rasa-pro 3.11.13__py3-none-any.whl → 3.11.14__py3-none-any.whl

rasa/api.py

@@ -81,6 +81,7 @@ def train(
     remote_storage: Optional[StorageType] = None,
     file_importer: Optional["TrainingDataImporter"] = None,
     keep_local_model_copy: bool = False,
+    remote_root_only: bool = False,
 ) -> "TrainingResult":
     """Runs Rasa Core and NLU training in `async` loop.
 
@@ -108,6 +109,8 @@ def train(
             If it is not provided, a new instance will be created.
         keep_local_model_copy: If `True` the model will be stored locally even if
             remote storage is configured.
+        remote_root_only: If `True`, the model will be stored in the root of the
+            remote model storage.
 
     Returns:
         An instance of `TrainingResult`.
@@ -131,6 +134,7 @@ def train(
             remote_storage=remote_storage,
             file_importer=file_importer,
             keep_local_model_copy=keep_local_model_copy,
+            remote_root_only=remote_root_only,
         )
     )
 

rasa/cli/arguments/default_arguments.py

@@ -172,7 +172,7 @@ def add_remote_storage_param(
 ) -> None:
     parser.add_argument(
         "--remote-storage",
-        help="Remote storage which should be used to store/load the model."
+        help="Remote storage which should be used to store/load the model. "
         f"Supported storages are: {RemoteStorageType.list()}. "
         "You can also provide your own implementation of the `Persistor` interface.",
         required=required,
@@ -180,6 +180,18 @@ def add_remote_storage_param(
     )
 
 
+def add_remote_root_only_param(
+    parser: argparse.ArgumentParser, required: bool = False
+) -> None:
+    parser.add_argument(
+        "--remote-root-only",
+        action="store_true",
+        help="If set, models will be stored only at the root directory "
+        "of the remote storage.",
+        required=required,
+    )
+
+
 def parse_remote_storage_arg(value: str) -> StorageType:
     try:
         return parse_remote_storage(value)

rasa/cli/arguments/train.py

@@ -8,6 +8,7 @@ from rasa.cli.arguments.default_arguments import (
     add_out_param,
     add_domain_param,
     add_endpoint_param,
+    add_remote_root_only_param,
     add_remote_storage_param,
 )
 from rasa.graph_components.providers.training_tracker_provider import (
@@ -41,6 +42,7 @@ def set_train_arguments(parser: argparse.ArgumentParser) -> None:
         parser, help_text="Configuration file for the connectors as a yml file."
     )
     add_remote_storage_param(parser)
+    add_remote_root_only_param(parser)
 
 
 def set_train_core_arguments(parser: argparse.ArgumentParser) -> None:

rasa/cli/train.py

@@ -153,6 +153,7 @@ def run_training(args: argparse.Namespace, can_exit: bool = False) -> Optional[T
         remote_storage=args.remote_storage,
         file_importer=training_data_importer,
         keep_local_model_copy=args.keep_local_model_copy,
+        remote_root_only=args.remote_root_only,
     )
     if training_result.code != 0 and can_exit:
         sys.exit(training_result.code)

rasa/constants.py

@@ -43,3 +43,5 @@ DEFAULT_BUCKET_NAME = "rasa-models"
 
 HTTP_STATUS_FORBIDDEN = 403
 HTTP_STATUS_NOT_FOUND = 404
+
+RASA_REMOTE_STORAGE_ENV_VAR_NAME = "RASA_REMOTE_STORAGE"

