revamped prompt action

kairon/actions/definitions/prompt.py

@@ -152,27 +152,31 @@ async def __get_llm_params(self, k_faq_action_config: dict, dispatcher: Collecti
         similarity_prompt = []
         params = {}
         num_bot_responses = k_faq_action_config['num_bot_responses']
-        for prompt in k_faq_action_config['llm_prompts']:
-            if prompt['type'] == LlmPromptType.system.value and prompt['is_enabled']:
-                system_prompt = f"{prompt['data']}\n"
-            elif prompt['type'] == LlmPromptType.user.value and prompt['is_enabled']:
+        system_prompt = k_faq_action_config.get("system_prompt", "")
+        user_prompt = k_faq_action_config.get("user_prompt", "")
+
+        for prompt in k_faq_action_config.get("contexts", []):
+            if prompt['type'] == LlmPromptType.user.value and prompt['is_enabled']:
                 if prompt['source'] == LlmPromptSource.history.value:
                     history_prompt = ActionUtility.prepare_bot_responses(tracker, num_bot_responses)
+
                 elif prompt['source'] == LlmPromptSource.bot_content.value and prompt['is_enabled']:
                     use_similarity_prompt = True
-                    hyperparameters = prompt.get("hyperparameters", {})
-                    similarity_prompt.append({'similarity_prompt_name': prompt['name'],
-                                              'similarity_prompt_instructions': prompt['instructions'],
-                                              'collection': prompt['data'],
-                                              'use_similarity_prompt': use_similarity_prompt,
-                                              'top_results': hyperparameters.get('top_results', 10),
-                                              'similarity_threshold': hyperparameters.get('similarity_threshold',
-                                                                                          0.70)})
+                    sim_cfg = prompt.get("similarity_config") or {}
+                    top_results = sim_cfg.get("top_results", 10)
+                    similarity_threshold = sim_cfg.get("similarity_threshold", 0.7)
+                    similarity_prompt.append({
+                        'similarity_prompt_name': prompt['name'],
+                        'collection': prompt['data'],
+                        'use_similarity_prompt': use_similarity_prompt,
+                        'top_results': top_results,
+                        'similarity_threshold': similarity_threshold
+                    })
+
                 elif prompt['source'] == LlmPromptSource.slot.value:
                     slot_data = tracker.get_slot(prompt['data'])
                     context_prompt += f"{prompt['name']}:\n{slot_data}\n"
-                    if prompt['instructions']:
-                        context_prompt += f"Instructions on how to use {prompt['name']}:\n{prompt['instructions']}\n\n"
+
                 elif prompt['source'] == LlmPromptSource.crud.value:
                     crud_config = prompt.get('crud_config', {})
                     collections = crud_config.get('collections', [])
@@ -210,29 +214,26 @@ async def __get_llm_params(self, k_faq_action_config: dict, dispatcher: Collecti
                         data_list = [rec["data"] for rec in records]
                         context_prompt += f"Collection data for {prompt['name']}:\n{data_list}\n"
 
-                    if prompt['instructions']:
-                        context_prompt += f"Instructions on how to use collection {prompt['name']}:\n{prompt['instructions']}\n\n"
                 elif prompt['source'] == LlmPromptSource.action.value:
                     action = ActionFactory.get_instance(self.bot, prompt['data'])
                     await action.execute(dispatcher, tracker, domain, **kwargs)
                     if action.is_success:
                         response = action.response
                         context_prompt += f"{prompt['name']}:\n{response}\n"
-                        if prompt['instructions']:
-                            context_prompt += f"Instructions on how to use {prompt['name']}:\n{prompt['instructions']}\n\n"
+                        # if prompt.get('instructions'):
+                        #     context_prompt += f"Instructions on how to use {prompt['name']}:\n{prompt['instructions']}\n\n"
+
                 else:
                     context_prompt += f"{prompt['name']}:\n{prompt['data']}\n"
-                    if prompt['instructions']:
-                        context_prompt += f"Instructions on how to use {prompt['name']}:\n{prompt['instructions']}\n\n"
+
             elif prompt['type'] == LlmPromptType.query.value and prompt['is_enabled']:
                 query_prompt += f"{prompt['name']}:\n{prompt['data']}\n"
-                if prompt['instructions']:
-                    query_prompt += f"Instructions on how to use {prompt['name']}:\n{prompt['instructions']}\n\n"
                 is_query_prompt_enabled = True
                 query_prompt_dict.update({'query_prompt': query_prompt, 'use_query_prompt': is_query_prompt_enabled})
 
