Model Runtime (#1858)
Co-authored-by: StyleZhang <jasonapring2015@outlook.com> Co-authored-by: Garfield Dai <dai.hai@foxmail.com> Co-authored-by: chenhe <guchenhe@gmail.com> Co-authored-by: jyong <jyong@dify.ai> Co-authored-by: Joel <iamjoel007@gmail.com> Co-authored-by: Yeuoly <admin@srmxy.cn>
This commit is contained in:
takatost
@@ -2,131 +2,199 @@ import json
|
||||
import os
|
||||
import re
|
||||
import enum
|
||||
from typing import List, Optional, Tuple
|
||||
from typing import List, Optional, Tuple, cast
|
||||
|
||||
from langchain.memory.chat_memory import BaseChatMemory
|
||||
from langchain.schema import BaseMessage
|
||||
|
||||
from core.model_providers.models.entity.model_params import ModelMode
|
||||
from core.model_providers.models.entity.message import PromptMessage, MessageType, to_prompt_messages, PromptMessageFile
|
||||
from core.model_providers.models.llm.base import BaseLLM
|
||||
from core.model_providers.models.llm.baichuan_model import BaichuanModel
|
||||
from core.model_providers.models.llm.huggingface_hub_model import HuggingfaceHubModel
|
||||
from core.model_providers.models.llm.openllm_model import OpenLLMModel
|
||||
from core.model_providers.models.llm.xinference_model import XinferenceModel
|
||||
from core.entities.application_entities import ModelConfigEntity, PromptTemplateEntity, \
|
||||
AdvancedCompletionPromptTemplateEntity
|
||||
from core.file.file_obj import FileObj
|
||||
from core.memory.token_buffer_memory import TokenBufferMemory
|
||||
from core.model_runtime.entities.message_entities import PromptMessage, SystemPromptMessage, UserPromptMessage, \
|
||||
TextPromptMessageContent, PromptMessageRole, AssistantPromptMessage
|
||||
from core.model_runtime.entities.model_entities import ModelPropertyKey
|
||||
from core.model_runtime.model_providers.__base.large_language_model import LargeLanguageModel
|
||||
from core.prompt.prompt_builder import PromptBuilder
|
||||
from core.prompt.prompt_template import PromptTemplateParser
|
||||
from models.model import AppModelConfig
|
||||
|
||||
|
||||
class AppMode(enum.Enum):
|
||||
COMPLETION = 'completion'
|
||||
CHAT = 'chat'
|
||||
|
||||
@classmethod
|
||||
def value_of(cls, value: str) -> 'AppMode':
|
||||
"""
|
||||
Get value of given mode.
|
||||
|
||||
:param value: mode value
|
||||
:return: mode
|
||||
"""
|
||||
for mode in cls:
|
||||
if mode.value == value:
|
||||
return mode
|
||||
raise ValueError(f'invalid mode value {value}')
|
||||
|
||||
|
||||
class ModelMode(enum.Enum):
|
||||
COMPLETION = 'completion'
|
||||
CHAT = 'chat'
|
||||
|
||||
@classmethod
|
||||
def value_of(cls, value: str) -> 'ModelMode':
|
||||
"""
|
||||
Get value of given mode.