rasa/core/nlg/contextual_response_rephraser.py

@@ -4,16 +4,21 @@ import structlog
 from jinja2 import Template
 from rasa import telemetry
 from rasa.core.nlg.response import TemplatedNaturalLanguageGenerator
-from rasa.core.nlg.summarize import summarize_conversation
+from rasa.core.nlg.summarize import (
+    _count_multiple_utterances_as_single_turn,
+    summarize_conversation,
+)
 from rasa.shared.constants import (
     LLM_CONFIG_KEY,
+    MAX_COMPLETION_TOKENS_CONFIG_KEY,
     MODEL_CONFIG_KEY,
+    MODEL_GROUP_ID_CONFIG_KEY,
     MODEL_NAME_CONFIG_KEY,
+    OPENAI_PROVIDER,
     PROMPT_CONFIG_KEY,
     PROVIDER_CONFIG_KEY,
-    OPENAI_PROVIDER,
+    TEMPERATURE_CONFIG_KEY,
     TIMEOUT_CONFIG_KEY,
-    MODEL_GROUP_ID_CONFIG_KEY,
 )
 from rasa.shared.core.domain import KEY_RESPONSES_TEXT, Domain
 from rasa.shared.core.events import BotUttered, UserUttered
@@ -45,12 +50,13 @@ RESPONSE_SUMMARISE_CONVERSATION_KEY = "summarize_conversation"
 DEFAULT_REPHRASE_ALL = False
 DEFAULT_SUMMARIZE_HISTORY = True
 DEFAULT_MAX_HISTORICAL_TURNS = 5
+DEFAULT_COUNT_MULTIPLE_UTTERANCES_AS_SINGLE_TURN = True
 
 DEFAULT_LLM_CONFIG = {
     PROVIDER_CONFIG_KEY: OPENAI_PROVIDER,
     MODEL_CONFIG_KEY: DEFAULT_OPENAI_GENERATE_MODEL_NAME,
-    "temperature": 0.3,
-    "max_tokens": DEFAULT_OPENAI_MAX_GENERATED_TOKENS,
+    TEMPERATURE_CONFIG_KEY: 0.3,
+    MAX_COMPLETION_TOKENS_CONFIG_KEY: DEFAULT_OPENAI_MAX_GENERATED_TOKENS,
     TIMEOUT_CONFIG_KEY: 5,
 }
 
@@ -62,6 +68,7 @@ its meaning. Use simple english.
 Context / previous conversation with the user:
 {{history}}
 
+Last user message:
 {{current_input}}
 
 Suggested AI Response: {{suggested_response}}
@@ -112,6 +119,11 @@ class ContextualResponseRephraser(
             "max_historical_turns", DEFAULT_MAX_HISTORICAL_TURNS
         )
 
+        self.count_multiple_utterances_as_single_turn = self.nlg_endpoint.kwargs.get(
+            "count_multiple_utterances_as_single_turn",
+            DEFAULT_COUNT_MULTIPLE_UTTERANCES_AS_SINGLE_TURN,
+        )
+
         self.llm_config = resolve_model_client_config(
             self.nlg_endpoint.kwargs.get(LLM_CONFIG_KEY),
             ContextualResponseRephraser.__name__,
@@ -198,8 +210,16 @@ class ContextualResponseRephraser(
         Returns:
         The history for the prompt.
         """
+        # Count multiple utterances by bot/user as single turn in conversation history
+        turns_wrapper = (
+            _count_multiple_utterances_as_single_turn
+            if self.count_multiple_utterances_as_single_turn
+            else None
+        )
         llm = llm_factory(self.llm_config, DEFAULT_LLM_CONFIG)
-        return await summarize_conversation(tracker, llm, max_turns=5)
+        return await summarize_conversation(
+            tracker, llm, max_turns=5, turns_wrapper=turns_wrapper
+        )
 
     async def rephrase(
         self,
@@ -211,7 +231,6 @@ class ContextualResponseRephraser(
         Args:
             response: The response to rephrase.
             tracker: The tracker to use for the prediction.
-            model_name: The name of the model to use for the prediction.
 
