+
**Dify** is an easy-to-use LLMOps platform designed to empower more people to create sustainable, AI-native applications. With visual orchestration for various application types, Dify offers out-of-the-box, ready-to-use applications that can also serve as Backend-as-a-Service APIs. Unify your development process with one API for plugins and datasets integration, and streamline your operations using a single interface for prompt engineering, visual analytics, and continuous improvement.
Applications created with Dify include:
diff --git a/README_CN.md b/README_CN.md
index 17cd027aee..e37399343e 100644
--- a/README_CN.md
+++ b/README_CN.md
@@ -8,6 +8,9 @@
[官方网站](https://dify.ai) • [文档](https://docs.dify.ai/v/zh-hans) • [Twitter](https://twitter.com/dify_ai) • [Discord](https://discord.gg/FngNHpbcY7)
+在 Product Hunt 上投我们一票吧 ↓
+
+
**Dify** 是一个易用的 LLMOps 平台，旨在让更多人可以创建可持续运营的原生 AI 应用。Dify 提供多种类型应用的可视化编排，应用可开箱即用，也能以“后端即服务”的 API 提供服务。
通过 Dify 创建的应用包含了：
diff --git a/README_JA.md b/README_JA.md
index 6b62747c9c..fed796d4f1 100644
--- a/README_JA.md
+++ b/README_JA.md
@@ -7,6 +7,10 @@
[Web サイト](https://dify.ai) • [ドキュメント](https://docs.dify.ai) • [Twitter](https://twitter.com/dify_ai) • [Discord](https://discord.gg/FngNHpbcY7)
+Product Huntで私たちに投票してください ↓
+
+
+
**Dify** は、より多くの人々が持続可能な AI ネイティブアプリケーションを作成できるように設計された、使いやすい LLMOps プラットフォームです。様々なアプリケーションタイプに対応したビジュアルオーケストレーションにより Dify は Backend-as-a-Service API としても機能する、すぐに使えるアプリケーションを提供します。プラグインやデータセットを統合するための1つの API で開発プロセスを統一し、プロンプトエンジニアリング、ビジュアル分析、継続的な改善のための1つのインターフェイスを使って業務を合理化します。
Difyで作成したアプリケーションは以下の通りです:
diff --git a/api/.env.example b/api/.env.example
index 5f307dc106..4e2d76f810 100644
--- a/api/.env.example
+++ b/api/.env.example
@@ -14,7 +14,7 @@ CONSOLE_URL=http://127.0.0.1:5001
API_URL=http://127.0.0.1:5001
# Web APP base URL
-APP_URL=http://127.0.0.1:5001
+APP_URL=http://127.0.0.1:3000
# celery configuration
CELERY_BROKER_URL=redis://:difyai123456@localhost:6379/1
diff --git a/api/controllers/console/app/__init__.py b/api/controllers/console/app/__init__.py
index 1f22ab30c6..f0c7956e0f 100644
--- a/api/controllers/console/app/__init__.py
+++ b/api/controllers/console/app/__init__.py
@@ -17,6 +17,6 @@ def _get_app(app_id, mode=None):
raise NotFound("App not found")
if mode and app.mode != mode:
- raise AppUnavailableError()
+ raise NotFound("The {} app not found".format(mode))
return app
diff --git a/api/controllers/console/app/error.py b/api/controllers/console/app/error.py
index c19f054be4..8923f90bf8 100644
--- a/api/controllers/console/app/error.py
+++ b/api/controllers/console/app/error.py
@@ -9,31 +9,33 @@ class AppNotFoundError(BaseHTTPException):
class ProviderNotInitializeError(BaseHTTPException):
error_code = 'provider_not_initialize'
- description = "Provider Token not initialize."
+ description = "No valid model provider credentials found. " \
+ "Please go to Settings -> Model Provider to complete your provider credentials."
code = 400
class ProviderQuotaExceededError(BaseHTTPException):
error_code = 'provider_quota_exceeded'
- description = "Provider quota exceeded."