|
||||
|
||||
:param value: mode value
|
||||
:return: mode
|
||||
"""
|
||||
for mode in cls:
|
||||
if mode.value == value:
|
||||
return mode
|
||||
raise ValueError(f'invalid mode value {value}')
|
||||
|
||||
|
||||
class PromptTransform:
|
||||
def get_prompt(self,
|
||||
app_mode: str,
|
||||
pre_prompt: str,
|
||||
prompt_template_entity: PromptTemplateEntity,
|
||||
inputs: dict,
|
||||
query: str,
|
||||
files: List[PromptMessageFile],
|
||||
files: List[FileObj],
|
||||
context: Optional[str],
|
||||
memory: Optional[BaseChatMemory],
|
||||
model_instance: BaseLLM) -> \
|
||||
memory: Optional[TokenBufferMemory],
|
||||
model_config: ModelConfigEntity) -> \
|
||||
Tuple[List[PromptMessage], Optional[List[str]]]:
|
||||
app_mode = AppMode.value_of(app_mode)
|
||||
model_mode = ModelMode.value_of(model_config.mode)
|
||||
|
||||
app_mode_enum = AppMode(app_mode)
|
||||
model_mode_enum = model_instance.model_mode
|
||||
prompt_rules = self._read_prompt_rules_from_file(self._prompt_file_name(
|
||||
app_mode=app_mode,
|
||||
provider=model_config.provider,
|
||||
model=model_config.model
|
||||
))
|
||||
|
||||
prompt_rules = self._read_prompt_rules_from_file(self._prompt_file_name(app_mode, model_instance))
|
||||
|
||||
if app_mode_enum == AppMode.CHAT and model_mode_enum == ModelMode.CHAT:
|
||||
if app_mode == AppMode.CHAT and model_mode == ModelMode.CHAT:
|
||||
stops = None
|
||||
|
||||
prompt_messages = self._get_simple_chat_app_chat_model_prompt_messages(prompt_rules, pre_prompt, inputs,
|
||||
query, context, memory,
|
||||
model_instance, files)
|
||||
prompt_messages = self._get_simple_chat_app_chat_model_prompt_messages(
|
||||
prompt_rules=prompt_rules,
|
||||
pre_prompt=prompt_template_entity.simple_prompt_template,
|
||||
inputs=inputs,
|
||||
query=query,
|
||||
files=files,
|
||||
context=context,
|
||||
memory=memory,
|
||||
model_config=model_config
|
||||
)
|
||||
else:
|
||||
stops = prompt_rules.get('stops')
|
||||
if stops is not None and len(stops) == 0:
|
||||
stops = None
|
||||
|
||||
prompt_messages = self._get_simple_others_prompt_messages(prompt_rules, pre_prompt, inputs, query, context,
|
||||
memory,
|
||||
model_instance, files)
|
||||
prompt_messages = self._get_simple_others_prompt_messages(
|
||||
prompt_rules=prompt_rules,
|
||||
pre_prompt=prompt_template_entity.simple_prompt_template,
|
||||
inputs=inputs,
|
||||
query=query,
|
||||
files=files,
|
||||
context=context,
|
||||
memory=memory,
|
||||
model_config=model_config
|
||||
)
|
||||
return prompt_messages, stops
|
||||
|
||||
def get_advanced_prompt(self,
|
||||
app_mode: str,
|
||||
app_model_config: AppModelConfig,
|
||||
def get_advanced_prompt(self, app_mode: str,
|
||||
prompt_template_entity: PromptTemplateEntity,
|
||||
inputs: dict,
|
||||
query: str,
|
||||
files: List[PromptMessageFile],
|
||||
files: List[FileObj],
|
||||
context: Optional[str],
|
||||
memory: Optional[BaseChatMemory],
|
||||
model_instance: BaseLLM) -> List[PromptMessage]:
|
||||
|
||||
model_mode = app_model_config.model_dict['mode']
|
||||
|
||||
app_mode_enum = AppMode(app_mode)
|
||||
model_mode_enum = ModelMode(model_mode)
|
||||
memory: Optional[TokenBufferMemory],
|
||||
model_config: ModelConfigEntity) -> List[PromptMessage]:
|
||||
app_mode = AppMode.value_of(app_mode)
