aiqtoolkit 1.1.0a20250429__py3-none-any.whl → 1.1.0a20250502__py3-none-any.whl

aiq/data_models/api_server.py

@@ -32,14 +32,63 @@ from aiq.data_models.interactive import HumanPrompt
 from aiq.utils.type_converter import GlobalTypeConverter
 
 
+class ChatContentType(str, Enum):
+    """
+    ChatContentType is an Enum that represents the type of Chat content.
+    """
+    TEXT = "text"
+    IMAGE_URL = "image_url"
+    INPUT_AUDIO = "input_audio"
+
+
+class InputAudio(BaseModel):
+    data: str = "default"
+    format: str = "default"
+
+
+class AudioContent(BaseModel):
+    model_config = ConfigDict(extra="forbid")
+
+    type: typing.Literal[ChatContentType.INPUT_AUDIO] = ChatContentType.INPUT_AUDIO
+    input_audio: InputAudio = InputAudio()
+
+
+class ImageUrl(BaseModel):
+    url: HttpUrl = HttpUrl(url="http://default.com")
+
+
+class ImageContent(BaseModel):
+    model_config = ConfigDict(extra="forbid")
+
+    type: typing.Literal[ChatContentType.IMAGE_URL] = ChatContentType.IMAGE_URL
+    image_url: ImageUrl = ImageUrl()
+
+
+class TextContent(BaseModel):
+    model_config = ConfigDict(extra="forbid")
+
+    type: typing.Literal[ChatContentType.TEXT] = ChatContentType.TEXT
+    text: str = "default"
+
+
+class Security(BaseModel):
+    model_config = ConfigDict(extra="forbid")
+
+    api_key: str = "default"
+    token: str = "default"
+
+
+UserContent = typing.Annotated[TextContent | ImageContent | AudioContent, Discriminator("type")]
+
+
 class Message(BaseModel):
-    content: str
+    content: str | list[UserContent]
     role: str
 
 
 class AIQChatRequest(BaseModel):
     """
-    AIQChatRequest is a data model that represents a request to the AgentIQ chat API.
+    AIQChatRequest is a data model that represents a request to the AIQ Toolkit chat API.
     """
 
     # Allow extra fields in the model_config to support derived models
@@ -65,6 +114,20 @@ class AIQChatRequest(BaseModel):
                               max_tokens=max_tokens,
                               top_p=top_p)
 
+    @staticmethod
+    def from_content(content: list[UserContent],
+                     *,
+                     model: str | None = None,
+                     temperature: float | None = None,
+                     max_tokens: int | None = None,
+                     top_p: float | None = None) -> "AIQChatRequest":
+
+        return AIQChatRequest(messages=[Message(content=content, role="user")],
+                              model=model,
+                              temperature=temperature,
+                              max_tokens=max_tokens,
+                              top_p=top_p)
+
 
 class AIQChoiceMessage(BaseModel):
     content: str | None = None
@@ -88,8 +151,8 @@ class AIQUsage(BaseModel):
 
 class AIQResponseSerializable(abc.ABC):
     """
-    AIQChatResponseSerializable is an abstract class that defines the interface for serializing output for the AgentIQ
-    chat streaming API.
+    AIQChatResponseSerializable is an abstract class that defines the interface for serializing output for the AIQ
+    Toolkit chat streaming API.
     """
 
     @abstractmethod
@@ -111,7 +174,7 @@ class AIQResponseBaseModelIntermediate(BaseModel, AIQResponseSerializable):
 
 class AIQChatResponse(AIQResponseBaseModelOutput):
     """
-    AIQChatResponse is a data model that represents a response from the AgentIQ chat API.
+    AIQChatResponse is a data model that represents a response from the AIQ Toolkit chat API.
     """
 