+ description = "Your quota for Dify Hosted OpenAI has been exhausted. " \
+ "Please go to Settings -> Model Provider to complete your own provider credentials."
code = 400
class ProviderModelCurrentlyNotSupportError(BaseHTTPException):
error_code = 'model_currently_not_support'
- description = "GPT-4 currently not support."
+ description = "Dify Hosted OpenAI trial currently not support the GPT-4 model."
code = 400
class ConversationCompletedError(BaseHTTPException):
error_code = 'conversation_completed'
- description = "Conversation was completed."
+ description = "The conversation has ended. Please start a new conversation."
code = 400
class AppUnavailableError(BaseHTTPException):
error_code = 'app_unavailable'
- description = "App unavailable."
+ description = "App unavailable, please check your app configurations."
code = 400
@@ -45,5 +47,5 @@ class CompletionRequestError(BaseHTTPException):
class AppMoreLikeThisDisabledError(BaseHTTPException):
error_code = 'app_more_like_this_disabled'
- description = "More like this disabled."
+ description = "The 'More like this' feature is disabled. Please refresh your page."
code = 403
diff --git a/api/controllers/console/datasets/datasets_document.py b/api/controllers/console/datasets/datasets_document.py
index 79e52d565b..3b9efeaab4 100644
--- a/api/controllers/console/datasets/datasets_document.py
+++ b/api/controllers/console/datasets/datasets_document.py
@@ -10,13 +10,14 @@ from werkzeug.exceptions import NotFound, Forbidden
import services
from controllers.console import api
-from controllers.console.app.error import ProviderNotInitializeError
+from controllers.console.app.error import ProviderNotInitializeError, ProviderQuotaExceededError, \
+ ProviderModelCurrentlyNotSupportError
from controllers.console.datasets.error import DocumentAlreadyFinishedError, InvalidActionError, DocumentIndexingError, \
InvalidMetadataError, ArchivedDocumentImmutableError
from controllers.console.setup import setup_required
from controllers.console.wraps import account_initialization_required
from core.indexing_runner import IndexingRunner
-from core.llm.error import ProviderTokenNotInitError
+from core.llm.error import ProviderTokenNotInitError, QuotaExceededError, ModelCurrentlyNotSupportError
from extensions.ext_redis import redis_client
from libs.helper import TimestampField
from extensions.ext_database import db
@@ -222,6 +223,10 @@ class DatasetDocumentListApi(Resource):
document = DocumentService.save_document_with_dataset_id(dataset, args, current_user)
except ProviderTokenNotInitError:
raise ProviderNotInitializeError()
+ except QuotaExceededError:
+ raise ProviderQuotaExceededError()
+ except ModelCurrentlyNotSupportError:
+ raise ProviderModelCurrentlyNotSupportError()
return document
@@ -259,6 +264,10 @@ class DatasetInitApi(Resource):
)
except ProviderTokenNotInitError:
raise ProviderNotInitializeError()
+ except QuotaExceededError:
+ raise ProviderQuotaExceededError()
+ except ModelCurrentlyNotSupportError:
+ raise ProviderModelCurrentlyNotSupportError()
response = {
'dataset': dataset,
diff --git a/api/controllers/console/datasets/error.py b/api/controllers/console/datasets/error.py
index 014822d565..29142b80e6 100644
--- a/api/controllers/console/datasets/error.py
+++ b/api/controllers/console/datasets/error.py
@@ -3,7 +3,7 @@ from libs.exception import BaseHTTPException
class NoFileUploadedError(BaseHTTPException):
error_code = 'no_file_uploaded'
- description = "No file uploaded."
+ description = "Please upload your file."
code = 400
@@ -27,25 +27,25 @@ class UnsupportedFileTypeError(BaseHTTPException):
class HighQualityDatasetOnlyError(BaseHTTPException):
error_code = 'high_quality_dataset_only'
- description = "High quality dataset only."