         params["hyperparameters"] = k_faq_action_config['hyperparameters']
         params["system_prompt"] = system_prompt
+        params["user_prompt"] = user_prompt
         params["context_prompt"] = context_prompt
         params["query_prompt"] = query_prompt_dict
         params["previous_bot_responses"] = history_prompt

kairon/importer/validator/file_validator.py

@@ -716,7 +716,7 @@ def __validate_prompt_actions(prompt_actions: list, bot: Text = None):
                         action['num_bot_responses'] > 5 or not isinstance(action['num_bot_responses'], int)):
                     data_error.append(
                         f'num_bot_responses should not be greater than 5 and of type int: {action.get("name")}')
-                llm_prompts_errors = TrainingDataValidator.__validate_llm_prompts(action['llm_prompts'])
+                llm_prompts_errors = TrainingDataValidator.__validate_llm_prompts(action['contexts'])
                 if action.get('hyperparameters'):
                     llm_hyperparameters_errors = TrainingDataValidator.__validate_llm_prompts_hyperparameters(
                         action.get('hyperparameters'), action.get("llm_type", "openai"), bot)
@@ -753,13 +753,13 @@ def __validate_database_actions(database_actions: list):
         return data_error
 
     @staticmethod
-    def __validate_llm_prompts(llm_prompts: dict):
+    def __validate_llm_prompts(contexts: dict):
         error_list = []
         system_prompt_count = 0
         history_prompt_count = 0
-        for prompt in llm_prompts:
-            if prompt.get('hyperparameters') is not None:
-                hyperparameters = prompt.get('hyperparameters')
+        for prompt in contexts:
+            if prompt.get('similarity_config') is not None:
+                hyperparameters = prompt.get('similarity_config')
                 for key, value in hyperparameters.items():
                     if key == 'similarity_threshold':
                         if not (0.3 <= value <= 1.0) or not (
@@ -769,13 +769,11 @@ def __validate_llm_prompts(llm_prompts: dict):
                     if key == 'top_results' and (value > 30 or not isinstance(value, int)):
                         error_list.append("top_results should not be greater than 30 and of type int!")
 
-            if prompt.get('type') == 'system':
-                system_prompt_count += 1
-            elif prompt.get('source') == 'history':
+            if prompt.get('source') == 'history':
                 history_prompt_count += 1
             if prompt.get('type') not in ['user', 'system', 'query']:
                 error_list.append('Invalid prompt type')
-            if prompt.get('source') not in ['static', 'slot', 'action', 'history', 'bot_content']:
+            if prompt.get('source') not in ['static', 'slot', 'action', 'history', 'bot_content', 'crud']:
                 error_list.append('Invalid prompt source')
             if prompt.get('type') and not isinstance(prompt.get('type'), str):
                 error_list.append('type in LLM Prompts should be of type string.')

kairon/shared/actions/data_objects.py

@@ -792,11 +792,9 @@ def validate(self, clean=True):
     def clean(self):
         self.collections = [col.strip() for col in self.collections if col and col.strip()]
 
-class LlmPrompt(EmbeddedDocument):
+class Context(EmbeddedDocument):
     name = StringField(required=True)
-    hyperparameters = EmbeddedDocumentField(PromptHyperparameter)
     data = StringField()
-    instructions = StringField()
     type = StringField(
         required=True,
         choices=[
@@ -818,6 +816,7 @@ class LlmPrompt(EmbeddedDocument):
     )
     is_enabled = BooleanField(default=True)
     crud_config = EmbeddedDocumentField(CrudConfig)
+    similarity_config = EmbeddedDocumentField(PromptHyperparameter)
 
     def validate(self, clean=True):
         if (
@@ -833,8 +832,6 @@ def validate(self, clean=True):
                 self.crud_config.validate()
         elif self.crud_config:
             raise ValidationError("crud_config should only be provided when source is 'crud'")
-        if self.hyperparameters:
-            self.hyperparameters.validate()
         if self.source == LlmPromptSource.bot_content.value and Utility.check_empty_string(self.data):
             self.data = "default"
 
