rasa-pro 3.13.0.dev1__py3-none-any.whl → 3.13.0.dev3__py3-none-any.whl

rasa/core/information_retrieval/faiss.py

@@ -31,12 +31,10 @@ class FAISS_Store(InformationRetrieval):
         index_path: str,
         docs_folder: Optional[str],
         create_index: Optional[bool] = False,
-        use_llm: bool = False,
     ):
         """Initializes the FAISS Store."""
         self.chunk_size = 1000
         self.chunk_overlap = 20
-        self.use_llm = use_llm
 
         path = Path(index_path) / "documents_faiss"
         if create_index:
@@ -73,57 +71,6 @@ class FAISS_Store(InformationRetrieval):
 
         return loader.load()
 
-    def _format_faqs(self, docs: List["Document"]) -> List["Document"]:
-        """Splits each loaded file into individual FAQs.
-
-        Args:
-            docs: Documents representing whole files containing FAQs.
-
-        Returns:
-            List of Document objects, each containing a separate FAQ.
-
-        Examples:
-            An example of a file containing FAQs:
-
-            Q: Who is Finley?
-            A: Finley is your smart assistant for the FinX App. You can add him to your
-               favorite messenger and tell him what you need help with.
-
-            Q: How does Finley work?
-            A: Finley is powered by the latest chatbot technology leveraging a unique
-               interplay of large language models and secure logic.
-
-        More details in documentation: https://rasa.com/docs/reference/config/policies/extractive-search/
-        """
-        structured_faqs = []
-        from langchain.schema import Document
-
-        for doc in docs:
-            faq_chunks = doc.page_content.strip().split("\n\n")
-
-            for chunk in faq_chunks:
-                lines = chunk.strip().split("\n")
-                if len(lines) < 2:
-                    continue  # Skip if something unexpected
-
-                question_line = lines[0].strip()
-                answer_line = lines[1].strip()
-
-                question = question_line.replace("Q: ", "").strip()
-                answer = answer_line.replace("A: ", "").strip()
-
-                doc_obj = Document(
-                    page_content=question,
-                    metadata={
-                        "title": question.lower().replace(" ", "_")[:-1],
-                        "type": "faq",
-                        "answer": answer,
-                    },
-                )
-
-                structured_faqs.append(doc_obj)
-        return structured_faqs
-
     def _create_document_index(
         self, docs_folder: Optional[str], embedding: "Embeddings"
     ) -> FAISS:
@@ -140,15 +87,12 @@ class FAISS_Store(InformationRetrieval):
             raise ValueError("parameter `docs_folder` needs to be specified")
 
         docs = self.load_documents(docs_folder)
-        if self.use_llm:
-            splitter = RecursiveCharacterTextSplitter(
-                chunk_size=self.chunk_size,
-                chunk_overlap=self.chunk_overlap,
-                length_function=len,
-            )
-            doc_chunks = splitter.split_documents(docs)
-        else:
-            doc_chunks = self._format_faqs(docs)
+        splitter = RecursiveCharacterTextSplitter(
+            chunk_size=self.chunk_size,
+            chunk_overlap=self.chunk_overlap,
+            length_function=len,
+        )
+        doc_chunks = splitter.split_documents(docs)
 
         logger.info(
             "information_retrieval.faiss_store._create_document_index",

rasa/core/nlg/contextual_response_rephraser.py

@@ -27,6 +27,7 @@ from rasa.shared.nlu.constants import (
     PROMPTS,
 )
 from rasa.shared.providers.llm.llm_response import LLMResponse, measure_llm_latency
+from rasa.shared.utils.constants import LOG_COMPONENT_SOURCE_METHOD_INIT
 from rasa.shared.utils.health_check.llm_health_check_mixin import LLMHealthCheckMixin
 from rasa.shared.utils.llm import (
     DEFAULT_OPENAI_GENERATE_MODEL_NAME,
@@ -105,6 +106,8 @@ class ContextualResponseRephraser(
         self.prompt_template = get_prompt_template(
             self.nlg_endpoint.kwargs.get(PROMPT_CONFIG_KEY),
             DEFAULT_RESPONSE_VARIATION_PROMPT_TEMPLATE,
+            log_source_component=ContextualResponseRephraser.__name__,
+            log_source_method=LOG_COMPONENT_SOURCE_METHOD_INIT,
         )
         self.rephrase_all = self.nlg_endpoint.kwargs.get(
             "rephrase_all", DEFAULT_REPHRASE_ALL