+ description = "Current operation only supports 'high-quality' datasets."
code = 400
class DatasetNotInitializedError(BaseHTTPException):
error_code = 'dataset_not_initialized'
- description = "Dataset not initialized."
+ description = "The dataset is still being initialized or indexing. Please wait a moment."
code = 400
class ArchivedDocumentImmutableError(BaseHTTPException):
error_code = 'archived_document_immutable'
- description = "Cannot process an archived document."
+ description = "The archived document is not editable."
code = 403
class DatasetNameDuplicateError(BaseHTTPException):
error_code = 'dataset_name_duplicate'
- description = "Dataset name already exists."
+ description = "The dataset name already exists. Please modify your dataset name."
code = 409
@@ -57,17 +57,17 @@ class InvalidActionError(BaseHTTPException):
class DocumentAlreadyFinishedError(BaseHTTPException):
error_code = 'document_already_finished'
- description = "Document already finished."
+ description = "The document has been processed. Please refresh the page or go to the document details."
code = 400
class DocumentIndexingError(BaseHTTPException):
error_code = 'document_indexing'
- description = "Document indexing."
+ description = "The document is being processed and cannot be edited."
code = 400
class InvalidMetadataError(BaseHTTPException):
error_code = 'invalid_metadata'
- description = "Invalid metadata."
+ description = "The metadata content is incorrect. Please check and verify."
code = 400
diff --git a/api/controllers/console/datasets/hit_testing.py b/api/controllers/console/datasets/hit_testing.py
index 16bb571df3..771d49045f 100644
--- a/api/controllers/console/datasets/hit_testing.py
+++ b/api/controllers/console/datasets/hit_testing.py
@@ -6,9 +6,12 @@ from werkzeug.exceptions import InternalServerError, NotFound, Forbidden
import services
from controllers.console import api
+from controllers.console.app.error import ProviderNotInitializeError, ProviderQuotaExceededError, \
+ ProviderModelCurrentlyNotSupportError
from controllers.console.datasets.error import HighQualityDatasetOnlyError, DatasetNotInitializedError
from controllers.console.setup import setup_required
from controllers.console.wraps import account_initialization_required
+from core.llm.error import ProviderTokenNotInitError, QuotaExceededError, ModelCurrentlyNotSupportError
from libs.helper import TimestampField
from services.dataset_service import DatasetService
from services.hit_testing_service import HitTestingService
@@ -92,6 +95,12 @@ class HitTestingApi(Resource):
return {"query": response['query'], 'records': marshal(response['records'], hit_testing_record_fields)}
except services.errors.index.IndexNotInitializedError:
raise DatasetNotInitializedError()
+ except ProviderTokenNotInitError:
+ raise ProviderNotInitializeError()
+ except QuotaExceededError:
+ raise ProviderQuotaExceededError()
+ except ModelCurrentlyNotSupportError:
+ raise ProviderModelCurrentlyNotSupportError()
except Exception as e:
logging.exception("Hit testing failed.")
raise InternalServerError(str(e))
diff --git a/api/controllers/console/error.py b/api/controllers/console/error.py
index 3040423d71..e563364f27 100644
--- a/api/controllers/console/error.py
+++ b/api/controllers/console/error.py
@@ -3,13 +3,14 @@ from libs.exception import BaseHTTPException
class AlreadySetupError(BaseHTTPException):
error_code = 'already_setup'
- description = "Application already setup."
+ description = "Dify has been successfully installed. Please refresh the page or return to the dashboard homepage."
code = 403
class NotSetupError(BaseHTTPException):
error_code = 'not_setup'
- description = "Application not setup."
