rasa-pro 3.12.22__py3-none-any.whl → 3.13.0__py3-none-any.whl

rasa/tracing/config.py

@@ -23,15 +23,13 @@ from rasa.core.actions.custom_action_executor import (
 from rasa.core.actions.grpc_custom_action_executor import GRPCCustomActionExecutor
 from rasa.core.agent import Agent
 from rasa.core.processor import MessageProcessor
-from rasa.core.tracker_store import TrackerStore
-from rasa.dialogue_understanding.commands import (
-    Command,
-    FreeFormAnswerCommand,
-)
+from rasa.core.tracker_stores.tracker_store import TrackerStore
+from rasa.dialogue_understanding.commands import Command, FreeFormAnswerCommand
 from rasa.dialogue_understanding.generator import (
     CompactLLMCommandGenerator,
     LLMCommandGenerator,
     MultiStepLLMCommandGenerator,
+    SearchReadyLLMCommandGenerator,
     SingleStepLLMCommandGenerator,
 )
 from rasa.dialogue_understanding.generator.flow_retrieval import FlowRetrieval
@@ -112,6 +110,7 @@ def configure_tracing(tracer_provider: Optional[TracerProvider]) -> None:
         grpc_custom_action_executor_class=GRPCCustomActionExecutor,
         single_step_llm_command_generator_class=SingleStepLLMCommandGenerator,
         compact_llm_command_generator_class=CompactLLMCommandGenerator,
+        search_ready_llm_command_generator_class=SearchReadyLLMCommandGenerator,
         multi_step_llm_command_generator_class=MultiStepLLMCommandGenerator,
         custom_action_executor_subclasses=custom_action_executor_subclasses,
         flow_retrieval_class=FlowRetrieval,

rasa/tracing/constants.py

@@ -44,6 +44,18 @@ COMPACT_LLM_COMMAND_GENERATOR_PROMPT_TOKEN_USAGE_METRIC_NAME = (
 COMPACT_LLM_COMMAND_GENERATOR_LLM_RESPONSE_DURATION_METRIC_NAME = (
     "compact_llm_command_generator_llm_response_duration"
 )
+SEARCH_READY_LLM_COMMAND_GENERATOR_CPU_USAGE_METRIC_NAME = (
+    "search_ready_llm_command_generator_cpu_usage"
+)
+SEARCH_READY_LLM_COMMAND_GENERATOR_MEMORY_USAGE_METRIC_NAME = (
+    "search_ready_llm_command_generator_memory_usage"
+)
+SEARCH_READY_LLM_COMMAND_GENERATOR_PROMPT_TOKEN_USAGE_METRIC_NAME = (
+    "search_ready_llm_command_generator_prompt_token_usage"
+)
+SEARCH_READY_LLM_COMMAND_GENERATOR_LLM_RESPONSE_DURATION_METRIC_NAME = (
+    "search_ready_llm_command_generator_llm_response_duration"
+)
 MULTI_STEP_LLM_COMMAND_GENERATOR_CPU_USAGE_METRIC_NAME = (
     "multi_step_llm_command_generator_cpu_usage"
 )
@@ -56,7 +68,13 @@ MULTI_STEP_LLM_COMMAND_GENERATOR_PROMPT_TOKEN_USAGE_METRIC_NAME = (
 MULTI_STEP_LLM_COMMAND_GENERATOR_LLM_RESPONSE_DURATION_METRIC_NAME = (
     "multi_step_llm_command_generator_llm_response_duration"
 )
-
+ENTERPRISE_SEARCH_POLICY_CPU_USAGE_METRIC_NAME = "enterprise_search_policy_cpu_usage"
+ENTERPRISE_SEARCH_POLICY_MEMORY_USAGE_METRIC_NAME = (
+    "enterprise_search_policy_memory_usage"
+)
+ENTERPRISE_SEARCH_POLICY_PROMPT_TOKEN_USAGE_METRIC_NAME = (
+    "enterprise_search_policy_prompt_token_usage"
+)
 ENTERPRISE_SEARCH_POLICY_LLM_RESPONSE_DURATION_METRIC_NAME = (
     "enterprise_search_policy_llm_response_duration"
 )