     # Allow extra fields in the model_config to support derived models
@@ -152,7 +215,7 @@ class AIQChatResponse(AIQResponseBaseModelOutput):
 
 class AIQChatResponseChunk(AIQResponseBaseModelOutput):
     """
-    AIQChatResponseChunk is a data model that represents a response chunk from the AgentIQ chat streaming API.
+    AIQChatResponseChunk is a data model that represents a response chunk from the AIQ Toolkit chat streaming API.
     """
 
     # Allow extra fields in the model_config to support derived models
@@ -191,7 +254,7 @@ class AIQChatResponseChunk(AIQResponseBaseModelOutput):
 
 class AIQResponseIntermediateStep(AIQResponseBaseModelIntermediate):
     """
-    AIQResponseSerializedStep is a data model that represents a serialized step in the AgentIQ chat streaming API.
+    AIQResponseSerializedStep is a data model that represents a serialized step in the AIQ Toolkit chat streaming API.
     """
 
     # Allow extra fields in the model_config to support derived models
@@ -231,15 +294,6 @@ class UserMessageContentRoleType(str, Enum):
     ASSISTANT = "assistant"
 
 
-class ChatContentType(str, Enum):
-    """
-    ChatContentType is an Enum that represents the type of Chat content.
-    """
-    TEXT = "text"
-    IMAGE_URL = "image_url"
-    INPUT_AUDIO = "input_audio"
-
-
 class WebSocketMessageType(str, Enum):
     """
     WebSocketMessageType is an Enum that represents WebSocket Message types.
@@ -270,46 +324,6 @@ class WebSocketMessageStatus(str, Enum):
     COMPLETE = "complete"
 
 
-class InputAudio(BaseModel):
-    data: str = "default"
-    format: str = "default"
-
-
-class AudioContent(BaseModel):
-    model_config = ConfigDict(extra="forbid")
-
-    type: typing.Literal[ChatContentType.INPUT_AUDIO] = ChatContentType.INPUT_AUDIO
-    input_audio: InputAudio = InputAudio()
-
-
-class ImageUrl(BaseModel):
-    url: HttpUrl = HttpUrl(url="http://default.com")
-
-
-class ImageContent(BaseModel):
-    model_config = ConfigDict(extra="forbid")
-
-    type: typing.Literal[ChatContentType.IMAGE_URL] = ChatContentType.IMAGE_URL
-    image_url: ImageUrl = ImageUrl()
-
-
-class TextContent(BaseModel):
-    model_config = ConfigDict(extra="forbid")
-
-    type: typing.Literal[ChatContentType.TEXT] = ChatContentType.TEXT
-    text: str = "default"
-
-
-class Security(BaseModel):
-    model_config = ConfigDict(extra="forbid")
-
-    api_key: str = "default"
-    token: str = "default"
-
-
-UserContent = typing.Annotated[TextContent | ImageContent | AudioContent, Discriminator("type")]
-
-
 class UserMessages(BaseModel):
     model_config = ConfigDict(extra="forbid")
 
@@ -487,7 +501,9 @@ GlobalTypeConverter.register_converter(_generate_response_to_chat_response)
 
 # ======== AIQChatRequest Converters ========
 def _aiq_chat_request_to_string(data: AIQChatRequest) -> str:
-    return data.messages[-1].content
+    if isinstance(data.messages[-1].content, str):
+        return data.messages[-1].content
+    return str(data.messages[-1].content)
 
 
 GlobalTypeConverter.register_converter(_aiq_chat_request_to_string)

aiq/data_models/component_ref.py

@@ -43,7 +43,7 @@ class ComponentRefNode(HashableBaseModel):
 
     Args:
         ref_name (ComponentRef): The name of the component runtime instance.
-        component_group (ComponentGroup): The component group in an AgentIQ configuration object.
+        component_group (ComponentGroup): The component group in an AIQ Toolkit configuration object.
     """
 
     ref_name: "ComponentRef"
@@ -70,7 +70,7 @@ class ComponentRef(str, ABC):
         """Provides the component group this ComponentRef object represents.
 