         Returns:
             The response with the rephrased text.
@@ -221,19 +240,26 @@ class ContextualResponseRephraser(
 
         prompt_template_text = self._template_for_response_rephrasing(response)
 
-        # Retrieve inputs for the dynamic prompt
-        latest_message = self._last_message_if_human(tracker)
-        current_input = f"{USER}: {latest_message}" if latest_message else ""
+        # Last user message (=current input) should always be in prompt if available
+        last_message_by_user = getattr(tracker.latest_message, "text", "")
+        current_input = (
+            f"{USER}: {last_message_by_user}" if last_message_by_user else ""
+        )
 
         # Only summarise conversation history if flagged
         if self.summarize_history:
             history = await self._create_history(tracker)
         else:
-            # make sure the transcript/history contains the last user utterance
+            # Count multiple utterances by bot/user as single turn
+            turns_wrapper = (
+                _count_multiple_utterances_as_single_turn
+                if self.count_multiple_utterances_as_single_turn
+                else None
+            )
             max_turns = max(self.max_historical_turns, 1)
-            history = tracker_as_readable_transcript(tracker, max_turns=max_turns)
-            # the history already contains the current input
-            current_input = ""
+            history = tracker_as_readable_transcript(
+                tracker, max_turns=max_turns, turns_wrapper=turns_wrapper
+            )
 
         prompt = Template(prompt_template_text).render(
             history=history,

rasa/core/nlg/summarize.py

@@ -1,4 +1,5 @@
-from typing import Optional
+from itertools import groupby
+from typing import Callable, List, Optional
 
 import structlog
 from jinja2 import Template
@@ -22,20 +23,47 @@ SUMMARY_PROMPT_TEMPLATE = Template(_DEFAULT_SUMMARIZER_TEMPLATE)
 MAX_TURNS_DEFAULT = 20
 
 
+def _count_multiple_utterances_as_single_turn(transcript: List[str]) -> List[str]:
+    """Counts multiple utterances as a single turn.
+    Args:
+        transcript: the lines of the transcript
+    Returns:
+        transcript: with multiple utterances counted as a single turn
+    """
+    if not transcript:
+        return []
+
+    def get_speaker_label(line: str) -> str:
+        return line.partition(": ")[0] if ": " in line else ""
+
+    modified_transcript = [
+        f"{speaker}: {' '.join(line.partition(': ')[2] for line in group)}"
+        for speaker, group in groupby(transcript, key=get_speaker_label)
+        if speaker
+    ]
+
+    return modified_transcript
+
+
 def _create_summarization_prompt(
-    tracker: DialogueStateTracker, max_turns: Optional[int]
+    tracker: DialogueStateTracker,
+    max_turns: Optional[int],
+    turns_wrapper: Optional[Callable[[List[str]], List[str]]],
 ) -> str:
     """Creates an LLM prompt to summarize the conversation in the tracker.
 
     Args:
         tracker: tracker of the conversation to be summarized
         max_turns: maximum number of turns to summarize
+        turns_wrapper: optional function to wrap the turns
 
 
     Returns:
         The prompt to summarize the conversation.
     """
-    transcript = tracker_as_readable_transcript(tracker, max_turns=max_turns)
+    transcript = tracker_as_readable_transcript(
+        tracker, max_turns=max_turns, turns_wrapper=turns_wrapper
+    )
     return SUMMARY_PROMPT_TEMPLATE.render(
         conversation=transcript,
     )
@@ -45,6 +73,7 @@ async def summarize_conversation(
     tracker: DialogueStateTracker,
     llm: LLMClient,
     max_turns: Optional[int] = MAX_TURNS_DEFAULT,
+    turns_wrapper: Optional[Callable[[List[str]], List[str]]] = None,
 ) -> str:
     """Summarizes the dialogue using the LLM.
 