rasa/tracing/instrumentation/attribute_extractors.py

@@ -16,14 +16,20 @@ from rasa.core.channels import UserMessage
 from rasa.core.lock_store import LOCK_LIFETIME, LockStore
 from rasa.core.nlg.contextual_response_rephraser import ContextualResponseRephraser
 from rasa.core.processor import MessageProcessor
-from rasa.core.tracker_store import TrackerStore
+from rasa.core.tracker_stores.tracker_store import TrackerStore
 from rasa.dialogue_understanding.commands import Command
 from rasa.dialogue_understanding.generator import LLMBasedCommandGenerator
 from rasa.dialogue_understanding.generator.constants import FLOW_RETRIEVAL_KEY
 from rasa.dialogue_understanding.stack.dialogue_stack import DialogueStack
 from rasa.dialogue_understanding_test.du_test_result import (
+    KEY_COMMANDS_F1_MACRO,
+    KEY_COMMANDS_F1_MICRO,
+    KEY_COMMANDS_F1_WEIGHTED,
     KEY_TEST_CASES_ACCURACY,
     KEY_USER_UTTERANCES_ACCURACY,
+    OUTPUT_COMMANDS_F1_MACRO_INSTRUMENTATION_ATTR,
+    OUTPUT_COMMANDS_F1_MICRO_INSTRUMENTATION_ATTR,
+    OUTPUT_COMMANDS_F1_WEIGHTED_INSTRUMENTATION_ATTR,
     OUTPUT_COMPLETION_TOKEN_METRICS,
     OUTPUT_LATENCY_METRICS,
     OUTPUT_NAMES_OF_FAILED_TESTS,
@@ -58,7 +64,10 @@ from rasa.shared.core.trackers import DialogueStateTracker
 from rasa.shared.core.training_data.structures import StoryGraph
 from rasa.shared.importers.importer import TrainingDataImporter
 from rasa.shared.nlu.constants import INTENT_NAME_KEY, SET_SLOT_COMMAND
-from rasa.shared.utils.llm import combine_custom_and_default_config
+from rasa.shared.utils.llm import (
+    combine_custom_and_default_config,
+    resolve_model_client_config,
+)
 from rasa.tracing.constants import (
     PROMPT_TOKEN_LENGTH_ATTRIBUTE_NAME,
     REQUEST_BODY_SIZE_IN_BYTES_ATTRIBUTE_NAME,
@@ -329,9 +338,8 @@ def extract_llm_config(
     else:
         config = self.config
 
-    llm_property = combine_custom_and_default_config(
-        config.get(LLM_CONFIG_KEY), default_llm_config
-    )
+    llm_config = resolve_model_client_config(config.get(LLM_CONFIG_KEY))
+    llm_property = combine_custom_and_default_config(llm_config, default_llm_config)
 
     if isinstance(self, LLMBasedCommandGenerator):
         flow_retrieval_config = config.get(FLOW_RETRIEVAL_KEY, {}) or {}
@@ -340,8 +348,11 @@ def extract_llm_config(
             default_embeddings_config,
         )
     else:
+        embeddings_config = resolve_model_client_config(
+            config.get(EMBEDDINGS_CONFIG_KEY)
+        )
         embeddings_property = combine_custom_and_default_config(
-            config.get(EMBEDDINGS_CONFIG_KEY), default_embeddings_config
+            embeddings_config, default_embeddings_config
         )
 
     attributes = {
@@ -614,6 +625,15 @@ def extract_attrs_for_du_print_test_results(
         OUTPUT_NAMES_OF_FAILED_TESTS: json.dumps(
             test_suite_result.names_of_failed_tests
         ),
+        OUTPUT_COMMANDS_F1_MACRO_INSTRUMENTATION_ATTR: (
+            test_suite_result.f1_score[KEY_COMMANDS_F1_MACRO]
+        ),
+        OUTPUT_COMMANDS_F1_MICRO_INSTRUMENTATION_ATTR: (
+            test_suite_result.f1_score[KEY_COMMANDS_F1_MICRO]
+        ),
+        OUTPUT_COMMANDS_F1_WEIGHTED_INSTRUMENTATION_ATTR: (
+            test_suite_result.f1_score[KEY_COMMANDS_F1_WEIGHTED]
+        ),
     }
     if test_suite_result.command_metrics:
         for (
@@ -638,12 +658,27 @@ def extract_attrs_for_du_print_test_results(
             for key, value in test_suite_result.llm_config.items():
                 attributes_dict[f"llm_config_0_{key}"] = value
 