|
||||
model_mode = ModelMode.value_of(model_config.mode)
|
||||
|
||||
prompt_messages = []
|
||||
|
||||
if app_mode_enum == AppMode.CHAT:
|
||||
if model_mode_enum == ModelMode.COMPLETION:
|
||||
prompt_messages = self._get_chat_app_completion_model_prompt_messages(app_model_config, inputs, query,
|
||||
files, context, memory,
|
||||
model_instance)
|
||||
elif model_mode_enum == ModelMode.CHAT:
|
||||
prompt_messages = self._get_chat_app_chat_model_prompt_messages(app_model_config, inputs, query, files,
|
||||
context, memory, model_instance)
|
||||
elif app_mode_enum == AppMode.COMPLETION:
|
||||
if model_mode_enum == ModelMode.CHAT:
|
||||
prompt_messages = self._get_completion_app_chat_model_prompt_messages(app_model_config, inputs,
|
||||
files, context)
|
||||
elif model_mode_enum == ModelMode.COMPLETION:
|
||||
prompt_messages = self._get_completion_app_completion_model_prompt_messages(app_model_config, inputs,
|
||||
files, context)
|
||||
if app_mode == AppMode.CHAT:
|
||||
if model_mode == ModelMode.COMPLETION:
|
||||
prompt_messages = self._get_chat_app_completion_model_prompt_messages(
|
||||
prompt_template_entity=prompt_template_entity,
|
||||
inputs=inputs,
|
||||
query=query,
|
||||
context=context,
|
||||
memory=memory,
|
||||
model_config=model_config
|
||||
)
|
||||
elif model_mode == ModelMode.CHAT:
|
||||
prompt_messages = self._get_chat_app_chat_model_prompt_messages(
|
||||
prompt_template_entity=prompt_template_entity,
|
||||
inputs=inputs,
|
||||
query=query,
|
||||
files=files,
|
||||
context=context,
|
||||
memory=memory,
|
||||
model_config=model_config
|
||||
)
|
||||
elif app_mode == AppMode.COMPLETION:
|
||||
if model_mode == ModelMode.CHAT:
|
||||
prompt_messages = self._get_completion_app_chat_model_prompt_messages(
|
||||
prompt_template_entity=prompt_template_entity,
|
||||
inputs=inputs,
|
||||
files=files,
|
||||
context=context,
|
||||
)
|
||||
elif model_mode == ModelMode.COMPLETION:
|
||||
prompt_messages = self._get_completion_app_completion_model_prompt_messages(
|
||||
prompt_template_entity=prompt_template_entity,
|
||||
inputs=inputs,
|
||||
context=context,
|
||||
)
|
||||
|
||||
return prompt_messages
|
||||
|
||||
def _get_history_messages_from_memory(self, memory: BaseChatMemory,
|
||||
max_token_limit: int) -> str:
|
||||
def _get_history_messages_from_memory(self, memory: TokenBufferMemory,
|
||||
max_token_limit: int,
|
||||
human_prefix: Optional[str] = None,
|
||||
ai_prefix: Optional[str] = None) -> str:
|
||||
"""Get memory messages."""
|
||||
memory.max_token_limit = max_token_limit
|
||||
memory_key = memory.memory_variables[0]
|
||||
external_context = memory.load_memory_variables({})
|
||||
return external_context[memory_key]
|
||||
kwargs = {
|
||||
"max_token_limit": max_token_limit
|
||||
}
|
||||
|
||||
def _get_history_messages_list_from_memory(self, memory: BaseChatMemory,
|
||||
if human_prefix:
|
||||
kwargs['human_prefix'] = human_prefix
|
||||
|
||||
if ai_prefix:
|
||||
kwargs['ai_prefix'] = ai_prefix
|
||||
|
||||
return memory.get_history_prompt_text(
|
||||
**kwargs
|
||||
)
|
||||
|
||||
def _get_history_messages_list_from_memory(self, memory: TokenBufferMemory,
|
||||
max_token_limit: int) -> List[PromptMessage]:
|
||||
"""Get memory messages."""