+ description = "Dify has not been initialized and installed yet. " \
+ "Please proceed with the initialization and installation process first."
code = 401
diff --git a/api/controllers/console/version.py b/api/controllers/console/version.py
index 0e6e75c361..285cdcba5c 100644
--- a/api/controllers/console/version.py
+++ b/api/controllers/console/version.py
@@ -19,6 +19,14 @@ class VersionApi(Resource):
args = parser.parse_args()
check_update_url = current_app.config['CHECK_UPDATE_URL']
+ if not check_update_url:
+ return {
+ 'version': '0.0.0',
+ 'release_date': '',
+ 'release_notes': '',
+ 'can_auto_update': False
+ }
+
try:
response = requests.get(check_update_url, {
'current_version': args.get('current_version')
diff --git a/api/controllers/console/workspace/error.py b/api/controllers/console/workspace/error.py
index c5e3a3fb6a..cb744232ec 100644
--- a/api/controllers/console/workspace/error.py
+++ b/api/controllers/console/workspace/error.py
@@ -21,11 +21,11 @@ class InvalidInvitationCodeError(BaseHTTPException):
class AccountAlreadyInitedError(BaseHTTPException):
error_code = 'account_already_inited'
- description = "Account already inited."
+ description = "The account has been initialized. Please refresh the page."
code = 400
class AccountNotInitializedError(BaseHTTPException):
error_code = 'account_not_initialized'
- description = "Account not initialized."
+ description = "The account has not been initialized yet. Please proceed with the initialization process first."
code = 400
diff --git a/api/controllers/console/workspace/providers.py b/api/controllers/console/workspace/providers.py
index dc9e9c45f1..f2baec29c1 100644
--- a/api/controllers/console/workspace/providers.py
+++ b/api/controllers/console/workspace/providers.py
@@ -90,8 +90,8 @@ class ProviderTokenApi(Resource):
configs=args['token']
)
token_is_valid = True
- except ValidateFailedError:
- token_is_valid = False
+ except ValidateFailedError as ex:
+ raise ValueError(str(ex))
base64_encrypted_token = ProviderService.get_encrypted_token(
tenant=current_user.current_tenant,
@@ -157,7 +157,7 @@ class ProviderTokenValidateApi(Resource):
args = parser.parse_args()
# todo: remove this when the provider is supported
- if provider in [ProviderName.ANTHROPIC.value, ProviderName.AZURE_OPENAI.value, ProviderName.COHERE.value,
+ if provider in [ProviderName.ANTHROPIC.value, ProviderName.COHERE.value,
ProviderName.HUGGINGFACEHUB.value]:
return {'result': 'success', 'warning': 'MOCK: This provider is not supported yet.'}
diff --git a/api/controllers/service_api/app/error.py b/api/controllers/service_api/app/error.py
index c59f570efd..b7f6e0f6fa 100644
--- a/api/controllers/service_api/app/error.py
+++ b/api/controllers/service_api/app/error.py
@@ -4,43 +4,45 @@ from libs.exception import BaseHTTPException
class AppUnavailableError(BaseHTTPException):
error_code = 'app_unavailable'
- description = "App unavailable."
+ description = "App unavailable, please check your app configurations."
code = 400
class NotCompletionAppError(BaseHTTPException):
error_code = 'not_completion_app'
- description = "Not Completion App"
+ description = "Please check if your Completion app mode matches the right API route."
code = 400
class NotChatAppError(BaseHTTPException):
error_code = 'not_chat_app'
- description = "Not Chat App"
+ description = "Please check if your Chat app mode matches the right API route."
code = 400
class ConversationCompletedError(BaseHTTPException):
error_code = 'conversation_completed'
- description = "Conversation Completed."
+ description = "The conversation has ended. Please start a new conversation."
code = 400
class ProviderNotInitializeError(BaseHTTPException):
error_code = 'provider_not_initialize'
- description = "Provider Token not initialize."
+ description = "No valid model provider credentials found. " \
+ "Please go to Settings -> Model Provider to complete your provider credentials."
code = 400
class ProviderQuotaExceededError(BaseHTTPException):
error_code = 'provider_quota_exceeded'
- description = "Provider quota exceeded."