-    for key, value in test_suite_result.latency_metrics.items():
-        attributes_dict[f"{OUTPUT_LATENCY_METRICS}_{key}"] = value
-    for key, value in test_suite_result.prompt_token_metrics.items():
-        attributes_dict[f"{OUTPUT_PROMPT_TOKEN_METRICS}_{key}"] = value
-    for key, value in test_suite_result.completion_token_metrics.items():
-        attributes_dict[f"{OUTPUT_COMPLETION_TOKEN_METRICS}_{key}"] = value
+    for component_name, latency_metric in test_suite_result.latency_metrics.items():
+        for metric_name, metric_value in latency_metric.items():
+            attributes_dict[
+                f"{OUTPUT_LATENCY_METRICS}_{component_name}_{metric_name}"
+            ] = metric_value
+    for (
+        component_name,
+        prompt_token_metric,
+    ) in test_suite_result.prompt_token_metrics.items():
+        for metric_name, metric_value in prompt_token_metric.items():
+            attributes_dict[
+                f"{OUTPUT_PROMPT_TOKEN_METRICS}_{component_name}_{metric_name}"
+            ] = metric_value
+    for (
+        component_name,
+        completion_token_metric,
+    ) in test_suite_result.completion_token_metrics.items():
+        for metric_name, metric_value in completion_token_metric.items():
+            attributes_dict[
+                f"{OUTPUT_COMPLETION_TOKEN_METRICS}_{component_name}_{metric_name}"
+            ] = metric_value
 
     return attributes_dict
 
@@ -737,8 +772,8 @@ def extract_attrs_for_intentless_policy_generate_llm_answer(
     return extend_attributes_with_prompt_tokens_length(self, attributes, prompt)
 
 
-def extract_attrs_for_enterprise_search_generate_llm_answer(
-    self: "EnterpriseSearchPolicy", llm: "BaseLLM", prompt: str
+def extract_attrs_for_enterprise_search_invoke_llm(
+    self: "EnterpriseSearchPolicy", prompt: str
 ) -> Dict[str, Any]:
     from rasa.core.policies.enterprise_search_policy import (
         DEFAULT_EMBEDDINGS_CONFIG,
@@ -754,6 +789,14 @@ def extract_attrs_for_enterprise_search_generate_llm_answer(
     return extend_attributes_with_prompt_tokens_length(self, attributes, prompt)
 
 
+def extract_attrs_for_enterprise_search_parse_llm_relevancy_check_response(
+    self: "EnterpriseSearchPolicy", llm_answer: str
+) -> Dict[str, Any]:
+    return {
+        "llm_answer": llm_answer,
+    }
+
+
 def extract_current_context_attribute(stack: DialogueStack) -> Dict[str, Any]:
     """Utility function to extract the current context from the dialogue stack."""
     current_context = stack.current_context()
@@ -776,14 +819,23 @@ def compute_prompt_tokens_length(
         )
         return None
 
-    if model_name in ["gpt-3.5-turbo", "gpt-4"]:
+    if model_name == "gpt-4":
         logger.debug(
             f"Model {model_name} may update over time. "
             f"Returning num tokens assuming model '{model_name}-0613.'"
         )
         model_name = f"{model_name}-0613"
 
-    encoding = tiktoken.encoding_for_model(model_name)
+    try:
+        encoding = tiktoken.encoding_for_model(model_name)
+    except KeyError:
+        # Fallback for unknown models
+        logger.warning(
+            f"Unknown model name '{model_name}', "
+            f"using 'cl100k_base' encoding as fallback."
+        )
+        encoding = tiktoken.get_encoding("cl100k_base")
+
     return len(encoding.encode(prompt))
 