         Returns:
-            ComponentGroup: A component group of the AgentIQ configuration object
+            ComponentGroup: A component group of the AIQ Toolkit configuration object
         """
 
         pass
@@ -82,7 +82,7 @@ class ComponentRef(str, ABC):
 
 class EmbedderRef(ComponentRef):
     """
-    A reference to an embedder in an AgentIQ configuration object.
+    A reference to an embedder in an AIQ Toolkit configuration object.
     """
 
     @property
@@ -93,7 +93,7 @@ class EmbedderRef(ComponentRef):
 
 class FunctionRef(ComponentRef):
     """
-    A reference to a function in an AgentIQ configuration object.
+    A reference to a function in an AIQ Toolkit configuration object.
     """
 
     @property
@@ -104,7 +104,7 @@ class FunctionRef(ComponentRef):
 
 class LLMRef(ComponentRef):
     """
-    A reference to an LLM in an AgentIQ configuration object.
+    A reference to an LLM in an AIQ Toolkit configuration object.
     """
 
     @property
@@ -115,7 +115,7 @@ class LLMRef(ComponentRef):
 
 class MemoryRef(ComponentRef):
     """
-    A reference to a memory in an AgentIQ configuration object.
+    A reference to a memory in an AIQ Toolkit configuration object.
     """
 
     @property
@@ -126,7 +126,7 @@ class MemoryRef(ComponentRef):
 
 class RetrieverRef(ComponentRef):
     """
-    A reference to a retriever in an AgentIQ configuration object.
+    A reference to a retriever in an AIQ Toolkit configuration object.
     """
 
     @property

aiq/data_models/discovery_metadata.py

@@ -55,11 +55,11 @@ class DiscoveryMetadata(BaseModel):
     """A data model representing metadata about each registered component to faciliate its discovery.
 
     Args:
-        package (str): The name of the package containing the AgentIQ component.
-        version (str): The version number of the package containing the AgentIQ component.
-        component_type (AIQComponentEnum): The type of AgentIQ component this metadata represents.
-        component_name (str): The registered name of the AgentIQ component.
-        description (str): Description of the AgentIQ component pulled from its config objects docstrings.
+        package (str): The name of the package containing the AIQ Toolkit component.
+        version (str): The version number of the package containing the AIQ Toolkit component.
+        component_type (AIQComponentEnum): The type of AIQ Toolkit component this metadata represents.
+        component_name (str): The registered name of the AIQ Toolkit component.
+        description (str): Description of the AIQ Toolkit component pulled from its config objects docstrings.
         developer_notes (str): Other notes to a developers to aid in the use of the component.
         status (DiscoveryStatusEnum): Provides the status of the metadata discovery process.
     """
@@ -129,7 +129,7 @@ class DiscoveryMetadata(BaseModel):
     @staticmethod
     def from_config_type(config_type: type["TypedBaseModelT"],
                          component_type: AIQComponentEnum = AIQComponentEnum.UNDEFINED) -> "DiscoveryMetadata":
-        """Generates discovery metadata from an AgentIQ config object.
+        """Generates discovery metadata from an AIQ Toolkit config object.
 
         Args:
             config_type (type[TypedBaseModelT]): A registered component's configuration object.
@@ -204,7 +204,7 @@ class DiscoveryMetadata(BaseModel):
         """Generates discovery metadata from an installed package name.
 
         Args:
-            package_name (str): The name of the AgentIQ plugin package containing registered components.
+            package_name (str): The name of the AIQ Toolkit plugin package containing registered components.
             package_version (str, optional): The version of the package, Defaults to None.
 