|
||||
memory.max_token_limit = max_token_limit
|
||||
memory.return_messages = True
|
||||
memory_key = memory.memory_variables[0]
|
||||
external_context = memory.load_memory_variables({})
|
||||
memory.return_messages = False
|
||||
return to_prompt_messages(external_context[memory_key])
|
||||
return memory.get_history_prompt_messages(
|
||||
max_token_limit=max_token_limit
|
||||
)
|
||||
|
||||
def _prompt_file_name(self, mode: str, model_instance: BaseLLM) -> str:
|
||||
def _prompt_file_name(self, app_mode: AppMode, provider: str, model: str) -> str:
|
||||
# baichuan
|
||||
if isinstance(model_instance, BaichuanModel):
|
||||
return self._prompt_file_name_for_baichuan(mode)
|
||||
if provider == 'baichuan':
|
||||
return self._prompt_file_name_for_baichuan(app_mode)
|
||||
|
||||
baichuan_model_hosted_platforms = (HuggingfaceHubModel, OpenLLMModel, XinferenceModel)
|
||||
if isinstance(model_instance, baichuan_model_hosted_platforms) and 'baichuan' in model_instance.name.lower():
|
||||
return self._prompt_file_name_for_baichuan(mode)
|
||||
baichuan_supported_providers = ["huggingface_hub", "openllm", "xinference"]
|
||||
if provider in baichuan_supported_providers and 'baichuan' in model.lower():
|
||||
return self._prompt_file_name_for_baichuan(app_mode)
|
||||
|
||||
# common
|
||||
if mode == 'completion':
|
||||
if app_mode == AppMode.COMPLETION:
|
||||
return 'common_completion'
|
||||
else:
|
||||
return 'common_chat'
|
||||
|
||||
def _prompt_file_name_for_baichuan(self, mode: str) -> str:
|
||||
if mode == 'completion':
|
||||
def _prompt_file_name_for_baichuan(self, app_mode: AppMode) -> str:
|
||||
if app_mode == AppMode.COMPLETION:
|
||||
return 'baichuan_completion'
|
||||
else:
|
||||
return 'baichuan_chat'
|
||||
@@ -142,12 +210,14 @@ class PromptTransform:
|
||||
with open(json_file_path, 'r') as json_file:
|
||||
return json.load(json_file)
|
||||
|
||||
def _get_simple_chat_app_chat_model_prompt_messages(self, prompt_rules: dict, pre_prompt: str, inputs: dict,
|
||||
def _get_simple_chat_app_chat_model_prompt_messages(self, prompt_rules: dict,
|
||||
pre_prompt: str,
|
||||
inputs: dict,
|
||||
query: str,
|
||||
context: Optional[str],
|
||||
memory: Optional[BaseChatMemory],
|
||||
model_instance: BaseLLM,
|
||||
files: List[PromptMessageFile]) -> List[PromptMessage]:
|
||||
files: List[FileObj],
|
||||
memory: Optional[TokenBufferMemory],
|
||||
model_config: ModelConfigEntity) -> List[PromptMessage]:
|
||||
prompt_messages = []
|
||||
|
||||
context_prompt_content = ''
|
||||
@@ -174,20 +244,34 @@ class PromptTransform:
|
||||
|
||||
prompt = re.sub(r'<\|.*?\|>', '', prompt)
|
||||
|
||||
prompt_messages.append(PromptMessage(type=MessageType.SYSTEM, content=prompt))
|
||||
if prompt:
|
||||
prompt_messages.append(SystemPromptMessage(content=prompt))
|
||||
|
||||
self._append_chat_histories(memory, prompt_messages, model_instance)
|
||||
self._append_chat_histories(
|
||||
memory=memory,
|
||||
prompt_messages=prompt_messages,
|
||||
model_config=model_config
|
||||
)
|
||||
|
||||
prompt_messages.append(PromptMessage(type=MessageType.USER, content=query, files=files))
|
||||
if files:
|
||||
prompt_message_contents = [TextPromptMessageContent(data=query)]
|
||||
for file in files:
|
||||
prompt_message_contents.append(file.prompt_message_content)
|
||||
|
||||
prompt_messages.append(UserPromptMessage(content=prompt_message_contents))