 

rasa/tracing/instrumentation/instrumentation.py

@@ -38,12 +38,13 @@ from rasa.core.nlg import NaturalLanguageGenerator
 from rasa.core.policies.flows.flow_step_result import FlowActionPrediction
 from rasa.core.policies.policy import Policy, PolicyPrediction
 from rasa.core.processor import MessageProcessor
-from rasa.core.tracker_store import TrackerStore
+from rasa.core.tracker_stores.tracker_store import TrackerStore
 from rasa.dialogue_understanding.commands import Command
 from rasa.dialogue_understanding.generator import (
     CompactLLMCommandGenerator,
     LLMCommandGenerator,
     MultiStepLLMCommandGenerator,
+    SearchReadyLLMCommandGenerator,
     SingleStepLLMCommandGenerator,
 )
 from rasa.dialogue_understanding.generator.flow_retrieval import FlowRetrieval
@@ -66,9 +67,11 @@ from rasa.tracing.instrumentation.intentless_policy_instrumentation import (
 from rasa.tracing.instrumentation.metrics import (
     record_callable_duration_metrics,
     record_compact_llm_command_generator_metrics,
+    record_enterprise_search_policy_metrics,
     record_llm_command_generator_metrics,
     record_multi_step_llm_command_generator_metrics,
     record_request_size_in_bytes,
+    record_search_ready_llm_command_generator_metrics,
     record_single_step_llm_command_generator_metrics,
 )
 from rasa.utils.endpoints import EndpointConfig, concat_url
@@ -291,6 +294,9 @@ SingleStepLLMCommandGeneratorType = TypeVar(
 CompactLLMCommandGeneratorType = TypeVar(
     "CompactLLMCommandGeneratorType", bound=CompactLLMCommandGenerator
 )
+SearchReadyLLMCommandGeneratorType = TypeVar(
+    "SearchReadyLLMCommandGeneratorType", bound=SearchReadyLLMCommandGenerator
+)
 MultiStepLLMCommandGeneratorType = TypeVar(
     "MultiStepLLMCommandGeneratorType", bound=MultiStepLLMCommandGenerator
 )
@@ -326,6 +332,9 @@ def instrument(
     compact_llm_command_generator_class: Optional[
         Type[CompactLLMCommandGeneratorType]
     ] = None,
+    search_ready_llm_command_generator_class: Optional[
+        Type[SearchReadyLLMCommandGeneratorType]
+    ] = None,
     multi_step_llm_command_generator_class: Optional[
         Type[MultiStepLLMCommandGeneratorType]
     ] = None,
@@ -379,6 +388,9 @@ def instrument(
     :param compact_llm_command_generator_class: The `CompactLLMCommandGenerator`
         to be instrumented. If `None` is given, no `CompactLLMCommandGenerator` will
         be instrumented.
+    :param search_ready_llm_command_generator_class: The`SearchReadyLLMCommandGenerator`
+        to be instrumented. If `None` is given, no `SearchReadyLLMCommandGenerator` will
+        be instrumented.
     :param multi_step_llm_command_generator_class: The `MultiStepLLMCommandGenerator`
         to be instrumented. If `None` is given, no `MultiStepLLMCommandGenerator` will
         be instrumented.
@@ -527,6 +539,37 @@ def instrument(
         )
         mark_class_as_instrumented(compact_llm_command_generator_class)
 
