Fix missing pipeline implementations. (#1514)

Author: manstis

ansible_ai_connect/ai/api/model_pipelines/README.md

@@ -0,0 +1,110 @@
+# Model Pipelines
+
+Ansible AI Connect is becoming feature rich.
+
+It supports API for the following features:
+- Code completions
+- Content match
+- Playbook Generation
+- Role Generation
+- Playbook Explanation
+- Chat Bot
+
+"Model Pipelines" provides a mechanism to support different _pipelines_ and configuration for each of these features for different providers. Different providers require different configuration information.
+
+## Pipelines
+
+A pipeline can exist for each feature for each type of provider.
+
+Types of provider are:
+- `grpc`
+- `http`
+- `dummy`
+- `wca`
+- `wca-onprem`
+- `wca-dummy`
+- `ollama`
+- `llamacpp`
+- `nop`
+
+### Implementing pipelines
+
+Implementations of a pipeline, for a particular provider, for a particular feature should extend the applicable base class; implementing the `invoke(..)` method accordingly:
+- `ModelPipelineCompletions`
+- `ModelPipelineContentMatch`
+- `ModelPipelinePlaybookGeneration`
+- `ModelPipelineRoleGeneration`
+- `ModelPipelinePlaybookExplanation`
+- `ModelPipelineChatBot`
+
+### Registering pipelines
+
+Implementations of pipelines, per provider, per feature are dynamically registered. To register a pipeline the implementing class should be decorated with `@Register(api_type="<type>")`.
+
+In addition to the supported features themselves implementations for the following must also be provided and registered:
+- `MetaData`
+
+  A class providing basic meta-data for all features for the applicable provider.
+
+  For example API Key, Model ID, Timeout etc.
+
+
+- `PipelineConfiguration`
+
+  A class representing the pipelines configuration parameters.
+
+
+- `Serializer`
+
+  A class that can deserialise configuration JSON/YAML into the target `PipelineConfiguration` class.
+
+### Default implementations
+
+A "No Operation" pipeline is registered by default for each provider and each feature where a concrete implementation is not explicitly available.
+
+### Lookup
+
+A registry is constructed at start-up, containing information of configured pipelines for all providers for all features.
+```
+REGISTRY = {
+    "http": {
+        MetaData: <Implementing class>,
+        ModelPipelineCompletions: <Implementing class>
+        ModelPipelineContentMatch: <Implementing class>
+        ModelPipelinePlaybookGeneration: <Implementing class>
+        ModelPipelineRoleGeneration: <Implementing class>
+        ModelPipelinePlaybookExplanation: <Implementing class>
+        ModelPipelineChatBot: <Implementing class>
+        PipelineConfiguration: <Implementing class>
+        Serializer: <Implementing class>
+    }
+    ...
+}
+```
+
+To invoke a pipeline for a particular feature the instance for the configured provider can be retrieved from the `ai` Django application:
+```
+pipeline: ModelPipelinePlaybookGeneration =
+    apps
+    .get_app_config("ai")
+    .get_model_pipeline(ModelPipelinePlaybookGeneration)
+```
+The pipeline can then be invoked:
+```
+playbook, outline, warnings = pipeline.invoke(
+    PlaybookGenerationParameters.init(
+        request=request,
+        text=self.validated_data["text"],
+        custom_prompt=self.validated_data["customPrompt"],
+        create_outline=self.validated_data["createOutline"],
+        outline=self.validated_data["outline"],
+        generation_id=self.validated_data["generationId"],
+        model_id=self.req_model_id,
+    )
+)
+```
+The code is identical irrespective of which provider is configured.
+
+### Configuration
+
+Refer to the [examples](../../../../docs/config).

ansible_ai_connect/ai/api/model_pipelines/__init__.py

@@ -17,3 +17,6 @@
 import ansible_ai_connect.ai.api.model_pipelines.wca.pipelines_dummy  # noqa
 import ansible_ai_connect.ai.api.model_pipelines.wca.pipelines_onprem  # noqa
 import ansible_ai_connect.ai.api.model_pipelines.wca.pipelines_saas  # noqa
+from ansible_ai_connect.ai.api.model_pipelines.registry import set_defaults
+
+set_defaults()

ansible_ai_connect/ai/api/model_pipelines/factory.py

@@ -21,6 +21,7 @@
     PipelineConfiguration,
 )
 from ansible_ai_connect.ai.api.model_pipelines.config_providers import Configuration
+from ansible_ai_connect.ai.api.model_pipelines.nop.pipelines import NopMetaData
 from ansible_ai_connect.ai.api.model_pipelines.pipelines import PIPELINE_TYPE
 from ansible_ai_connect.ai.api.model_pipelines.registry import REGISTRY, REGISTRY_ENTRY
 