@@ -52,11 +81,12 @@ async def summarize_conversation(
         tracker: the tracker to summarize
         llm: the LLM to use for summarization
         max_turns: maximum number of turns to summarize
+        turns_wrapper: optional function to wrap the turns
 
     Returns:
         The summary of the dialogue.
     """
-    prompt = _create_summarization_prompt(tracker, max_turns)
+    prompt = _create_summarization_prompt(tracker, max_turns, turns_wrapper)
     try:
         llm_response = await llm.acompletion(prompt)
         summarization = llm_response.choices[0].strip()
@@ -65,6 +95,8 @@ async def summarize_conversation(
         )
         return summarization
     except Exception as e:
-        transcript = tracker_as_readable_transcript(tracker, max_turns=max_turns)
+        transcript = tracker_as_readable_transcript(
+            tracker, max_turns=max_turns, turns_wrapper=turns_wrapper
+        )
         structlogger.error("summarization.error", error=e)
         return transcript

rasa/core/persistor.py

@@ -121,10 +121,12 @@ def get_persistor(storage: StorageType) -> Optional[Persistor]:
 class Persistor(abc.ABC):
     """Store models in cloud and fetch them when needed."""
 
-    def persist(self, trained_model: str) -> None:
+    def persist(self, trained_model: str, remote_root_only: bool = False) -> None:
         """Uploads a trained model persisted in the `target_dir` to cloud storage."""
         absolute_file_key = self._create_file_key(trained_model)
-        file_key = Path(absolute_file_key).name
+        file_key = (
+            Path(absolute_file_key).name if remote_root_only else absolute_file_key
+        )
         self._persist_tar(file_key, trained_model)
 
     def retrieve(self, model_name: Text, target_path: Text) -> Text:
@@ -143,30 +145,32 @@ class Persistor(abc.ABC):
             # ensure backward compatibility
             tar_name = self._tar_name(model_name)
         tar_name = self._create_file_key(tar_name)
-        target_filename = os.path.basename(tar_name)
-        self._retrieve_tar(target_filename)
-        self._copy(os.path.basename(tar_name), target_path)
+        self._retrieve_tar(tar_name, target_path)
 
         if os.path.isdir(target_path):
             return os.path.join(target_path, model_name)
 
         return target_path
 
-    def size_of_persisted_model(self, model_name: Text) -> int:
+    def size_of_persisted_model(
+        self, model_name: Text, target_path: Optional[str] = None
+    ) -> int:
         """Returns the size of the model that has been persisted to cloud storage.
 
         Args:
             model_name: The name of the model to retrieve.
+            target_path: The path to which the model should be saved.
         """
         tar_name = model_name
         if not model_name.endswith(MODEL_ARCHIVE_EXTENSION):
             # ensure backward compatibility
             tar_name = self._tar_name(model_name)
         tar_name = self._create_file_key(tar_name)
-        target_filename = os.path.basename(tar_name)
-        return self._retrieve_tar_size(target_filename)
+        return self._retrieve_tar_size(tar_name, target_path)
 
-    def _retrieve_tar_size(self, filename: Text) -> int:
+    def _retrieve_tar_size(
+        self, filename: Text, target_path: Optional[str] = None
+    ) -> int:
         """Returns the size of the model that has been persisted to cloud storage."""
         structlogger.warning(
             "persistor.retrieve_tar_size.not_implemented",
@@ -179,11 +183,11 @@ class Persistor(abc.ABC):
                 "size directly from the cloud storage."
             ),
         )
-        self._retrieve_tar(filename)
+        self._retrieve_tar(filename, target_path)
         return os.path.getsize(os.path.basename(filename))
 
     @abc.abstractmethod
-    def _retrieve_tar(self, filename: Text) -> None:
+    def _retrieve_tar(self, filename: str, target_path: Optional[str] = None) -> None:
         """Downloads a model previously persisted to cloud storage."""
         raise NotImplementedError
 
@@ -302,7 +306,9 @@ class AWSPersistor(Persistor):
         with open(tar_path, "rb") as f:
             self.s3.Object(self.bucket_name, file_key).put(Body=f)
 