+    if (
+        search_ready_llm_command_generator_class is not None
+        and not class_is_instrumented(search_ready_llm_command_generator_class)
+    ):
+        _instrument_method(
+            tracer_provider.get_tracer(
+                search_ready_llm_command_generator_class.__module__
+            ),
+            search_ready_llm_command_generator_class,
+            "invoke_llm",
+            attribute_extractors.extract_attrs_for_llm_based_command_generator,
+            metrics_recorder=record_search_ready_llm_command_generator_metrics,
+        )
+        _instrument_method(
+            tracer_provider.get_tracer(
+                search_ready_llm_command_generator_class.__module__
+            ),
+            search_ready_llm_command_generator_class,
+            "_check_commands_against_startable_flows",
+            attribute_extractors.extract_attrs_for_check_commands_against_startable_flows,
+        )
+        _instrument_perform_health_check_method_for_component(
+            tracer_provider.get_tracer(
+                search_ready_llm_command_generator_class.__module__
+            ),
+            search_ready_llm_command_generator_class,
+            "perform_llm_health_check",
+            attribute_extractors.extract_attrs_for_performing_health_check,
+        )
+        mark_class_as_instrumented(search_ready_llm_command_generator_class)
+
     if multi_step_llm_command_generator_class is not None and not class_is_instrumented(
         multi_step_llm_command_generator_class
     ):
@@ -562,6 +605,7 @@ def instrument(
                 llm_command_generator_class,
                 single_step_llm_command_generator_class,
                 compact_llm_command_generator_class,
+                search_ready_llm_command_generator_class,
                 multi_step_llm_command_generator_class,
             )
         )
@@ -847,8 +891,15 @@ def _instrument_enterprise_search_policy(
     _instrument_method(
         tracer,
         policy_class,
-        "_generate_llm_answer",
-        attribute_extractors.extract_attrs_for_enterprise_search_generate_llm_answer,
+        "_invoke_llm",
+        attribute_extractors.extract_attrs_for_enterprise_search_invoke_llm,
+        metrics_recorder=record_enterprise_search_policy_metrics,
+    )
+    _instrument_method(
+        tracer,
+        policy_class,
+        "_parse_llm_relevancy_check_response",
+        attribute_extractors.extract_attrs_for_enterprise_search_parse_llm_relevancy_check_response,
     )
     _instrument_perform_health_check_method_for_component(
         tracer_provider.get_tracer(policy_class.__module__),

rasa/tracing/instrumentation/metrics.py

@@ -9,6 +9,7 @@ from rasa.dialogue_understanding.generator import (
     CompactLLMCommandGenerator,
     LLMCommandGenerator,
     MultiStepLLMCommandGenerator,
+    SearchReadyLLMCommandGenerator,
     SingleStepLLMCommandGenerator,
 )
 from rasa.tracing.constants import (
@@ -17,7 +18,10 @@ from rasa.tracing.constants import (
     COMPACT_LLM_COMMAND_GENERATOR_MEMORY_USAGE_METRIC_NAME,
     COMPACT_LLM_COMMAND_GENERATOR_PROMPT_TOKEN_USAGE_METRIC_NAME,
     CONTEXTUAL_RESPONSE_REPHRASER_LLM_RESPONSE_DURATION_METRIC_NAME,
+    ENTERPRISE_SEARCH_POLICY_CPU_USAGE_METRIC_NAME,
     ENTERPRISE_SEARCH_POLICY_LLM_RESPONSE_DURATION_METRIC_NAME,
+    ENTERPRISE_SEARCH_POLICY_MEMORY_USAGE_METRIC_NAME,
+    ENTERPRISE_SEARCH_POLICY_PROMPT_TOKEN_USAGE_METRIC_NAME,
     INTENTLESS_POLICY_LLM_RESPONSE_DURATION_METRIC_NAME,
     LLM_COMMAND_GENERATOR_CPU_USAGE_METRIC_NAME,
     LLM_COMMAND_GENERATOR_LLM_RESPONSE_DURATION_METRIC_NAME,
@@ -31,6 +35,10 @@ from rasa.tracing.constants import (
     RASA_CLIENT_REQUEST_BODY_SIZE_METRIC_NAME,
     RASA_CLIENT_REQUEST_DURATION_METRIC_NAME,
     REQUEST_BODY_SIZE_IN_BYTES_ATTRIBUTE_NAME,
+    SEARCH_READY_LLM_COMMAND_GENERATOR_CPU_USAGE_METRIC_NAME,
+    SEARCH_READY_LLM_COMMAND_GENERATOR_LLM_RESPONSE_DURATION_METRIC_NAME,
+    SEARCH_READY_LLM_COMMAND_GENERATOR_MEMORY_USAGE_METRIC_NAME,
+    SEARCH_READY_LLM_COMMAND_GENERATOR_PROMPT_TOKEN_USAGE_METRIC_NAME,
     SINGLE_STEP_LLM_COMMAND_GENERATOR_CPU_USAGE_METRIC_NAME,
     SINGLE_STEP_LLM_COMMAND_GENERATOR_LLM_RESPONSE_DURATION_METRIC_NAME,
     SINGLE_STEP_LLM_COMMAND_GENERATOR_MEMORY_USAGE_METRIC_NAME,
@@ -40,7 +48,7 @@ from rasa.tracing.metric_instrument_provider import MetricInstrumentProvider
 from rasa.utils.endpoints import EndpointConfig
 
