assert bedrock_provider.top_p is None
         assert bedrock_provider.top_k == 250
 
+    def test_initialization_explicit_generation_params(self):
+        """Explicitly configured sampling params are kept."""
+        provider = BedrockProvider(
+            api_key="key", model=_SONNET_MODEL, temperature=0.2, top_p=0.9
+        )
+        assert provider.temperature == 0.2
+        assert provider.top_p == 0.9
+
     def test_initialization_missing_api_key_raises_error(self):
         """Test that missing api_key raises ValueError."""
         with pytest.raises(ValueError, match="API key is required"):
@@ -288,7 +296,59 @@ class TestBedrockProviderRequestPreparation:
 
         assert "inferenceConfig" in body
         assert body["inferenceConfig"]["maxTokens"] == bedrock_provider.max_tokens
-        assert body["inferenceConfig"]["temperature"] == bedrock_provider.temperature
+
+    def test_unset_sampling_params_are_omitted_from_inference_config(
+        self, bedrock_provider, sample_chat_context, sample_tools
+    ):
+        """Bedrock hosts models that reject temperature/topP outright.
+
+        Nothing in Omnimancer lets a user choose these values, so an
+        unconfigured provider must not send them and trigger a
+        ValidationException on models that do not accept them.
+        """
+        for body_str in (
+            bedrock_provider._prepare_bedrock_request("Hello", sample_chat_context),
+            bedrock_provider._prepare_bedrock_request_with_tools(
+                "Hello", sample_chat_context, sample_tools
+            ),
+        ):
+            inference = json.loads(body_str)["inferenceConfig"]
+            assert "temperature" not in inference
+            assert "topP" not in inference
+            assert inference["maxTokens"] == bedrock_provider.max_tokens
+
+    def test_configured_sampling_params_are_sent(
+        self, sample_chat_context, sample_tools
+    ):
+        """A user who configures temperature/top_p still gets them sent."""
+        provider = BedrockProvider(
+            api_key="key", model=_SONNET_MODEL, temperature=0.2, top_p=0.9
+        )
+
+        for body_str in (
+            provider._prepare_bedrock_request("Hello", sample_chat_context),
+            provider._prepare_bedrock_request_with_tools(
+                "Hello", sample_chat_context, sample_tools
+            ),
+        ):
+            inference = json.loads(body_str)["inferenceConfig"]
+            assert inference["temperature"] == 0.2
+            assert inference["topP"] == 0.9
+
+    @pytest.mark.asyncio
+    async def test_validate_model_access_omits_temperature(self, bedrock_provider):
+        """The model-access probe must not send temperature either."""
+        mock_response = MagicMock()
+        mock_response.status_code = 200
+
+        with patch("httpx.AsyncClient") as mock_client:
+            mock_post = AsyncMock(return_value=mock_response)
+            mock_client.return_value.__aenter__.return_value.post = mock_post
+
+            await bedrock_provider.validate_model_access()
+
+            sent = json.loads(mock_post.call_args.kwargs["content"])
+            assert "temperature" not in sent["inferenceConfig"]
 
     def test_prepare_request_standard_model_has_no_model_id_field(
         self, bedrock_provider, sample_chat_context