-    def _retrieve_tar_size(self, model_path: Text) -> int:
+    def _retrieve_tar_size(
+        self, model_path: Text, target_path: Optional[str] = None
+    ) -> int:
         """Returns the size of the model that has been persisted to s3."""
         try:
             obj = self.s3.Object(self.bucket_name, model_path)
@@ -310,7 +316,9 @@ class AWSPersistor(Persistor):
         except Exception:
             raise ModelNotFound()
 
-    def _retrieve_tar(self, target_filename: str) -> None:
+    def _retrieve_tar(
+        self, target_filename: str, target_path: Optional[str] = None
+    ) -> None:
         """Downloads a model that has previously been persisted to s3."""
         from botocore import exceptions
 
@@ -320,8 +328,14 @@ class AWSPersistor(Persistor):
             f"in the bucket."
         )
 
+        tar_name = (
+            os.path.join(target_path, os.path.basename(target_filename))
+            if target_path
+            else os.path.basename(target_filename)
+        )
+
         try:
-            with open(target_filename, "wb") as f:
+            with open(tar_name, "wb") as f:
                 self.bucket.download_fileobj(target_filename, f)
 
             structlogger.debug(
@@ -425,7 +439,9 @@ class GCSPersistor(Persistor):
         blob = self.bucket.blob(file_key)
         blob.upload_from_filename(tar_path)
 
-    def _retrieve_tar_size(self, target_filename: Text) -> int:
+    def _retrieve_tar_size(
+        self, target_filename: Text, target_path: Optional[str] = None
+    ) -> int:
         """Returns the size of the model that has been persisted to GCS."""
         try:
             blob = self.bucket.blob(target_filename)
@@ -433,13 +449,22 @@ class GCSPersistor(Persistor):
         except Exception:
             raise ModelNotFound()
 
-    def _retrieve_tar(self, target_filename: Text) -> None:
+    def _retrieve_tar(
+        self, target_filename: str, target_path: Optional[str] = None
+    ) -> None:
         """Downloads a model that has previously been persisted to GCS."""
         from google.api_core import exceptions
 
         blob = self.bucket.blob(target_filename)
+
+        destination = (
+            os.path.join(target_path, os.path.basename(target_filename))
+            if target_path
+            else target_filename
+        )
+
         try:
-            blob.download_to_filename(target_filename)
+            blob.download_to_filename(destination)
 
             structlogger.debug(
                 "gcs_persistor.retrieve_tar.object_found", object_key=target_filename
@@ -500,7 +525,9 @@ class AzurePersistor(Persistor):
         with open(tar_path, "rb") as data:
             self._container_client().upload_blob(name=file_key, data=data)
 
-    def _retrieve_tar_size(self, target_filename: Text) -> int:
+    def _retrieve_tar_size(
+        self, target_filename: Text, target_path: Optional[str] = None
+    ) -> int:
         """Returns the size of the model that has been persisted to Azure."""
         try:
             blob_client = self._container_client().get_blob_client(target_filename)
@@ -509,12 +536,20 @@ class AzurePersistor(Persistor):
         except Exception:
             raise ModelNotFound()
 
-    def _retrieve_tar(self, target_filename: Text) -> None:
+    def _retrieve_tar(
+        self, target_filename: Text, target_path: Optional[str] = None
+    ) -> None:
         """Downloads a model that has previously been persisted to Azure."""
         from azure.core.exceptions import AzureError
 