 
-def record_llm_based_command_generator_cpu_usage(
+def record_llm_based_component_cpu_usage(
     metric_instrument_provider: MetricInstrumentProvider,
     metric_name: str,
 ) -> None:
@@ -60,8 +68,10 @@ def record_llm_based_command_generator_cpu_usage(
     cpu_usage = psutil.cpu_percent()
     metric_instrument.record(amount=cpu_usage)
 
+    return None
 
-def record_llm_based_command_generator_memory_usage(
+
+def record_llm_based_component_memory_usage(
     metric_instrument_provider: MetricInstrumentProvider,
     metric_name: str,
 ) -> None:
@@ -81,8 +91,10 @@ def record_llm_based_command_generator_memory_usage(
     memory_usage = psutil.virtual_memory().percent
     metric_instrument.record(amount=memory_usage)
 
+    return None
+
 
-def record_llm_based_command_generator_prompt_token(
+def record_llm_based_component_prompt_token(
     metric_instrument_provider: MetricInstrumentProvider,
     attributes: Dict[str, Any],
     metric_name: str,
@@ -116,6 +128,8 @@ def record_llm_based_command_generator_prompt_token(
         amount=prompt_tokens_len,
     )
 
+    return None
+
 
 def record_llm_command_generator_metrics(attributes: Dict[str, Any]) -> None:
     """
@@ -132,17 +146,18 @@ def record_llm_command_generator_metrics(attributes: Dict[str, Any]) -> None:
     if not instrument_provider.instruments:
         return None
 
-    record_llm_based_command_generator_cpu_usage(
+    record_llm_based_component_cpu_usage(
         instrument_provider, LLM_COMMAND_GENERATOR_CPU_USAGE_METRIC_NAME
     )
-    record_llm_based_command_generator_memory_usage(
+    record_llm_based_component_memory_usage(
         instrument_provider, LLM_COMMAND_GENERATOR_MEMORY_USAGE_METRIC_NAME
     )
-    record_llm_based_command_generator_prompt_token(
+    record_llm_based_component_prompt_token(
         instrument_provider,
         attributes,
         LLM_COMMAND_GENERATOR_PROMPT_TOKEN_USAGE_METRIC_NAME,
     )
+    return None
 
 
 def record_single_step_llm_command_generator_metrics(
@@ -162,17 +177,18 @@ def record_single_step_llm_command_generator_metrics(
     if not instrument_provider.instruments:
         return None
 
-    record_llm_based_command_generator_cpu_usage(
+    record_llm_based_component_cpu_usage(
         instrument_provider, SINGLE_STEP_LLM_COMMAND_GENERATOR_CPU_USAGE_METRIC_NAME
     )
-    record_llm_based_command_generator_memory_usage(
+    record_llm_based_component_memory_usage(
         instrument_provider, SINGLE_STEP_LLM_COMMAND_GENERATOR_MEMORY_USAGE_METRIC_NAME
     )
-    record_llm_based_command_generator_prompt_token(
+    record_llm_based_component_prompt_token(
         instrument_provider,
         attributes,
         SINGLE_STEP_LLM_COMMAND_GENERATOR_PROMPT_TOKEN_USAGE_METRIC_NAME,
     )
+    return None
 
 
 def record_compact_llm_command_generator_metrics(
@@ -192,17 +208,49 @@ def record_compact_llm_command_generator_metrics(
     if not instrument_provider.instruments:
         return None
 