rasa/core/policies/enterprise_search_policy.py

@@ -78,6 +78,10 @@ from rasa.shared.providers.embedding._langchain_embedding_client_adapter import
 from rasa.shared.providers.llm.llm_client import LLMClient
 from rasa.shared.providers.llm.llm_response import LLMResponse, measure_llm_latency
 from rasa.shared.utils.cli import print_error_and_exit
+from rasa.shared.utils.constants import (
+    LOG_COMPONENT_SOURCE_METHOD_FINGERPRINT_ADDON,
+    LOG_COMPONENT_SOURCE_METHOD_INIT,
+)
 from rasa.shared.utils.health_check.embeddings_health_check_mixin import (
     EmbeddingsHealthCheckMixin,
 )
@@ -254,10 +258,14 @@ class EnterpriseSearchPolicy(LLMHealthCheckMixin, EmbeddingsHealthCheckMixin, Po
         self.prompt_template = prompt_template or get_prompt_template(
             self.config.get(PROMPT_CONFIG_KEY),
             DEFAULT_ENTERPRISE_SEARCH_PROMPT_TEMPLATE,
+            log_source_component=EnterpriseSearchPolicy.__name__,
+            log_source_method=LOG_COMPONENT_SOURCE_METHOD_INIT,
         )
         self.citation_prompt_template = get_prompt_template(
             self.config.get(PROMPT_CONFIG_KEY),
             DEFAULT_ENTERPRISE_SEARCH_PROMPT_WITH_CITATION_TEMPLATE,
+            log_source_component=EnterpriseSearchPolicy.__name__,
+            log_source_method=LOG_COMPONENT_SOURCE_METHOD_INIT,
         )
         # If citation is enabled, use the citation prompt template
         if self.citation_enabled:
@@ -373,7 +381,6 @@ class EnterpriseSearchPolicy(LLMHealthCheckMixin, EmbeddingsHealthCheckMixin, Po
                     embeddings=embeddings,
                     index_path=path,
                     create_index=True,
-                    use_llm=self.use_llm,
                 )
         else:
             logger.info("enterprise_search_policy.train.custom", store_type=store_type)
@@ -844,6 +851,8 @@ class EnterpriseSearchPolicy(LLMHealthCheckMixin, EmbeddingsHealthCheckMixin, Po
         prompt_template = get_prompt_template(
             config.get(PROMPT_CONFIG_KEY),
             DEFAULT_ENTERPRISE_SEARCH_PROMPT_TEMPLATE,
+            log_source_component=EnterpriseSearchPolicy.__name__,
+            log_source_method=LOG_COMPONENT_SOURCE_METHOD_FINGERPRINT_ADDON,
         )
 
         llm_config = resolve_model_client_config(

rasa/core/policies/flows/flow_executor.py

@@ -23,7 +23,6 @@ from rasa.core.policies.flows.flow_step_result import (
 )
 from rasa.dialogue_understanding.commands import CancelFlowCommand
 from rasa.dialogue_understanding.patterns.cancel import CancelPatternFlowStackFrame
-from rasa.dialogue_understanding.patterns.clarify import ClarifyPatternFlowStackFrame
 from rasa.dialogue_understanding.patterns.collect_information import (
     CollectInformationPatternFlowStackFrame,
 )
@@ -51,7 +50,6 @@ from rasa.dialogue_understanding.stack.frames.flow_stack_frame import (
 )
 from rasa.dialogue_understanding.stack.utils import (
     top_user_flow_frame,
-    user_flows_on_the_stack,
 )
 from rasa.shared.constants import RASA_PATTERN_HUMAN_HANDOFF
 from rasa.shared.core.constants import (
@@ -280,33 +278,6 @@ def trigger_pattern_continue_interrupted(
     return events
 