+        destination = (
+            os.path.join(target_path, os.path.basename(target_filename))
+            if target_path
+            else target_filename
+        )
+
         try:
-            with open(target_filename, "wb") as model_file:
+            with open(destination, "wb") as model_file:
                 blob_client = self._container_client().get_blob_client(target_filename)
                 download_stream = blob_client.download_blob()
                 model_file.write(download_stream.readall())

rasa/core/policies/enterprise_search_policy.py

@@ -45,13 +45,16 @@ from rasa.graph_components.providers.forms_provider import Forms
 from rasa.graph_components.providers.responses_provider import Responses
 from rasa.shared.constants import (
     EMBEDDINGS_CONFIG_KEY,
+    MAX_COMPLETION_TOKENS_CONFIG_KEY,
+    MAX_RETRIES_CONFIG_KEY,
     MODEL_CONFIG_KEY,
+    MODEL_GROUP_ID_CONFIG_KEY,
+    MODEL_NAME_CONFIG_KEY,
+    OPENAI_PROVIDER,
     PROMPT_CONFIG_KEY,
     PROVIDER_CONFIG_KEY,
-    OPENAI_PROVIDER,
+    TEMPERATURE_CONFIG_KEY,
     TIMEOUT_CONFIG_KEY,
-    MODEL_NAME_CONFIG_KEY,
-    MODEL_GROUP_ID_CONFIG_KEY,
 )
 from rasa.shared.core.constants import (
     ACTION_CANCEL_FLOW,
@@ -121,14 +124,14 @@ DEFAULT_LLM_CONFIG = {
     PROVIDER_CONFIG_KEY: OPENAI_PROVIDER,
     MODEL_CONFIG_KEY: DEFAULT_OPENAI_CHAT_MODEL_NAME,
     TIMEOUT_CONFIG_KEY: 10,
-    "temperature": 0.0,
-    "max_tokens": 256,
-    "max_retries": 1,
+    TEMPERATURE_CONFIG_KEY: 0.0,
+    MAX_COMPLETION_TOKENS_CONFIG_KEY: 256,
+    MAX_RETRIES_CONFIG_KEY: 1,
 }
 
 DEFAULT_EMBEDDINGS_CONFIG = {
     PROVIDER_CONFIG_KEY: OPENAI_PROVIDER,
-    "model": DEFAULT_OPENAI_EMBEDDING_MODEL_NAME,
+    MODEL_CONFIG_KEY: DEFAULT_OPENAI_EMBEDDING_MODEL_NAME,
 }
 
 ENTERPRISE_SEARCH_PROMPT_FILE_NAME = "enterprise_search_policy_prompt.jinja2"

rasa/core/policies/intentless_policy.py

@@ -31,13 +31,15 @@ from rasa.graph_components.providers.responses_provider import Responses
 from rasa.shared.constants import (
     EMBEDDINGS_CONFIG_KEY,
     LLM_CONFIG_KEY,
+    MAX_COMPLETION_TOKENS_CONFIG_KEY,
     MODEL_CONFIG_KEY,
+    MODEL_GROUP_ID_CONFIG_KEY,
     MODEL_NAME_CONFIG_KEY,
+    OPENAI_PROVIDER,
     PROMPT_CONFIG_KEY,
     PROVIDER_CONFIG_KEY,
-    OPENAI_PROVIDER,
+    TEMPERATURE_CONFIG_KEY,
     TIMEOUT_CONFIG_KEY,
-    MODEL_GROUP_ID_CONFIG_KEY,
 )
 from rasa.shared.core.constants import ACTION_LISTEN_NAME
 from rasa.shared.core.domain import KEY_RESPONSES_TEXT, Domain
@@ -110,14 +112,14 @@ NLU_ABSTENTION_THRESHOLD = "nlu_abstention_threshold"
 DEFAULT_LLM_CONFIG = {
     PROVIDER_CONFIG_KEY: OPENAI_PROVIDER,
     MODEL_CONFIG_KEY: DEFAULT_OPENAI_CHAT_MODEL_NAME,
-    "temperature": 0.0,
-    "max_tokens": DEFAULT_OPENAI_MAX_GENERATED_TOKENS,
+    TEMPERATURE_CONFIG_KEY: 0.0,
+    MAX_COMPLETION_TOKENS_CONFIG_KEY: DEFAULT_OPENAI_MAX_GENERATED_TOKENS,
     TIMEOUT_CONFIG_KEY: 5,
 }
 