-    record_llm_based_command_generator_cpu_usage(
+    record_llm_based_component_cpu_usage(
         instrument_provider, COMPACT_LLM_COMMAND_GENERATOR_CPU_USAGE_METRIC_NAME
     )
-    record_llm_based_command_generator_memory_usage(
+    record_llm_based_component_memory_usage(
         instrument_provider, COMPACT_LLM_COMMAND_GENERATOR_MEMORY_USAGE_METRIC_NAME
     )
-    record_llm_based_command_generator_prompt_token(
+    record_llm_based_component_prompt_token(
         instrument_provider,
         attributes,
         COMPACT_LLM_COMMAND_GENERATOR_PROMPT_TOKEN_USAGE_METRIC_NAME,
     )
+    return None
+
+
+def record_search_ready_llm_command_generator_metrics(
+    attributes: Dict[str, Any],
+) -> None:
+    """
+    Record measurements for SearchReadyLLMCommandGenerator specific metrics.
+
+    The recording is done by the opentelemetry.metrics.Histogram instruments.
+    These instruments are registered to the MetricInstrumentProvider internal singleton.
+
+    :param attributes: Extracted tracing attributes
+    :return: None
+    """
+    instrument_provider = MetricInstrumentProvider()
+
+    if not instrument_provider.instruments:
+        return None
+
+    record_llm_based_component_cpu_usage(
+        instrument_provider, SEARCH_READY_LLM_COMMAND_GENERATOR_CPU_USAGE_METRIC_NAME
+    )
+    record_llm_based_component_memory_usage(
+        instrument_provider, SEARCH_READY_LLM_COMMAND_GENERATOR_MEMORY_USAGE_METRIC_NAME
+    )
+    record_llm_based_component_prompt_token(
+        instrument_provider,
+        attributes,
+        SEARCH_READY_LLM_COMMAND_GENERATOR_PROMPT_TOKEN_USAGE_METRIC_NAME,
+    )
+    return None
 
 
 def record_multi_step_llm_command_generator_metrics(attributes: Dict[str, Any]) -> None:
@@ -220,17 +268,47 @@ def record_multi_step_llm_command_generator_metrics(attributes: Dict[str, Any])
     if not instrument_provider.instruments:
         return None
 
-    record_llm_based_command_generator_cpu_usage(
+    record_llm_based_component_cpu_usage(
         instrument_provider, MULTI_STEP_LLM_COMMAND_GENERATOR_CPU_USAGE_METRIC_NAME
     )
-    record_llm_based_command_generator_memory_usage(
+    record_llm_based_component_memory_usage(
         instrument_provider, MULTI_STEP_LLM_COMMAND_GENERATOR_MEMORY_USAGE_METRIC_NAME
     )
-    record_llm_based_command_generator_prompt_token(
+    record_llm_based_component_prompt_token(
         instrument_provider,
         attributes,
         MULTI_STEP_LLM_COMMAND_GENERATOR_PROMPT_TOKEN_USAGE_METRIC_NAME,
     )
+    return None
+
+
+def record_enterprise_search_policy_metrics(attributes: Dict[str, Any]) -> None:
+    """
+    Record measurements for EnterpriseSearchPolicy specific metrics.
+
+    The recording is done by the opentelemetry.metrics.Histogram instruments.
+    These instruments are registered to the MetricInstrumentProvider internal singleton.
+
+    :param attributes: Extracted tracing attributes
+    :return: None
+    """
+    instrument_provider = MetricInstrumentProvider()
+
+    if not instrument_provider.instruments:
+        return None
+
+    record_llm_based_component_cpu_usage(
+        instrument_provider, ENTERPRISE_SEARCH_POLICY_CPU_USAGE_METRIC_NAME
+    )
+    record_llm_based_component_memory_usage(
+        instrument_provider, ENTERPRISE_SEARCH_POLICY_MEMORY_USAGE_METRIC_NAME
+    )
+    record_llm_based_component_prompt_token(
+        instrument_provider,
+        attributes,
+        ENTERPRISE_SEARCH_POLICY_PROMPT_TOKEN_USAGE_METRIC_NAME,
+    )
+    return None
 
 
 def record_callable_duration_metrics(
@@ -276,6 +354,11 @@ def record_callable_duration_metrics(
             COMPACT_LLM_COMMAND_GENERATOR_LLM_RESPONSE_DURATION_METRIC_NAME
         )
 
+    if type(self) == SearchReadyLLMCommandGenerator:
+        metric_instrument = instrument_provider.get_instrument(
+            SEARCH_READY_LLM_COMMAND_GENERATOR_LLM_RESPONSE_DURATION_METRIC_NAME
+        )
+
     if type(self) == MultiStepLLMCommandGenerator:
         metric_instrument = instrument_provider.get_instrument(
             MULTI_STEP_LLM_COMMAND_GENERATOR_LLM_RESPONSE_DURATION_METRIC_NAME