 
-def trigger_pattern_clarification(
-    current_frame: DialogueStackFrame, stack: DialogueStack, flows: FlowsList
-) -> None:
-    """Trigger the pattern to clarify which topic to continue if needed."""
-    if not isinstance(current_frame, UserFlowStackFrame):
-        return None
-
-    if current_frame.frame_type in [
-        FlowStackFrameType.CALL,
-        FlowStackFrameType.INTERRUPT,
-    ]:
-        # we want to return to the flow that called
-        # the current flow or the flow that was interrupted
-        # by the current flow
-        return None
-
-    pending_flows = [
-        flows.flow_by_id(frame.flow_id)
-        for frame in stack.frames
-        if isinstance(frame, UserFlowStackFrame)
-        and frame.flow_id != current_frame.flow_id
-    ]
-
-    flow_names = [flow.readable_name() for flow in pending_flows if flow is not None]
-    stack.push(ClarifyPatternFlowStackFrame(names=flow_names))
-
-
 def trigger_pattern_completed(
     current_frame: DialogueStackFrame, stack: DialogueStack, flows: FlowsList
 ) -> None:
@@ -675,15 +646,9 @@ def _run_end_step(
     structlogger.debug("flow.step.run.flow_end")
     current_frame = stack.pop()
     trigger_pattern_completed(current_frame, stack, flows)
-    resumed_events = []
-    if len(user_flows_on_the_stack(stack)) > 1:
-        # if there are more user flows on the stack,
-        # we need to trigger the pattern clarify
-        trigger_pattern_clarification(current_frame, stack, flows)
-    else:
-        resumed_events = trigger_pattern_continue_interrupted(
-            current_frame, stack, flows, tracker
-        )
+    resumed_events = trigger_pattern_continue_interrupted(
+        current_frame, stack, flows, tracker
+    )
     reset_events: List[Event] = reset_scoped_slots(current_frame, flow, tracker)
     return ContinueFlowWithNextStep(
         events=initial_events + reset_events + resumed_events, has_flow_ended=True

rasa/core/policies/intentless_policy.py

@@ -58,6 +58,7 @@ from rasa.shared.providers.embedding._langchain_embedding_client_adapter import
     _LangchainEmbeddingClientAdapter,
 )
 from rasa.shared.providers.llm.llm_client import LLMClient
+from rasa.shared.utils.constants import LOG_COMPONENT_SOURCE_METHOD_FINGERPRINT_ADDON
 from rasa.shared.utils.health_check.embeddings_health_check_mixin import (
     EmbeddingsHealthCheckMixin,
 )
@@ -939,6 +940,8 @@ class IntentlessPolicy(LLMHealthCheckMixin, EmbeddingsHealthCheckMixin, Policy):
         prompt_template = get_prompt_template(
             config.get(PROMPT_CONFIG_KEY),
             DEFAULT_INTENTLESS_PROMPT_TEMPLATE,
+            log_source_component=IntentlessPolicy.__name__,
+            log_source_method=LOG_COMPONENT_SOURCE_METHOD_FINGERPRINT_ADDON,
         )
 
         llm_config = resolve_model_client_config(

rasa/core/processor.py

@@ -76,6 +76,7 @@ from rasa.shared.core.constants import (
     SLOT_SILENCE_TIMEOUT,
     USER_INTENT_RESTART,
     USER_INTENT_SILENCE_TIMEOUT,
+    SetSlotExtractor,
 )
 from rasa.shared.core.events import (
     ActionExecuted,
@@ -766,13 +767,26 @@ class MessageProcessor:
         if self.http_interpreter:
             parse_data = await self.http_interpreter.parse(message)
         else:
-            regex_reader = create_regex_pattern_reader(message, self.domain)
-
             processed_message = Message({TEXT: message.text})
-            if regex_reader:
-                processed_message = regex_reader.unpack_regex_message(
-                    message=processed_message, domain=self.domain
+
+            all_flows = await self.get_flows()
+            should_force_slot_command, slot_name = (
+                rasa.core.utils.should_force_slot_filling(tracker, all_flows)
+            )
+
+            if should_force_slot_command:
+                command = SetSlotCommand(
+                    name=slot_name,
+                    value=message.text,
+                    extractor=SetSlotExtractor.COMMAND_PAYLOAD_READER.value,
                 )
+                processed_message.set(COMMANDS, [command.as_dict()], add_to_output=True)
+            else:
+                regex_reader = create_regex_pattern_reader(message, self.domain)
+                if regex_reader:
+                    processed_message = regex_reader.unpack_regex_message(
+                        message=processed_message, domain=self.domain
+                    )
 