 DEFAULT_EMBEDDINGS_CONFIG = {
     PROVIDER_CONFIG_KEY: OPENAI_PROVIDER,
-    "model": DEFAULT_OPENAI_EMBEDDING_MODEL_NAME,
+    MODEL_CONFIG_KEY: DEFAULT_OPENAI_EMBEDDING_MODEL_NAME,
 }
 
 DEFAULT_INTENTLESS_PROMPT_TEMPLATE = importlib.resources.open_text(
@@ -343,8 +345,6 @@ class IntentlessPolicy(LLMHealthCheckMixin, EmbeddingsHealthCheckMixin, Policy):
             # ensures that the policy will not override a deterministic policy
             # which utilizes the nlu predictions confidence (e.g. Memoization).
             NLU_ABSTENTION_THRESHOLD: 0.9,
-            LLM_CONFIG_KEY: DEFAULT_LLM_CONFIG,
-            EMBEDDINGS_CONFIG_KEY: DEFAULT_EMBEDDINGS_CONFIG,
             PROMPT_CONFIG_KEY: DEFAULT_INTENTLESS_PROMPT_TEMPLATE,
         }
 
@@ -380,13 +380,19 @@ class IntentlessPolicy(LLMHealthCheckMixin, EmbeddingsHealthCheckMixin, Policy):
         super().__init__(config, model_storage, resource, execution_context, featurizer)
 
         # Resolve LLM config
-        self.config[LLM_CONFIG_KEY] = resolve_model_client_config(
-            self.config.get(LLM_CONFIG_KEY), IntentlessPolicy.__name__
+        self.config[LLM_CONFIG_KEY] = combine_custom_and_default_config(
+            resolve_model_client_config(
+                self.config.get(LLM_CONFIG_KEY), IntentlessPolicy.__name__
+            ),
+            DEFAULT_LLM_CONFIG,
         )
 
         # Resolve embeddings config
-        self.config[EMBEDDINGS_CONFIG_KEY] = resolve_model_client_config(
-            self.config.get(EMBEDDINGS_CONFIG_KEY), IntentlessPolicy.__name__
+        self.config[EMBEDDINGS_CONFIG_KEY] = combine_custom_and_default_config(
+            resolve_model_client_config(
+                self.config.get(EMBEDDINGS_CONFIG_KEY), IntentlessPolicy.__name__
+            ),
+            DEFAULT_EMBEDDINGS_CONFIG,
         )
 
         self.nlu_abstention_threshold: float = self.config[NLU_ABSTENTION_THRESHOLD]

rasa/core/run.py

@@ -86,13 +86,15 @@ def _create_single_channel(channel: Text, credentials: Dict[Text, Any]) -> Any:
             )
 
 
-def _create_app_without_api(cors: Optional[Union[Text, List[Text]]] = None) -> Sanic:
+def _create_app_without_api(
+    cors: Optional[Union[Text, List[Text]]] = None, is_inspector_enabled: bool = False
+) -> Sanic:
     app = Sanic("rasa_core_no_api", configure_logging=False)
 
     # Reset Sanic warnings filter that allows the triggering of Sanic warnings
     warnings.filterwarnings("ignore", category=DeprecationWarning, module=r"sanic.*")
 
-    server.add_root_route(app)
+    server.add_root_route(app, is_inspector_enabled)
     server.configure_cors(app, cors)
     return app
 