+ description = "Your quota for Dify Hosted OpenAI has been exhausted. " \
+ "Please go to Settings -> Model Provider to complete your own provider credentials."
code = 400
class ProviderModelCurrentlyNotSupportError(BaseHTTPException):
error_code = 'model_currently_not_support'
- description = "GPT-4 currently not support."
+ description = "Dify Hosted OpenAI trial currently not support the GPT-4 model."
code = 400
diff --git a/api/controllers/service_api/dataset/error.py b/api/controllers/service_api/dataset/error.py
index d231e0b40a..2131fe0bac 100644
--- a/api/controllers/service_api/dataset/error.py
+++ b/api/controllers/service_api/dataset/error.py
@@ -16,5 +16,5 @@ class DocumentIndexingError(BaseHTTPException):
class DatasetNotInitedError(BaseHTTPException):
error_code = 'dataset_not_inited'
- description = "Dataset not inited."
+ description = "The dataset is still being initialized or indexing. Please wait a moment."
code = 403
diff --git a/api/controllers/web/error.py b/api/controllers/web/error.py
index ea72422a1b..fdfe36f6d1 100644
--- a/api/controllers/web/error.py
+++ b/api/controllers/web/error.py
@@ -4,43 +4,45 @@ from libs.exception import BaseHTTPException
class AppUnavailableError(BaseHTTPException):
error_code = 'app_unavailable'
- description = "App unavailable."
+ description = "App unavailable, please check your app configurations."
code = 400
class NotCompletionAppError(BaseHTTPException):
error_code = 'not_completion_app'
- description = "Not Completion App"
+ description = "Please check if your Completion app mode matches the right API route."
code = 400
class NotChatAppError(BaseHTTPException):
error_code = 'not_chat_app'
- description = "Not Chat App"
+ description = "Please check if your Chat app mode matches the right API route."
code = 400
class ConversationCompletedError(BaseHTTPException):
error_code = 'conversation_completed'
- description = "Conversation Completed."
+ description = "The conversation has ended. Please start a new conversation."
code = 400
class ProviderNotInitializeError(BaseHTTPException):
error_code = 'provider_not_initialize'
- description = "Provider Token not initialize."
+ description = "No valid model provider credentials found. " \
+ "Please go to Settings -> Model Provider to complete your provider credentials."
code = 400
class ProviderQuotaExceededError(BaseHTTPException):
error_code = 'provider_quota_exceeded'
- description = "Provider quota exceeded."
+ description = "Your quota for Dify Hosted OpenAI has been exhausted. " \
+ "Please go to Settings -> Model Provider to complete your own provider credentials."
code = 400
class ProviderModelCurrentlyNotSupportError(BaseHTTPException):
error_code = 'model_currently_not_support'
- description = "GPT-4 currently not support."
+ description = "Dify Hosted OpenAI trial currently not support the GPT-4 model."
code = 400
@@ -52,11 +54,11 @@ class CompletionRequestError(BaseHTTPException):
class AppMoreLikeThisDisabledError(BaseHTTPException):
error_code = 'app_more_like_this_disabled'
- description = "More like this disabled."
+ description = "The 'More like this' feature is disabled. Please refresh your page."
code = 403
class AppSuggestedQuestionsAfterAnswerDisabledError(BaseHTTPException):
error_code = 'app_suggested_questions_after_answer_disabled'
- description = "Function Suggested questions after answer disabled."
+ description = "The 'Suggested Questions After Answer' feature is disabled. Please refresh your page."