         Returns:

aiq/data_models/intermediate_step.py

@@ -98,7 +98,7 @@ class TraceMetadata(BaseModel):
 
 class IntermediateStepPayload(BaseModel):
     """
-    AIQIntermediateStep is a data model that represents an intermediate step in the AgentIQ. Intermediate steps are
+    AIQIntermediateStep is a data model that represents an intermediate step in the AIQ Toolkit. Intermediate steps are
     captured while a request is running and can be used to show progress or to evaluate the path a workflow took to get
     a response.
     """
@@ -203,7 +203,7 @@ class IntermediateStepPayload(BaseModel):
 
 class IntermediateStep(BaseModel):
     """
-    AIQIntermediateStep is a data model that represents an intermediate step in the AgentIQ. Intermediate steps are
+    AIQIntermediateStep is a data model that represents an intermediate step in the AIQ Toolkit. Intermediate steps are
     captured while a request is running and can be used to show progress or to evaluate the path a workflow took to get
     a response.
     """

aiq/eval/register.py

@@ -20,3 +20,4 @@
 from .rag_evaluator.register import register_ragas_evaluator
 from .swe_bench_evaluator.register import register_swe_bench_evaluator
 from .trajectory_evaluator.register import register_trajectory_evaluator
+from .tunable_rag_evaluator.register import register_tunable_rag_evaluator

aiq/eval/remote_workflow.py

@@ -52,7 +52,7 @@ class EvaluationRemoteWorkflowHandler:
 
         try:
             # Use the streaming endpoint
-            endpoint = f"{self.config.endpoint}/generate/stream/full"
+            endpoint = f"{self.config.endpoint}/generate/full"
             async with session.post(endpoint, json=payload) as response:
                 response.raise_for_status()  # Raise an exception for HTTP errors
 