@@ -857,11 +854,12 @@ class PromptAction(Auditlog):
     timestamp = DateTimeField(default=datetime.utcnow)
     llm_type = StringField(default=DEFAULT_LLM, choices=Utility.get_llms())
     hyperparameters = DictField(default=Utility.get_default_llm_hyperparameters)
-    llm_prompts = EmbeddedDocumentListField(LlmPrompt, required=True)
-    instructions = ListField(StringField())
+    contexts = EmbeddedDocumentListField(Context)
+    system_prompt = StringField(required=True)
+    user_prompt = StringField(required=True)
     set_slots = EmbeddedDocumentListField(SetSlotsFromResponse)
     dispatch_response = BooleanField(default=True)
-    process_media=BooleanField(default=False)
+    process_media = BooleanField(default=False)
     status = BooleanField(default=True)
 
     meta = {"indexes": [{"fields": ["bot", ("bot", "name", "status")]}]}
@@ -876,13 +874,13 @@ def validate(self, clean=True):
             self.clean()
         if self.num_bot_responses > 5:
             raise ValidationError("num_bot_responses should not be greater than 5")
-        if not self.llm_prompts:
+        if not self.contexts:
             raise ValidationError("llm_prompts are required!")
-        for prompts in self.llm_prompts:
+        for prompts in self.contexts:
             prompts.validate()
         dict_data = self.to_mongo().to_dict()
         Utility.validate_kairon_faq_llm_prompts(
-            dict_data["llm_prompts"], ValidationError
+            dict_data["contexts"], ValidationError
         )
         Utility.validate_llm_hyperparameters(
             dict_data["hyperparameters"], self.llm_type, self.bot, ValidationError

kairon/shared/chat/user_media.py

@@ -448,14 +448,6 @@ def add_media_extraction_flow_if_not_exist( bot: str):
         try:
             instructions = ['Extract all relevant information from the media file and return as a markdown']
 
-            llm_prompts = [
-                {'name': 'System Prompt',
-                 'instructions': f'{instructions[0]}',
-                 'type': 'system',
-                 'data': 'Extract information',
-                 },
-            ]
-
             action_name = f"{UserMedia.MEDIA_EXTRACTION_FLOW_NAME}_prompt_action"
             if not Actions.objects(bot=bot, name=action_name).first():
                 action = Actions(
@@ -471,8 +463,7 @@ def add_media_extraction_flow_if_not_exist( bot: str):
             if not PromptAction.objects(bot=bot, name=action_name).first():
                 prompt_action = PromptAction(
                     name=action_name,
-                    instructions=instructions,
-                    llm_prompts=llm_prompts,
+                    system_prompt = f'{instructions[0]}',
                     dispatch_response=True,
                     process_media=True,
                     bot=bot,

kairon/shared/cognition/processor.py

@@ -314,7 +314,7 @@ def find_matching_metadata(bot: Text, data: Any, collection: Text = None):
 
     @staticmethod
     def validate_collection_name(bot: Text, collection: Text):
-        prompt_action = list(PromptAction.objects(bot=bot, llm_prompts__data__iexact=collection))
+        prompt_action = list(PromptAction.objects(bot=bot, contexts__data__iexact=collection))
         database_action = list(DatabaseAction.objects(bot=bot, collection__iexact=collection))
         if prompt_action:
             raise AppException(f'Cannot remove collection {collection} linked to action "{prompt_action[0].name}"!')

kairon/shared/data/data_models.py

@@ -1102,9 +1102,9 @@ class CrudConfigRequest(BaseModel):
     result_limit: int = 10
     query_source: Optional[Literal["value", "slot"]] = None
 
-class LlmPromptRequest(BaseModel, use_enum_values=True):
+class Context(BaseModel, use_enum_values=True):
     name: str
-    hyperparameters: PromptHyperparameters = None
+    similarity_config: Optional[PromptHyperparameters] = None
     data: str = None
     instructions: str = None
     type: LlmPromptType
@@ -1135,8 +1135,11 @@ def check(cls, values):
         else:
             values.pop('crud_config', None)
 
-        if values.get('source') == LlmPromptSource.bot_content.value and Utility.check_empty_string(values.get('data')):
-            values['data'] = "default"
+        if values.get('source') == LlmPromptSource.bot_content.value:
+            if Utility.check_empty_string(values.get('data')):
+                values['data'] = "default"
+        else:
+            values.pop('similarity_config', None)
 
         return values
 
@@ -1153,8 +1156,9 @@ class PromptActionConfigUploadValidation(BaseModel):
     user_question: UserQuestionModel = UserQuestionModel()
     llm_type: str
     hyperparameters: dict
-    llm_prompts: List[LlmPromptRequest]
-    instructions: List[str] = []
+    contexts: List[Context] = []
+    system_prompt: str
+    user_prompt: str
     set_slots: List[SetSlotsUsingActionResponse] = []
     dispatch_response: bool = True
 
@@ -1166,8 +1170,9 @@ class PromptActionConfigRequest(BaseModel):
     user_question: UserQuestionModel = UserQuestionModel()
     llm_type: str
     hyperparameters: dict
-    llm_prompts: List[LlmPromptRequest]
-    instructions: List[str] = []
+    contexts: List[Context] = Field(default_factory=list)
+    system_prompt: str
+    user_prompt: str
     set_slots: List[SetSlotsUsingActionResponse] = []
     dispatch_response: bool = True
     process_media: bool = False
@@ -1180,7 +1185,7 @@ def validate_llm_type(cls, v, values, **kwargs):
             raise ValueError("Invalid llm type")
         return v
 