code = 403
diff --git a/api/core/completion.py b/api/core/completion.py
index d05d8417da..5e559ac7c7 100644
--- a/api/core/completion.py
+++ b/api/core/completion.py
@@ -1,4 +1,4 @@
-from typing import Optional, List, Union
+from typing import Optional, List, Union, Tuple
from langchain.callbacks import CallbackManager
from langchain.chat_models.base import BaseChatModel
@@ -39,7 +39,8 @@ class Completion:
memory = cls.get_memory_from_conversation(
tenant_id=app.tenant_id,
app_model_config=app_model_config,
- conversation=conversation
+ conversation=conversation,
+ return_messages=False
)
inputs = conversation.inputs
@@ -96,7 +97,7 @@ class Completion:
)
# get llm prompt
- prompt = cls.get_main_llm_prompt(
+ prompt, stop_words = cls.get_main_llm_prompt(
mode=mode,
llm=final_llm,
pre_prompt=app_model_config.pre_prompt,
@@ -114,14 +115,22 @@ class Completion:
mode=mode
)
- response = final_llm.generate([prompt])
+ response = final_llm.generate([prompt], stop_words)
return response
@classmethod
- def get_main_llm_prompt(cls, mode: str, llm: BaseLanguageModel, pre_prompt: str, query: str, inputs: dict, chain_output: Optional[str],
+ def get_main_llm_prompt(cls, mode: str, llm: BaseLanguageModel, pre_prompt: str, query: str, inputs: dict,
+ chain_output: Optional[str],
memory: Optional[ReadOnlyConversationTokenDBBufferSharedMemory]) -> \
- Union[str | List[BaseMessage]]:
+ Tuple[Union[str | List[BaseMessage]], Optional[List[str]]]:
+ # disable template string in query
+ query_params = OutLinePromptTemplate.from_template(template=query).input_variables
+ if query_params:
+ for query_param in query_params:
+ if query_param not in inputs:
+ inputs[query_param] = '{' + query_param + '}'
+
pre_prompt = PromptBuilder.process_template(pre_prompt) if pre_prompt else pre_prompt
if mode == 'completion':
prompt_template = OutLinePromptTemplate.from_template(
@@ -142,6 +151,11 @@ And answer according to the language of the user's question.
if chain_output:
inputs['context'] = chain_output
+ context_params = OutLinePromptTemplate.from_template(template=chain_output).input_variables
+ if context_params:
+ for context_param in context_params:
+ if context_param not in inputs:
+ inputs[context_param] = '{' + context_param + '}'
prompt_inputs = {k: inputs[k] for k in prompt_template.input_variables if k in inputs}
prompt_content = prompt_template.format(
@@ -151,9 +165,9 @@ And answer according to the language of the user's question.
if isinstance(llm, BaseChatModel):
# use chat llm as completion model
- return [HumanMessage(content=prompt_content)]
+ return [HumanMessage(content=prompt_content)], None
else:
- return prompt_content
+ return prompt_content, None
else:
messages: List[BaseMessage] = []
@@ -161,11 +175,19 @@ And answer according to the language of the user's question.
"query": query
}
- human_message_prompt = "{query}"
+ human_message_prompt = ""
+
+ if pre_prompt:
+ pre_prompt_inputs = {k: inputs[k] for k in
+ OutLinePromptTemplate.from_template(template=pre_prompt).input_variables
+ if k in inputs}
+
+ if pre_prompt_inputs:
+ human_inputs.update(pre_prompt_inputs)
if chain_output:
human_inputs['context'] = chain_output
- human_message_instruction = """Use the following CONTEXT as your learned knowledge.
+ human_message_prompt += """Use the following CONTEXT as your learned knowledge.
[CONTEXT]
{context}
[END CONTEXT]
@@ -176,15 +198,35 @@ When answer to user:
Avoid mentioning that you obtained the information from the context.
And answer according to the language of the user's question.