@@ -46,16 +47,19 @@ def get_pipeline(self, pipeline_type: Type[PIPELINE_TYPE]) -> PIPELINE_TYPE:
         try:
             # Get the configuration for the requested pipeline
             pipeline_config: PipelineConfiguration = self.pipelines_config[pipeline_type.__name__]
+
             # Get the pipeline class for the configured provider
             pipelines = REGISTRY[pipeline_config.provider]
             pipeline = pipelines[pipeline_type]
-            config = pipeline_config.config
-            # No explicit implementation defined; fallback to NOP
-            if pipeline_config.provider == "nop":
+
+            # Ensure NOP instances are created with NOP configuration
+            if issubclass(pipeline, NopMetaData):
                 logger.info(f"Using NOP implementation for '{pipeline_type.__name__}'.")
+                pipelines = REGISTRY["nop"]
+                pipeline_config = pipelines[PipelineConfiguration]()
 
             # Construct an instance of the pipeline class with the applicable configuration
-            self.cache[pipeline_type] = pipeline(config)
+            self.cache[pipeline_type] = pipeline(pipeline_config.config)
 
         except KeyError:
             pass

ansible_ai_connect/ai/api/model_pipelines/nop/configuration.py

@@ -40,5 +40,5 @@ def __init__(self, **kwargs):
 
 
 @Register(api_type="nop")
-class LlamaCppConfigurationSerializer(serializers.Serializer):
+class NopConfigurationSerializer(serializers.Serializer):
     pass

ansible_ai_connect/ai/api/model_pipelines/registry.py

@@ -70,3 +70,23 @@ def __call__(self, cls):
         elif issubclass(cls, Serializer):
             REGISTRY[self.api_type][Serializer] = cls
         return cls
+
+
+def set_defaults():
+
+    def set_defaults_for_api_type(pipeline_provider):
+
+        def v_or_default(k, v):
+            defaults = REGISTRY["nop"]
+            if v is None:
+                logger.warning(
+                    f"'{k.alias()}' is not available for provider '{pipeline_provider}',"
+                    " failing back to 'nop'"
+                )
+                return defaults[k]
+            return v
+
+        return {k: v_or_default(k, v) for k, v in REGISTRY[pipeline_provider].items()}
+
+    for model_mesh_api_type in get_args(t_model_mesh_api_type):
+        REGISTRY[model_mesh_api_type] = set_defaults_for_api_type(model_mesh_api_type)

ansible_ai_connect/ai/api/model_pipelines/tests/test_default_pipelines.py

@@ -0,0 +1,57 @@
+#  Copyright Red Hat
+#
+#  Licensed under the Apache License, Version 2.0 (the "License");
+#  you may not use this file except in compliance with the License.
+#  You may obtain a copy of the License at
+#
+#      http://www.apache.org/licenses/LICENSE-2.0
+#
+#  Unless required by applicable law or agreed to in writing, software
+#  distributed under the License is distributed on an "AS IS" BASIS,
+#  WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+#  See the License for the specific language governing permissions and
+#  limitations under the License.
+from django.test import override_settings
+
+from ansible_ai_connect.ai.api.model_pipelines.factory import ModelPipelineFactory
+from ansible_ai_connect.ai.api.model_pipelines.nop.pipelines import NopChatBotPipeline
+from ansible_ai_connect.ai.api.model_pipelines.pipelines import (
+    MetaData,
+    ModelPipelineChatBot,
+)
+from ansible_ai_connect.ai.api.model_pipelines.registry import REGISTRY, REGISTRY_ENTRY
+from ansible_ai_connect.test_utils import WisdomServiceAPITestCaseBaseOIDC
+
+
+class TestDefaultModelPipelines(WisdomServiceAPITestCaseBaseOIDC):
+
+    @override_settings(ANSIBLE_AI_MODEL_MESH_CONFIG="{}")
+    def test_default_pipeline_when_not_defined(self):
+        factory = ModelPipelineFactory()
+
+        # The configuration is empty. All pipelines should fall back to the NOP variety
+        pipelines = list(filter(lambda p: issubclass(p, MetaData), REGISTRY_ENTRY.keys()))
+        for pipeline in pipelines:
+            nop = REGISTRY["nop"][pipeline]
+            with self.assertLogs(logger="root", level="INFO") as log:
+                implementation = factory.get_pipeline(pipeline)
+                self.assertIsNotNone(implementation)
+                self.assertIsInstance(implementation, nop)
+                self.assertInLog(
+                    f"Using NOP implementation for '{pipeline.__name__}'.",
+                    log,
+                )
+
+    @override_settings(ANSIBLE_AI_MODEL_MESH_CONFIG="ModelPipelineChatBot:\n  provider: dummy")
+    def test_default_pipeline_when_not_implemented(self):
+        factory = ModelPipelineFactory()
+
+        # ChatBot is configured to use "dummy" however there is no "dummy" ChatBot implementation
+        with self.assertLogs(logger="root", level="INFO") as log:
+            pipeline = factory.get_pipeline(ModelPipelineChatBot)
+            self.assertIsNotNone(pipeline)
+            self.assertIsInstance(pipeline, NopChatBotPipeline)
+            self.assertInLog(
+                "Using NOP implementation for 'ModelPipelineChatBot'.",
+                log,
+            )