|
||||
else:
|
||||
prompt_messages.append(UserPromptMessage(content=query))
|
||||
|
||||
return prompt_messages
|
||||
|
||||
def _get_simple_others_prompt_messages(self, prompt_rules: dict, pre_prompt: str, inputs: dict,
|
||||
def _get_simple_others_prompt_messages(self, prompt_rules: dict,
|
||||
pre_prompt: str,
|
||||
inputs: dict,
|
||||
query: str,
|
||||
context: Optional[str],
|
||||
memory: Optional[BaseChatMemory],
|
||||
model_instance: BaseLLM,
|
||||
files: List[PromptMessageFile]) -> List[PromptMessage]:
|
||||
memory: Optional[TokenBufferMemory],
|
||||
files: List[FileObj],
|
||||
model_config: ModelConfigEntity) -> List[PromptMessage]:
|
||||
context_prompt_content = ''
|
||||
if context and 'context_prompt' in prompt_rules:
|
||||
prompt_template = PromptTemplateParser(template=prompt_rules['context_prompt'])
|
||||
@@ -214,19 +298,23 @@ class PromptTransform:
|
||||
|
||||
if memory and 'histories_prompt' in prompt_rules:
|
||||
# append chat histories
|
||||
tmp_human_message = PromptBuilder.to_human_message(
|
||||
prompt_content=prompt + query_prompt,
|
||||
inputs={
|
||||
'query': query
|
||||
}
|
||||
tmp_human_message = UserPromptMessage(
|
||||
content=PromptBuilder.parse_prompt(
|
||||
prompt=prompt + query_prompt,
|
||||
inputs={
|
||||
'query': query
|
||||
}
|
||||
)
|
||||
)
|
||||
|
||||
rest_tokens = self._calculate_rest_token(tmp_human_message, model_instance)
|
||||
rest_tokens = self._calculate_rest_token([tmp_human_message], model_config)
|
||||
|
||||
memory.human_prefix = prompt_rules['human_prefix'] if 'human_prefix' in prompt_rules else 'Human'
|
||||
memory.ai_prefix = prompt_rules['assistant_prefix'] if 'assistant_prefix' in prompt_rules else 'Assistant'
|
||||
|
||||
histories = self._get_history_messages_from_memory(memory, rest_tokens)
|
||||
histories = self._get_history_messages_from_memory(
|
||||
memory=memory,
|
||||
max_token_limit=rest_tokens,
|
||||
ai_prefix=prompt_rules['human_prefix'] if 'human_prefix' in prompt_rules else 'Human',
|
||||
human_prefix=prompt_rules['assistant_prefix'] if 'assistant_prefix' in prompt_rules else 'Assistant'
|
||||
)
|
||||
prompt_template = PromptTemplateParser(template=prompt_rules['histories_prompt'])
|
||||
histories_prompt_content = prompt_template.format({'histories': histories})
|
||||
|
||||
@@ -246,7 +334,7 @@ class PromptTransform:
|
||||
|
||||
prompt = re.sub(r'<\|.*?\|>', '', prompt)
|
||||
|
||||
return [PromptMessage(content=prompt, files=files)]
|
||||
return [UserPromptMessage(content=prompt)]
|
||||
|
||||
def _set_context_variable(self, context: str, prompt_template: PromptTemplateParser, prompt_inputs: dict) -> None:
|
||||
if '#context#' in prompt_template.variable_keys:
|
||||
@@ -262,42 +350,63 @@ class PromptTransform:
|
||||
else:
|
||||
prompt_inputs['#query#'] = ''
|
||||
|
||||
def _set_histories_variable(self, memory: BaseChatMemory, raw_prompt: str, conversation_histories_role: dict,
|
||||
prompt_template: PromptTemplateParser, prompt_inputs: dict,
|
||||
model_instance: BaseLLM) -> None:
|
||||
def _set_histories_variable(self, memory: TokenBufferMemory,
|
||||
raw_prompt: str,
|
||||
role_prefix: AdvancedCompletionPromptTemplateEntity.RolePrefixEntity,
|
||||
prompt_template: PromptTemplateParser,
|
||||
prompt_inputs: dict,
|
||||
model_config: ModelConfigEntity) -> None:
|
||||
if '#histories#' in prompt_template.variable_keys:
|
||||
if memory:
|
||||
tmp_human_message = PromptBuilder.to_human_message(
|
||||
prompt_content=raw_prompt,
|
||||
inputs={'#histories#': '', **prompt_inputs}
|
||||
tmp_human_message = UserPromptMessage(
|
||||
content=PromptBuilder.parse_prompt(
|
||||
prompt=raw_prompt,
|
||||
inputs={'#histories#': '', **prompt_inputs}
|
||||
)
|
||||
)
|
||||
|
||||
rest_tokens = self._calculate_rest_token(tmp_human_message, model_instance)
|
||||
rest_tokens = self._calculate_rest_token([tmp_human_message], model_config)
|
||||
|
||||
memory.human_prefix = conversation_histories_role['user_prefix']
|
||||
memory.ai_prefix = conversation_histories_role['assistant_prefix']
|
||||
histories = self._get_history_messages_from_memory(memory, rest_tokens)
|
||||
histories = self._get_history_messages_from_memory(
|
||||
memory=memory,
|
||||
max_token_limit=rest_tokens,
|
||||
human_prefix=role_prefix.user,
|
||||
ai_prefix=role_prefix.assistant
|
||||
)
|
||||
prompt_inputs['#histories#'] = histories
|
||||
else:
|
||||
prompt_inputs['#histories#'] = ''
|
||||
|
||||
def _append_chat_histories(self, memory: BaseChatMemory, prompt_messages: list[PromptMessage],
|
||||
model_instance: BaseLLM) -> None:
|
||||
def _append_chat_histories(self, memory: TokenBufferMemory,
|
||||
prompt_messages: list[PromptMessage],
|
||||
model_config: ModelConfigEntity) -> None:
|
||||
if memory:
|
||||
rest_tokens = self._calculate_rest_token(prompt_messages, model_instance)
|
||||
|
||||
memory.human_prefix = MessageType.USER.value
|
||||
memory.ai_prefix = MessageType.ASSISTANT.value
|
||||
rest_tokens = self._calculate_rest_token(prompt_messages, model_config)
|
||||
histories = self._get_history_messages_list_from_memory(memory, rest_tokens)
|
||||
prompt_messages.extend(histories)
|
||||
|
||||
def _calculate_rest_token(self, prompt_messages: BaseMessage, model_instance: BaseLLM) -> int:
|
||||
def _calculate_rest_token(self, prompt_messages: list[PromptMessage], model_config: ModelConfigEntity) -> int:
|
||||
rest_tokens = 2000
|
||||
|
||||
if model_instance.model_rules.max_tokens.max:
|
||||
curr_message_tokens = model_instance.get_num_tokens(to_prompt_messages(prompt_messages))
|
||||
max_tokens = model_instance.model_kwargs.max_tokens
|
||||
rest_tokens = model_instance.model_rules.max_tokens.max - max_tokens - curr_message_tokens
|
||||
model_context_tokens = model_config.model_schema.model_properties.get(ModelPropertyKey.CONTEXT_SIZE)
|
||||
if model_context_tokens:
|
||||
model_type_instance = model_config.provider_model_bundle.model_type_instance
|
||||
model_type_instance = cast(LargeLanguageModel, model_type_instance)
|
||||
|
||||
curr_message_tokens = model_type_instance.get_num_tokens(
|
||||
model_config.model,
|
||||
model_config.credentials,
|
||||
prompt_messages
|
||||
)
|
||||
|
||||
max_tokens = 0
|
||||
for parameter_rule in model_config.model_schema.parameter_rules:
|
||||
if (parameter_rule.name == 'max_tokens'
|
||||
or (parameter_rule.use_template and parameter_rule.use_template == 'max_tokens')):
|
||||
max_tokens = (model_config.parameters.get(parameter_rule.name)
|
||||
or model_config.parameters.get(parameter_rule.use_template)) or 0
|
||||
|
||||
rest_tokens = model_context_tokens - max_tokens - curr_message_tokens
|
||||
rest_tokens = max(rest_tokens, 0)
|
||||
|
||||
return rest_tokens
|
||||
@@ -311,16 +420,15 @@ class PromptTransform:
|
||||
return prompt
|
||||
|
||||
def _get_chat_app_completion_model_prompt_messages(self,
|
||||
app_model_config: AppModelConfig,
|
||||
prompt_template_entity: PromptTemplateEntity,
|
||||
inputs: dict,
|
||||
query: str,
|
||||
files: List[PromptMessageFile],
|
||||
context: Optional[str],
|
||||
memory: Optional[BaseChatMemory],
|
||||
model_instance: BaseLLM) -> List[PromptMessage]:
|
||||
memory: Optional[TokenBufferMemory],
|
||||
model_config: ModelConfigEntity) -> List[PromptMessage]:
|
||||
|
||||
raw_prompt = app_model_config.completion_prompt_config_dict['prompt']['text']
|
||||
conversation_histories_role = app_model_config.completion_prompt_config_dict['conversation_histories_role']
|
||||
raw_prompt = prompt_template_entity.advanced_completion_prompt_template.prompt
|
||||
role_prefix = prompt_template_entity.advanced_completion_prompt_template.role_prefix
|
||||
|
||||
prompt_messages = []
|
||||
|
||||
@@ -331,29 +439,35 @@ class PromptTransform:
|
||||
|
||||
self._set_query_variable(query, prompt_template, prompt_inputs)
|
||||
|
||||
self._set_histories_variable(memory, raw_prompt, conversation_histories_role, prompt_template, prompt_inputs,
|
||||
model_instance)
|
||||
self._set_histories_variable(
|
||||
memory=memory,
|
||||
raw_prompt=raw_prompt,
|
||||
role_prefix=role_prefix,
|
||||
prompt_template=prompt_template,
|
||||
prompt_inputs=prompt_inputs,
|
||||
model_config=model_config
|
||||
)
|
||||
|
||||
prompt = self._format_prompt(prompt_template, prompt_inputs)
|
||||
|
||||
prompt_messages.append(PromptMessage(type=MessageType.USER, content=prompt, files=files))
|
||||
prompt_messages.append(UserPromptMessage(content=prompt))
|
||||
|
||||
return prompt_messages
|
||||
|
||||
def _get_chat_app_chat_model_prompt_messages(self,
|
||||
app_model_config: AppModelConfig,
|
||||
prompt_template_entity: PromptTemplateEntity,
|
||||
inputs: dict,
|
||||
query: str,
|
||||
files: List[PromptMessageFile],
|
||||
files: List[FileObj],
|
||||
context: Optional[str],
|
||||
memory: Optional[BaseChatMemory],
|
||||
model_instance: BaseLLM) -> List[PromptMessage]:
|
||||
raw_prompt_list = app_model_config.chat_prompt_config_dict['prompt']
|
||||
memory: Optional[TokenBufferMemory],
|
||||
model_config: ModelConfigEntity) -> List[PromptMessage]:
|
||||
raw_prompt_list = prompt_template_entity.advanced_chat_prompt_template.messages
|
||||
|
||||
prompt_messages = []
|
||||
|
||||
for prompt_item in raw_prompt_list:
|
||||
raw_prompt = prompt_item['text']
|
||||
raw_prompt = prompt_item.text
|
||||
|
||||
prompt_template = PromptTemplateParser(template=raw_prompt)
|
||||
prompt_inputs = {k: inputs[k] for k in prompt_template.variable_keys if k in inputs}
|
||||
@@ -362,20 +476,31 @@ class PromptTransform:
|
||||
|
||||
prompt = self._format_prompt(prompt_template, prompt_inputs)
|
||||
|
||||
prompt_messages.append(PromptMessage(type=MessageType(prompt_item['role']), content=prompt))
|
||||
if prompt_item.role == PromptMessageRole.USER:
|
||||
prompt_messages.append(UserPromptMessage(content=prompt))
|
||||
elif prompt_item.role == PromptMessageRole.SYSTEM and prompt:
|
||||
prompt_messages.append(SystemPromptMessage(content=prompt))
|
||||
elif prompt_item.role == PromptMessageRole.ASSISTANT:
|
||||
prompt_messages.append(AssistantPromptMessage(content=prompt))
|
||||
|
||||
self._append_chat_histories(memory, prompt_messages, model_instance)