aiq/eval/tunable_rag_evaluator/evaluate.py

@@ -0,0 +1,263 @@
+# SPDX-FileCopyrightText: Copyright (c) 2025, NVIDIA CORPORATION & AFFILIATES. All rights reserved.
+# SPDX-License-Identifier: Apache-2.0
+#
+# 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.
+
+import asyncio
+import logging
+
+from langchain.output_parsers import ResponseSchema
+from langchain.output_parsers import StructuredOutputParser
+from langchain.schema import HumanMessage
+from langchain.schema import SystemMessage
+from langchain_core.language_models import BaseChatModel
+from tqdm import tqdm
+
+from aiq.eval.evaluator.evaluator_model import EvalInput
+from aiq.eval.evaluator.evaluator_model import EvalInputItem
+from aiq.eval.evaluator.evaluator_model import EvalOutput
+from aiq.eval.evaluator.evaluator_model import EvalOutputItem
+from aiq.eval.utils.tqdm_position_registry import TqdmPositionRegistry
+
+logger = logging.getLogger(__name__)
+
+# pylint: disable=line-too-long
+# flake8: noqa: E501
+
+
+def evaluation_prompt(judge_llm_prompt: str,
+                      question: str,
+                      answer_description: str,
+                      generated_answer: str,
+                      format_instructions: str,
+                      default_scoring: bool):
+    """
+    This function generates a prompt for the judge LLM to evaluate the generated answer.
+    """
+
+    DEFAULT_SCORING_INSTRUCTIONS = """
+    The coverage score is a measure of how well the generated answer covers the critical aspects mentioned in the expected answer. A low coverage score indicates that the generated answer misses critical aspects of the expected answer. A middle coverage score indicates that the generated answer covers some of the must-haves of the expected answer but lacks other details. A high coverage score indicates that all of the expected aspects are present in the generated answer.
+    The correctness score is a measure of how well the generated answer matches the expected answer. A low correctness score indicates that the generated answer is incorrect or does not match the expected answer. A middle correctness score indicates that the generated answer is correct but lacks some details. A high correctness score indicates that the generated answer is exactly the same as the expected answer.
+    The relevance score is a measure of how well the generated answer is relevant to the question. A low relevance score indicates that the generated answer is not relevant to the question. A middle relevance score indicates that the generated answer is somewhat relevant to the question. A high relevance score indicates that the generated answer is exactly relevant to the question.
+    The reasoning is a 1-2 sentence explanation for the scoring.
+    """
+
+    DEFAULT_EVAL_PROMPT = (f"You are an intelligent assistant that responds strictly in JSON format."
+                           f"Judge based on the following scoring rubric: {DEFAULT_SCORING_INSTRUCTIONS}"
+                           f"{judge_llm_prompt}\n"
+                           f"{format_instructions}\n"
+                           f"Here is the user's query: {question}"
+                           f"Here is the description of the expected answer: {answer_description}"
+                           f"Here is the generated answer: {generated_answer}")
+
+    EVAL_PROMPT = (f"You are an intelligent assistant that responds strictly in JSON format. {judge_llm_prompt}\n"
+                   f"{format_instructions}\n"
+                   f"Here is the user's query: {question}"
+                   f"Here is the description of the expected answer: {answer_description}"
+                   f"Here is the generated answer: {generated_answer}")
+
+    return EVAL_PROMPT if not default_scoring else DEFAULT_EVAL_PROMPT
+
+
+class TunableRagEvaluator:
+    '''Tunable RAG evaluator class with customizable LLM prompt for scoring.'''
+
+    def __init__(self,
+                 llm: BaseChatModel,
+                 judge_llm_prompt: str,
+                 max_concurrency: int,
+                 default_scoring: bool,
+                 default_score_weights: dict):
+        self.llm = llm
+        self.max_concurrency = max_concurrency
+        self.judge_llm_prompt = judge_llm_prompt
+        self.semaphore = asyncio.Semaphore(self.max_concurrency)
+        self.default_scoring = default_scoring
+        # Use user-provided weights if available; otherwise, set equal weights for each score
+        self.default_score_weights = default_score_weights if default_score_weights else {
+            "coverage": 1 / 3, "correctness": 1 / 3, "relevance": 1 / 3
+        }
+
+    async def evaluate(self, eval_input: EvalInput) -> EvalOutput:
+        '''Evaluate function'''
+
+        async def process_item(item):
+            """Compute RAG evaluation for an individual item"""
+            question = item.input_obj
+            answer_description = item.expected_output_obj
+            generated_answer = item.output_obj
+
+            # Call judge LLM to generate score
+            score = 0.0
+
+            default_evaluation_schema = [
+                ResponseSchema(
+                    name="coverage_score",
+                    description=
+                    "Score for the coverage of all critical aspects mentioned in the expected answer. Ex. 0.5",
+                    type="float"),
+                ResponseSchema(
+                    name="correctness_score",
+                    description=
+                    "Score for the accuracy of the generated answer compared to the expected answer. Ex. 0.5",
+                    type="float"),
+                ResponseSchema(name="relevance_score",
+                               description="Score for the relevance of the generated answer to the question. Ex. 0.5",
+                               type="float"),
+                ResponseSchema(
+                    name="reasoning",
+                    description=
+                    "1-2 summarized sentences of reasoning for the scores. Ex. 'The generated answer covers all critical aspects mentioned in the expected answer, is correct, and is relevant to the question.'",
+                    type="string"),
+            ]
+
+            custom_evaluation_schema = [
+                ResponseSchema(name="score", description="Score for the generated answer. Ex. 0.5", type="float"),
+                ResponseSchema(
+                    name="reasoning",
+                    description=
+                    "1-2 sentence reasoning for the score. Ex. 'The generated answer is exactly the same as the description of the expected answer.'",
+                    type="string"),
+            ]
+
+            if self.default_scoring:
+                evaluation_schema = default_evaluation_schema
+            else:
+                evaluation_schema = custom_evaluation_schema
+
+            llm_input_response_parser = StructuredOutputParser.from_response_schemas(evaluation_schema)
+            format_instructions = llm_input_response_parser.get_format_instructions()
+
+            eval_prompt = evaluation_prompt(judge_llm_prompt=self.judge_llm_prompt,
+                                            question=question,
+                                            answer_description=answer_description,
+                                            generated_answer=generated_answer,
+                                            format_instructions=format_instructions,
+                                            default_scoring=self.default_scoring)
+
+            messages = [
+                SystemMessage(content="You must respond only in JSON format."), HumanMessage(content=eval_prompt)
+            ]
+
+            response = await self.llm.ainvoke(messages)
+
+            # Initialize default values to handle service errors
+            coverage_score = 0.0
+            correctness_score = 0.0
+            relevance_score = 0.0
+            reasoning = "Error in evaluator from parsing judge LLM response."
+
+            try:
+                parsed_response = llm_input_response_parser.parse(response.content)
+                if self.default_scoring:
+                    try:
+                        coverage_score = parsed_response["coverage_score"]
+                        correctness_score = parsed_response["correctness_score"]
+                        relevance_score = parsed_response["relevance_score"]
+                        reasoning = parsed_response["reasoning"]
+                    except KeyError as e:
+                        logger.error("Missing required keys in default scoring response: %s",
+                                     ", ".join(str(arg) for arg in e.args))
+                        reasoning = f"Error in evaluator from parsing judge LLM response. Missing required key(s): {', '.join(str(arg) for arg in e.args)}"
+
+                    coverage_weight = self.default_score_weights.get("coverage", 1 / 3)
+                    correctness_weight = self.default_score_weights.get("correctness", 1 / 3)
+                    relevance_weight = self.default_score_weights.get("relevance", 1 / 3)
+
+                    # Calculate score
+                    total_weight = coverage_weight + correctness_weight + relevance_weight
+                    coverage_weight = coverage_weight / total_weight
+                    correctness_weight = correctness_weight / total_weight
+                    relevance_weight = relevance_weight / total_weight
+
+                    if round(coverage_weight + correctness_weight + relevance_weight, 2) != 1:
+                        logger.warning("The sum of the default score weights is not 1. The weights will be normalized.")
+                        coverage_weight = coverage_weight / (coverage_weight + correctness_weight + relevance_weight)
+                        correctness_weight = correctness_weight / (coverage_weight + correctness_weight +
+                                                                   relevance_weight)
+                        relevance_weight = relevance_weight / (coverage_weight + correctness_weight + relevance_weight)
+
+                    score = (coverage_weight * coverage_score + correctness_weight * correctness_score +
+                             relevance_weight * relevance_score)
+
+                else:
+                    try:
+                        score = parsed_response["score"]
+                        reasoning = parsed_response["reasoning"]
+                    except KeyError as e:
+                        logger.error("Missing required keys in custom scoring response: %s",
+                                     ", ".join(str(arg) for arg in e.args))
+                        reasoning = f"Error in evaluator from parsing judge LLM response. Missing required key(s): {', '.join(str(arg) for arg in e.args)}"
+                        raise
+            except (KeyError, ValueError) as e:
+                logger.error("Error parsing judge LLM response: %s", e)
+                score = 0.0
+                reasoning = "Error in evaluator from parsing judge LLM response."
+
+            if self.default_scoring:
+                reasoning = {
+                    "question": question,
+                    "answer_description": answer_description,
+                    "generated_answer": generated_answer,
+                    "score_breakdown": {
+                        "coverage_score": coverage_score,
+                        "correctness_score": correctness_score,
+                        "relevance_score": relevance_score,
+                    },
+                    "reasoning": reasoning,
+                }
+            else:
+                reasoning = {
+                    "question": question,
+                    "answer_description": answer_description,
+                    "generated_answer": generated_answer,
+                    "reasoning": reasoning
+                }
+
+            return score, reasoning
+
+        async def wrapped_process(item: EvalInputItem) -> tuple[float, dict]:
+            """
+            Process an item asynchronously and update the progress bar.
+            Use the semaphore to limit the number of concurrent items.
+            """
+            async with self.semaphore:
+                result = await process_item(item)
+                # Update the progress bar
+                pbar.update(1)
+                return result
+
+        try:
+            # Claim a tqdm position to display the progress bar
+            tqdm_position = TqdmPositionRegistry.claim()
+            # Create a progress bar
+            pbar = tqdm(total=len(eval_input.eval_input_items), desc="Evaluating RAG", position=tqdm_position)
+            # Process items concurrently with a limit on concurrency
+            results = await asyncio.gather(*[wrapped_process(item) for item in eval_input.eval_input_items])
+        finally:
+            pbar.close()
+            TqdmPositionRegistry.release(tqdm_position)
+
+        # Extract scores and reasonings
+        sample_scores, sample_reasonings = zip(*results) if results else ([], [])
+
+        # Compute average score
+        avg_score = round(sum(sample_scores) / len(sample_scores), 2) if sample_scores else 0.0
+
+        # Construct EvalOutputItems
+        eval_output_items = [
+            EvalOutputItem(id=item.id, score=score, reasoning=reasoning)
+            for item, score, reasoning in zip(eval_input.eval_input_items, sample_scores, sample_reasonings)
+        ]
+
+        return EvalOutput(average_score=avg_score, eval_output_items=eval_output_items)