-    @validator("llm_prompts")
+    @validator("contexts")
     def validate_llm_prompts(cls, v, values, **kwargs):
         from kairon.shared.utils import Utility
 
@@ -1195,15 +1200,34 @@ def validate_num_bot_responses(cls, v, values, **kwargs):
             raise ValueError("num_bot_responses should not be greater than 5")
         return v
 
-    @validator("hyperparameters")
-    def validate_hyperparameters(cls, v, values, **kwargs):
+    @validator("user_prompt", "system_prompt", pre=True, always=True)
+    def validate_user_and_system_prompts(cls, v, field):
         from kairon.shared.utils import Utility
-        bot = values.get('bot')
-        llm_type = values.get('llm_type')
-        if llm_type and v:
-            Utility.validate_llm_hyperparameters(v, llm_type, bot, ValueError)
+
+        if isinstance(v, list):
+            if len(v) != 1:
+                raise ValueError(f"Only one {field.name.replace('_', ' ')} is allowed.")
+            v = v[0]
+
+        if Utility.check_empty_string(v):
+            raise ValueError(f"{field.name.replace('_', ' ').capitalize()} cannot be empty!")
+
+        if not isinstance(v, str):
+            raise ValueError(f"{field.name.replace('_', ' ').capitalize()} must be a string.")
+
         return v
 
+    @root_validator()
+    def validate_hyperparameters(cls, values):
+        from kairon.shared.utils import Utility
+        bot = values.get('bot')
+        llm_type = values.get("llm_type")
+        hyperparams = values.get("hyperparameters")
+        if llm_type and hyperparams:
+            Utility.validate_llm_hyperparameters(hyperparams, llm_type, bot, ValueError)
+
+        return values
+
     @root_validator(pre=True)
     def validate_required_fields(cls, values):
         bot = values.get('bot')

kairon/shared/data/data_validation.py

@@ -89,7 +89,7 @@ def validate_prompt_action(bot: str, data: dict):
                 data['num_bot_responses'] > 5 or not isinstance(data['num_bot_responses'], int)):
             data_error.append(
                 f'num_bot_responses should not be greater than 5 and of type int: {data.get("name")}')
-        llm_prompts_errors = DataValidation.validate_llm_prompts(data['llm_prompts'])
+        llm_prompts_errors = DataValidation.validate_llm_prompts(data['contexts'])
         data_error.extend(llm_prompts_errors)
         if hyperparameters := data.get('hyperparameters'):
             if not data.get("llm_type"):
@@ -132,19 +132,18 @@ def validate_llm_prompts_hyperparameters(hyperparameters: dict, llm_type: str, b
         return error_list
 
     @staticmethod
-    def validate_llm_prompts(llm_prompts: list):
+    def validate_llm_prompts(contexts: list):
         error_list = []
         system_prompt_count = 0
         history_prompt_count = 0
-        for prompt in llm_prompts:
-            if prompt.get('hyperparameters') is not None:
-                hyperparameters = prompt.get('hyperparameters')
+        for prompt in contexts:
+            if prompt.get('similarity_config') is not None:
+                hyperparameters = prompt.get('similarity_config')
                 for key, value in hyperparameters.items():
                     if key == 'similarity_threshold':
                         if not (0.3 <= value <= 1.0) or not (
                                 isinstance(value, float) or isinstance(value, int)):
-                            error_list.append(
-                                "similarity_threshold should be within 0.3 and 1.0 and of type int or float!")
+                            error_list.append("similarity_threshold should be within 0.3 and 1.0 and of type int or float!")
                     if key == 'top_results' and (value > 30 or not isinstance(value, int)):
                         error_list.append("top_results should not be greater than 30 and of type int!")
 
@@ -154,7 +153,7 @@ def validate_llm_prompts(llm_prompts: list):
                 history_prompt_count += 1
             if prompt.get('type') not in ['user', 'system', 'query']:
                 error_list.append('Invalid prompt type')
-            if prompt.get('source') not in ['static', 'slot', 'action', 'history', 'bot_content']:
+            if prompt.get('source') not in ['static', 'slot', 'action', 'history', 'bot_content', 'crud']:
                 error_list.append('Invalid prompt source')
             if prompt.get('type') and not isinstance(prompt.get('type'), str):
                 error_list.append('type in LLM Prompts should be of type string.')