|
||||
self._append_chat_histories(memory, prompt_messages, model_config)
|
||||
|
||||
prompt_messages.append(PromptMessage(type=MessageType.USER, content=query, files=files))
|
||||
if files:
|
||||
prompt_message_contents = [TextPromptMessageContent(data=query)]
|
||||
for file in files:
|
||||
prompt_message_contents.append(file.prompt_message_content)
|
||||
|
||||
prompt_messages.append(UserPromptMessage(content=prompt_message_contents))
|
||||
else:
|
||||
prompt_messages.append(UserPromptMessage(content=query))
|
||||
|
||||
return prompt_messages
|
||||
|
||||
def _get_completion_app_completion_model_prompt_messages(self,
|
||||
app_model_config: AppModelConfig,
|
||||
prompt_template_entity: PromptTemplateEntity,
|
||||
inputs: dict,
|
||||
files: List[PromptMessageFile],
|
||||
context: Optional[str]) -> List[PromptMessage]:
|
||||
raw_prompt = app_model_config.completion_prompt_config_dict['prompt']['text']
|
||||
raw_prompt = prompt_template_entity.advanced_completion_prompt_template.prompt
|
||||
|
||||
prompt_messages = []
|
||||
|
||||
@@ -386,21 +511,21 @@ class PromptTransform:
|
||||
|
||||
prompt = self._format_prompt(prompt_template, prompt_inputs)
|
||||
|
||||
prompt_messages.append(PromptMessage(type=MessageType(MessageType.USER), content=prompt, files=files))
|
||||
prompt_messages.append(UserPromptMessage(content=prompt))
|
||||
|
||||
return prompt_messages
|
||||
|
||||
def _get_completion_app_chat_model_prompt_messages(self,
|
||||
app_model_config: AppModelConfig,
|
||||
prompt_template_entity: PromptTemplateEntity,
|
||||
inputs: dict,
|
||||
files: List[PromptMessageFile],
|
||||
files: List[FileObj],
|
||||
context: Optional[str]) -> List[PromptMessage]:
|
||||
raw_prompt_list = app_model_config.chat_prompt_config_dict['prompt']
|
||||
raw_prompt_list = prompt_template_entity.advanced_chat_prompt_template.messages
|
||||
|
||||
prompt_messages = []
|
||||
|
||||
for prompt_item in raw_prompt_list:
|
||||
raw_prompt = prompt_item['text']
|
||||
raw_prompt = prompt_item.text
|
||||
|
||||
prompt_template = PromptTemplateParser(template=raw_prompt)
|
||||
prompt_inputs = {k: inputs[k] for k in prompt_template.variable_keys if k in inputs}
|
||||
@@ -409,11 +534,21 @@ class PromptTransform:
|
||||
|
||||
prompt = self._format_prompt(prompt_template, prompt_inputs)
|
||||
|
||||
prompt_messages.append(PromptMessage(type=MessageType(prompt_item['role']), content=prompt))
|
||||
if prompt_item.role == PromptMessageRole.USER:
|
||||
prompt_messages.append(UserPromptMessage(content=prompt))
|
||||
elif prompt_item.role == PromptMessageRole.SYSTEM and prompt:
|
||||
prompt_messages.append(SystemPromptMessage(content=prompt))
|
||||
elif prompt_item.role == PromptMessageRole.ASSISTANT:
|
||||
prompt_messages.append(AssistantPromptMessage(content=prompt))
|
||||
|
||||
for prompt_message in prompt_messages[::-1]:
|
||||
if prompt_message.type == MessageType.USER:
|
||||
prompt_message.files = files
|
||||
if prompt_message.role == PromptMessageRole.USER:
|
||||
if files:
|
||||
prompt_message_contents = [TextPromptMessageContent(data=prompt_message.content)]
|
||||
for file in files:
|
||||
prompt_message_contents.append(file.prompt_message_content)
|
||||
|
||||
prompt_message.content = prompt_message_contents
|
||||
break
|
||||
|
||||
return prompt_messages
|
||||
|
||||
Reference in New Issue
Block a user