aiq/eval/tunable_rag_evaluator/register.py

@@ -0,0 +1,50 @@
+# SPDX-FileCopyrightText: Copyright (c) 2025, NVIDIA CORPORATION & AFFILIATES. All rights reserved.
+# SPDX-License-Identifier: Apache-2.0
+#
+# 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 pydantic import Field
+
+from aiq.builder.builder import EvalBuilder
+from aiq.builder.evaluator import EvaluatorInfo
+from aiq.builder.framework_enum import LLMFrameworkEnum
+from aiq.cli.register_workflow import register_evaluator
+from aiq.data_models.component_ref import LLMRef
+from aiq.data_models.evaluator import EvaluatorBaseConfig
+
+
+class TunableRagEvaluatorConfig(EvaluatorBaseConfig, name="tunable_rag_evaluator"):
+    '''Configuration for tunable RAG evaluator'''
+    llm_name: LLMRef = Field(description="Name of the judge LLM")
+    judge_llm_prompt: str = Field(description="LLM prompt for the judge LLM")
+    default_scoring: bool = Field(description="Whether to use default scoring", default=False)
+    default_score_weights: dict = Field(
+        default={
+            "coverage": 0.5, "correctness": 0.3, "relevance": 0.2
+        },
+        description="Weights for the different scoring components when using default scoring")
+
+
+@register_evaluator(config_type=TunableRagEvaluatorConfig)
+async def register_tunable_rag_evaluator(config: TunableRagEvaluatorConfig, builder: EvalBuilder):
+    '''Register tunable RAG evaluator'''
+    from .evaluate import TunableRagEvaluator
+
+    llm = await builder.get_llm(config.llm_name, wrapper_type=LLMFrameworkEnum.LANGCHAIN)
+    evaluator = TunableRagEvaluator(llm,
+                                    config.judge_llm_prompt,
+                                    builder.get_max_concurrency(),
+                                    config.default_scoring,
+                                    config.default_score_weights)
+
+    yield EvaluatorInfo(config=config, evaluate_fn=evaluator.evaluate, description="Tunable RAG Evaluator")

aiq/front_ends/console/console_front_end_config.py

@@ -22,7 +22,7 @@ from aiq.data_models.front_end import FrontEndBaseConfig
 
 class ConsoleFrontEndConfig(FrontEndBaseConfig, name="console"):
     """
-    A front end that allows an AgentIQ workflow to be run from the console.
+    A front end that allows an AIQ Toolkit workflow to be run from the console.
     """
 
     input_query: list[str] | None = Field(default=None,