             # Invalid use of slash syntax, sanitize the message before passing
             # it to the graph
@@ -1009,7 +1023,14 @@ class MessageProcessor:
 
     @staticmethod
     def _should_handle_message(tracker: DialogueStateTracker) -> bool:
-        return not tracker.is_paused() or (
+        return not tracker.is_paused() or MessageProcessor._last_user_intent_is_restart(
+            tracker
+        )
+
+    @staticmethod
+    def _last_user_intent_is_restart(tracker: DialogueStateTracker) -> bool:
+        """Check if the last user intent is a restart intent."""
+        return (
             tracker.latest_message is not None
             and tracker.latest_message.intent.get(INTENT_NAME_KEY)
             == USER_INTENT_RESTART

rasa/core/utils.py

@@ -19,6 +19,7 @@ from rasa.core.constants import (
 )
 from rasa.core.lock_store import InMemoryLockStore, LockStore, RedisLockStore
 from rasa.shared.constants import DEFAULT_ENDPOINTS_PATH, TCP_PROTOCOL
+from rasa.shared.core.constants import SlotMappingType
 from rasa.shared.core.trackers import DialogueStateTracker
 from rasa.utils.endpoints import (
     EndpointConfig,
@@ -30,6 +31,7 @@ from rasa.utils.io import write_yaml
 if TYPE_CHECKING:
     from rasa.core.nlg import NaturalLanguageGenerator
     from rasa.shared.core.domain import Domain
+    from rasa.shared.core.flows.flows_list import FlowsList
 
 structlogger = structlog.get_logger()
 
@@ -364,3 +366,54 @@ def add_bot_utterance_metadata(
         ]
 
     return message
+
+
+def should_force_slot_filling(
+    tracker: Optional[DialogueStateTracker], flows: "FlowsList"
+) -> Tuple[bool, Optional[str]]:
+    """Check if the flow should force slot filling.
+
+    This is only valid when the flow is at a collect information step which
+    has set `force_slot_filling` to true and the slot has a valid `from_text` mapping.
+
+    Args:
+        tracker: The dialogue state tracker.
+        flows: The list of flows.
+
+    Returns:
+        A tuple of a boolean indicating if the flow should force slot filling
+        and the name of the slot if applicable.
+    """
+    from rasa.dialogue_understanding.processor.command_processor import (
+        get_current_collect_step,
+    )
+
+    if tracker is None:
+        structlogger.error(
+            "slot.force_slot_filling.error",
+            event_info="Tracker is None. Cannot force slot filling.",
+        )
+        return False, None
+
+    stack = tracker.stack
+    step = get_current_collect_step(stack, flows)
+    if step is None or not step.force_slot_filling:
+        return False, None
+
+    slot_name = step.collect
+    slot = tracker.slots.get(slot_name)
+
+    if not slot:
+        structlogger.debug(
+            "slot.force_slot_filling.error",
+            event_info=f"Slot '{slot_name}' not found in tracker. "
+            f"Cannot force slot filling. "
+            f"Please check if the slot is defined in the domain.",
+        )
+        return False, None
+
+    for slot_mapping in slot.mappings:
+        if slot_mapping.type == SlotMappingType.FROM_TEXT:
+            return True, slot_name
+
+    return False, None

rasa/dialogue_understanding/coexistence/llm_based_router.py

@@ -35,6 +35,10 @@ from rasa.shared.exceptions import FileIOException, InvalidConfigException
 from rasa.shared.nlu.constants import COMMANDS, TEXT
 from rasa.shared.nlu.training_data.message import Message
 from rasa.shared.nlu.training_data.training_data import TrainingData
+from rasa.shared.utils.constants import (
+    LOG_COMPONENT_SOURCE_METHOD_FINGERPRINT_ADDON,
+    LOG_COMPONENT_SOURCE_METHOD_INIT,
+)
 from rasa.shared.utils.health_check.llm_health_check_mixin import LLMHealthCheckMixin
 from rasa.shared.utils.io import deep_container_fingerprint
 from rasa.shared.utils.llm import (
@@ -107,6 +111,8 @@ class LLMBasedRouter(LLMHealthCheckMixin, GraphComponent):
             or get_prompt_template(
                 config.get(PROMPT_CONFIG_KEY),
                 DEFAULT_COMMAND_PROMPT_TEMPLATE,
+                log_source_component=LLMBasedRouter.__name__,
+                log_source_method=LOG_COMPONENT_SOURCE_METHOD_INIT,
             ).strip()
         )
 