@@ -127,6 +129,7 @@ def configure_app(
     server_listeners: Optional[List[Tuple[Callable, Text]]] = None,
     use_uvloop: Optional[bool] = True,
     keep_alive_timeout: int = constants.DEFAULT_KEEP_ALIVE_TIMEOUT,
+    is_inspector_enabled: bool = False,
 ) -> Sanic:
     """Run the agent."""
     rasa.core.utils.configure_file_logging(
@@ -144,6 +147,7 @@ def configure_app(
                 jwt_private_key=jwt_private_key,
                 jwt_method=jwt_method,
                 endpoints=endpoints,
+                is_inspector_enabled=is_inspector_enabled,
             )
         )
     else:
@@ -259,6 +263,7 @@ def serve_application(
         syslog_protocol=syslog_protocol,
         request_timeout=request_timeout,
         server_listeners=server_listeners,
+        is_inspector_enabled=inspect,
     )
 
     ssl_context = server.create_ssl_context(

rasa/dialogue_understanding/coexistence/llm_based_router.py

@@ -23,11 +23,14 @@ from rasa.engine.recipes.default_recipe import DefaultV1Recipe
 from rasa.engine.storage.resource import Resource
 from rasa.engine.storage.storage import ModelStorage
 from rasa.shared.constants import (
-    ROUTE_TO_CALM_SLOT,
-    PROMPT_CONFIG_KEY,
-    PROVIDER_CONFIG_KEY,
+    LOGIT_BIAS_CONFIG_KEY,
+    MAX_COMPLETION_TOKENS_CONFIG_KEY,
     MODEL_CONFIG_KEY,
     OPENAI_PROVIDER,
+    PROMPT_CONFIG_KEY,
+    PROVIDER_CONFIG_KEY,
+    ROUTE_TO_CALM_SLOT,
+    TEMPERATURE_CONFIG_KEY,
     TIMEOUT_CONFIG_KEY,
 )
 from rasa.shared.core.trackers import DialogueStateTracker
@@ -62,9 +65,11 @@ DEFAULT_LLM_CONFIG = {
     PROVIDER_CONFIG_KEY: OPENAI_PROVIDER,
     MODEL_CONFIG_KEY: DEFAULT_OPENAI_CHAT_MODEL_NAME,
     TIMEOUT_CONFIG_KEY: 7,
-    "temperature": 0.0,
-    "max_tokens": 1,
-    "logit_bias": {str(token_id): 100 for token_id in A_TO_C_TOKEN_IDS_CHATGPT},
+    TEMPERATURE_CONFIG_KEY: 0.0,
+    MAX_COMPLETION_TOKENS_CONFIG_KEY: 1,
+    LOGIT_BIAS_CONFIG_KEY: {
+        str(token_id): 100 for token_id in A_TO_C_TOKEN_IDS_CHATGPT
+    },
 }
 
 structlogger = structlog.get_logger()

rasa/dialogue_understanding/generator/constants.py

@@ -1,8 +1,10 @@
 from rasa.shared.constants import (
-    PROVIDER_CONFIG_KEY,
-    OPENAI_PROVIDER,
+    MAX_COMPLETION_TOKENS_CONFIG_KEY,
     MODEL_CONFIG_KEY,
+    OPENAI_PROVIDER,
+    PROVIDER_CONFIG_KEY,
     TIMEOUT_CONFIG_KEY,
+    TEMPERATURE_CONFIG_KEY,
 )
 from rasa.shared.utils.llm import (
     DEFAULT_OPENAI_CHAT_MODEL_NAME_ADVANCED,
@@ -12,8 +14,8 @@ from rasa.shared.utils.llm import (
 DEFAULT_LLM_CONFIG = {
     PROVIDER_CONFIG_KEY: OPENAI_PROVIDER,
     MODEL_CONFIG_KEY: DEFAULT_OPENAI_CHAT_MODEL_NAME_ADVANCED,
-    "temperature": 0.0,
-    "max_tokens": DEFAULT_OPENAI_MAX_GENERATED_TOKENS,
+    TEMPERATURE_CONFIG_KEY: 0.0,
+    MAX_COMPLETION_TOKENS_CONFIG_KEY: DEFAULT_OPENAI_MAX_GENERATED_TOKENS,
     TIMEOUT_CONFIG_KEY: 7,
 }