@@ -318,6 +324,8 @@ class LLMBasedRouter(LLMHealthCheckMixin, GraphComponent):
         prompt_template = get_prompt_template(
             config.get(PROMPT_CONFIG_KEY),
             DEFAULT_COMMAND_PROMPT_TEMPLATE,
+            log_source_component=LLMBasedRouter.__name__,
+            log_source_method=LOG_COMPONENT_SOURCE_METHOD_FINGERPRINT_ADDON,
         )
 
         llm_config = resolve_model_client_config(

rasa/dialogue_understanding/commands/cancel_flow_command.py

@@ -1,6 +1,5 @@
 from __future__ import annotations
 
-import copy
 import re
 from dataclasses import dataclass
 from typing import Any, Dict, List
@@ -13,10 +12,11 @@ from rasa.dialogue_understanding.commands.command_syntax_manager import (
     CommandSyntaxVersion,
 )
 from rasa.dialogue_understanding.patterns.cancel import CancelPatternFlowStackFrame
-from rasa.dialogue_understanding.patterns.clarify import ClarifyPatternFlowStackFrame
 from rasa.dialogue_understanding.stack.dialogue_stack import DialogueStack
-from rasa.dialogue_understanding.stack.frames import UserFlowStackFrame
-from rasa.dialogue_understanding.stack.frames.flow_stack_frame import FlowStackFrameType
+from rasa.dialogue_understanding.stack.frames.flow_stack_frame import (
+    FlowStackFrameType,
+    UserFlowStackFrame,
+)
 from rasa.dialogue_understanding.stack.utils import top_user_flow_frame
 from rasa.shared.core.events import Event, FlowCancelled
 from rasa.shared.core.flows import FlowsList
@@ -95,8 +95,6 @@ class CancelFlowCommand(Command):
         original_stack = original_tracker.stack
 
         applied_events: List[Event] = []
-        # capture the top frame before we push new frames onto the stack
-        initial_top_frame = stack.top()
         user_frame = top_user_flow_frame(original_stack)
         current_flow = user_frame.flow(all_flows) if user_frame else None
 
@@ -123,21 +121,6 @@ class CancelFlowCommand(Command):
         if user_frame:
             applied_events.append(FlowCancelled(user_frame.flow_id, user_frame.step_id))
 
-        if initial_top_frame and isinstance(
-            initial_top_frame, ClarifyPatternFlowStackFrame
-        ):
-            structlogger.debug(
-                "command_executor.cancel_flow.cancel_clarification_options",
-                clarification_options=initial_top_frame.clarification_options,
-            )
-            applied_events += cancel_all_pending_clarification_options(
-                initial_top_frame,
-                original_stack,
-                canceled_frames,
-                all_flows,
-                stack,
-            )
-
         return applied_events + tracker.create_stack_updated_events(stack)
 
     def __hash__(self) -> int:
@@ -172,41 +155,3 @@ class CancelFlowCommand(Command):
             CommandSyntaxManager.get_syntax_version(),
             mapper[CommandSyntaxManager.get_default_syntax_version()],
         )
-
-
-def cancel_all_pending_clarification_options(
-    initial_top_frame: ClarifyPatternFlowStackFrame,
-    original_stack: DialogueStack,
-    canceled_frames: List[str],
-    all_flows: FlowsList,
-    stack: DialogueStack,
-) -> List[FlowCancelled]:
-    """Cancel all pending clarification options.
-
-    This is a special case when the assistant asks the user to clarify
-    which pending digression flow to start after the completion of an active flow.
-    If the user chooses to cancel all options, this function takes care of
-    updating the stack by removing all pending flow stack frames
-    listed as clarification options.
-    """
-    clarification_names = set(initial_top_frame.names)
-    to_be_canceled_frames = []
-    applied_events = []
-    for frame in reversed(original_stack.frames):
-        if frame.frame_id in canceled_frames:
-            continue
-
-        to_be_canceled_frames.append(frame.frame_id)
-        if isinstance(frame, UserFlowStackFrame):
-            readable_flow_name = frame.flow(all_flows).readable_name()
-            if readable_flow_name in clarification_names:
-                stack.push(
-                    CancelPatternFlowStackFrame(
-                        canceled_name=readable_flow_name,
-                        canceled_frames=copy.deepcopy(to_be_canceled_frames),
-                    )
-                )
-                applied_events.append(FlowCancelled(frame.flow_id, frame.step_id))
-                to_be_canceled_frames.clear()
-
-    return applied_events

rasa/dialogue_understanding/commands/knowledge_answer_command.py

@@ -65,7 +65,7 @@ class KnowledgeAnswerCommand(FreeFormAnswerCommand):
         """Converts the command to a DSL string."""
         mapper = {
             CommandSyntaxVersion.v1: "SearchAndReply()",
-            CommandSyntaxVersion.v2: "search and reply",
+            CommandSyntaxVersion.v2: "provide info",
         }
         return mapper.get(
             CommandSyntaxManager.get_syntax_version(),
@@ -81,7 +81,7 @@ class KnowledgeAnswerCommand(FreeFormAnswerCommand):
     def regex_pattern() -> str:
         mapper = {
             CommandSyntaxVersion.v1: r"SearchAndReply\(\)",
-            CommandSyntaxVersion.v2: r"""^[\s\W\d]*search and reply['"`]*$""",
+            CommandSyntaxVersion.v2: r"""^[\s\W\d]*provide info['"`]*$""",
         }
         return mapper.get(
             CommandSyntaxManager.get_syntax_version(),

rasa/dialogue_understanding/commands/start_flow_command.py

@@ -11,11 +11,6 @@ from rasa.dialogue_understanding.commands.command_syntax_manager import (
     CommandSyntaxManager,
     CommandSyntaxVersion,
 )
-from rasa.dialogue_understanding.patterns.clarify import FLOW_PATTERN_CLARIFICATION
-from rasa.dialogue_understanding.patterns.continue_interrupted import (
-    ContinueInterruptedPatternFlowStackFrame,
-)
-from rasa.dialogue_understanding.stack.dialogue_stack import DialogueStack
 from rasa.dialogue_understanding.stack.frames.flow_stack_frame import (
     FlowStackFrameType,
     UserFlowStackFrame,
@@ -77,10 +72,6 @@ class StartFlowCommand(Command):
         applied_events: List[Event] = []
 
         if self.flow in user_flows_on_the_stack(stack):
-            top_frame = stack.top()
-            if top_frame is not None and top_frame.type() == FLOW_PATTERN_CLARIFICATION:
-                return self.change_flow_frame_position_in_the_stack(stack, tracker)
-
             structlogger.debug(
                 "command_executor.skip_command.already_started_flow", command=self
             )
@@ -149,35 +140,3 @@ class StartFlowCommand(Command):
             CommandSyntaxManager.get_syntax_version(),
             mapper[CommandSyntaxManager.get_default_syntax_version()],
         )
-
-    def change_flow_frame_position_in_the_stack(
-        self, stack: DialogueStack, tracker: DialogueStateTracker
-    ) -> List[Event]:
-        """Changes the position of the flow frame in the stack.
-
-        This is a special case when pattern clarification is the active flow and
-        the same flow is selected to start. In this case, the existing flow frame
-        should be moved up in the stack.
-        """
-        frames = stack.frames[:]
-
-        for idx, frame in enumerate(frames):
-            if isinstance(frame, UserFlowStackFrame) and frame.flow_id == self.flow:
-                structlogger.debug(
-                    "command_executor.change_flow_position_during_clarification",
-                    command=self,
-                    index=idx,
-                )
-                # pop the continue interrupted flow frame if it exists
-                next_frame = frames[idx + 1] if idx + 1 < len(frames) else None
-                if (
-                    isinstance(next_frame, ContinueInterruptedPatternFlowStackFrame)
-                    and next_frame.previous_flow_name == self.flow
-                ):
-                    stack.frames.pop(idx + 1)
-                # move up the existing flow from the stack
-                stack.frames.pop(idx)
-                stack.push(frame)
-                return tracker.create_stack_updated_events(stack)
-
-        return []

rasa/dialogue_understanding/generator/command_generator.py

@@ -4,6 +4,7 @@ from typing import Any, Dict, List, Optional, Set, Text, Tuple
 import structlog
 
 from rasa.dialogue_understanding.commands import (
+    CannotHandleCommand,
     Command,
     CorrectSlotsCommand,
     ErrorCommand,
@@ -107,6 +108,14 @@ class CommandGenerator:
             commands = self._check_commands_against_startable_flows(
                 commands, startable_flows
             )
+
+            # During force slot filling, keep only the command that sets the
+            # slot asked by the active collect step.
+            # Or return a CannotHandleCommand if no matching command is found.
+            commands = self._filter_commands_during_force_slot_filling(
+                commands, available_flows, tracker
+            )
+
             commands_dicts = [command.as_dict() for command in commands]
             message.set(COMMANDS, commands_dicts, add_to_output=True)
 
@@ -370,6 +379,64 @@ class CommandGenerator:
             Command.command_from_json(command) for command in message.get(COMMANDS, [])
         ]
 
+    @staticmethod
+    def _filter_commands_during_force_slot_filling(
+        commands: List[Command],
+        available_flows: FlowsList,
+        tracker: Optional[DialogueStateTracker] = None,
+    ) -> List[Command]:
+        """Filter commands during a collect step that has set `force_slot_filling`.
+
+        Args:
+            commands: The commands to filter.
+            available_flows: The available flows.
+            tracker: The tracker.
+
+        Returns:
+            The filtered commands.
+        """
+        from rasa.dialogue_understanding.processor.command_processor import (
+            get_current_collect_step,
+        )
+
+        if tracker is None:
+            structlogger.error(
+                "command_generator.filter_commands_during_force_slot_filling.tracker_not_found",
+            )
+            return commands
+
+        stack = tracker.stack
+        step = get_current_collect_step(stack, available_flows)
+
+        if step is None or not step.force_slot_filling:
+            return commands
+
+        # Retain only the command that sets the slot asked by
+        # the active collect step
+        filtered_commands: List[Command] = [
+            command
+            for command in commands
+            if (isinstance(command, SetSlotCommand) and command.name == step.collect)
+        ]
+
+        if not filtered_commands:
+            # If no commands were predicted, we need to return a CannotHandleCommand
+            structlogger.debug(
+                "command_generator.filter_commands_during_force_slot_filling.no_commands",
+                event_info=f"The command generator did not find any SetSlot "
+                f"command at the collect step for the slot '{step.collect}'. "
+                f"Returning a CannotHandleCommand instead.",
+            )
+            return [CannotHandleCommand()]
+
+        structlogger.debug(
+            "command_generator.filter_commands_during_force_slot_filling.filtered_commands",
+            slot_name=step.collect,
+            filtered_commands=filtered_commands,
+        )
+
+        return filtered_commands
+
 
 def gather_slot_names(commands: List[Command]) -> Set[str]:
     """Gather all slot names from the commands."""

rasa/dialogue_understanding/generator/flow_retrieval.py

@@ -52,7 +52,6 @@ from rasa.shared.utils.llm import (
     USER,
     allowed_values_for_slot,
     embedder_factory,
-    get_prompt_template,
     resolve_model_client_config,
     tracker_as_readable_transcript,
 )
@@ -103,9 +102,7 @@ class FlowRetrieval(EmbeddingsHealthCheckMixin):
             self.config.get(EMBEDDINGS_CONFIG_KEY), FlowRetrieval.__name__
         )
         self.vector_store: Optional[FAISS] = None
-        self.flow_document_template = get_prompt_template(
-            None, DEFAULT_FLOW_DOCUMENT_TEMPLATE
-        )
+        self.flow_document_template = DEFAULT_FLOW_DOCUMENT_TEMPLATE
         self._model_storage = model_storage
         self._resource = resource