"""
- if pre_prompt:
- human_inputs.update(inputs)
- human_message_instruction += pre_prompt + "\n"
- human_message_prompt = human_message_instruction + "Q:{query}\nA:"
- else:
- if pre_prompt:
- human_inputs.update(inputs)
- human_message_prompt = pre_prompt + "\n" + human_message_prompt
+ if pre_prompt:
+ human_message_prompt += pre_prompt
+
+ query_prompt = "\nHuman: {query}\nAI: "
+
+ if memory:
+ # append chat histories
+ tmp_human_message = PromptBuilder.to_human_message(
+ prompt_content=human_message_prompt + query_prompt,
+ inputs=human_inputs
+ )
+
+ curr_message_tokens = memory.llm.get_messages_tokens([tmp_human_message])
+ rest_tokens = llm_constant.max_context_token_length[memory.llm.model_name] \
+ - memory.llm.max_tokens - curr_message_tokens
+ rest_tokens = max(rest_tokens, 0)
+ histories = cls.get_history_messages_from_memory(memory, rest_tokens)
+
+ # disable template string in query
+ histories_params = OutLinePromptTemplate.from_template(template=histories).input_variables
+ if histories_params:
+ for histories_param in histories_params:
+ if histories_param not in human_inputs:
+ human_inputs[histories_param] = '{' + histories_param + '}'
+
+ human_message_prompt += "\n\n" + histories
+
+ human_message_prompt += query_prompt
# construct main prompt
human_message = PromptBuilder.to_human_message(
@@ -192,23 +234,14 @@ And answer according to the language of the user's question.
inputs=human_inputs
)
- if memory:
- # append chat histories
- tmp_messages = messages.copy() + [human_message]
- curr_message_tokens = memory.llm.get_messages_tokens(tmp_messages)
- rest_tokens = llm_constant.max_context_token_length[
- memory.llm.model_name] - memory.llm.max_tokens - curr_message_tokens
- rest_tokens = max(rest_tokens, 0)
- history_messages = cls.get_history_messages_from_memory(memory, rest_tokens)
- messages += history_messages
-
messages.append(human_message)
- return messages
+ return messages, ['\nHuman:']
@classmethod
def get_llm_callback_manager(cls, llm: Union[StreamableOpenAI, StreamableChatOpenAI],
- streaming: bool, conversation_message_task: ConversationMessageTask) -> CallbackManager:
+ streaming: bool,
+ conversation_message_task: ConversationMessageTask) -> CallbackManager:
llm_callback_handler = LLMCallbackHandler(llm, conversation_message_task)
if streaming:
callback_handlers = [llm_callback_handler, DifyStreamingStdOutCallbackHandler()]
@@ -220,7 +253,7 @@ And answer according to the language of the user's question.
@classmethod
def get_history_messages_from_memory(cls, memory: ReadOnlyConversationTokenDBBufferSharedMemory,
max_token_limit: int) -> \
- List[BaseMessage]:
+ str:
"""Get memory messages."""
memory.max_token_limit = max_token_limit
memory_key = memory.memory_variables[0]
@@ -290,7 +323,7 @@ And answer according to the language of the user's question.
)
# get llm prompt
- original_prompt = cls.get_main_llm_prompt(
+ original_prompt, _ = cls.get_main_llm_prompt(
mode="completion",
llm=llm,
pre_prompt=pre_prompt,
diff --git a/api/core/conversation_message_task.py b/api/core/conversation_message_task.py
index 0c2e9983ce..b23d6664bd 100644
--- a/api/core/conversation_message_task.py
+++ b/api/core/conversation_message_task.py
@@ -56,6 +56,9 @@ class ConversationMessageTask:
)
def init(self):
+ provider_name = LLMBuilder.get_default_provider(self.app.tenant_id)
+ self.model_dict['provider'] = provider_name
+
override_model_configs = None
if self.is_override:
override_model_configs = {
diff --git a/api/core/embedding/openai_embedding.py b/api/core/embedding/openai_embedding.py
index 0f7cb252e2..d1179180f6 100644
--- a/api/core/embedding/openai_embedding.py
+++ b/api/core/embedding/openai_embedding.py
@@ -173,6 +173,13 @@ class OpenAIEmbedding(BaseEmbedding):
Can be overriden for batch queries.
"""
+ if self.openai_api_type and self.openai_api_type == 'azure':
+ embeddings = []
+ for text in texts:
+ embeddings.append(self._get_text_embedding(text))
+
+ return embeddings
+
if self.deployment_name is not None:
engine = self.deployment_name
else:
@@ -187,6 +194,13 @@ class OpenAIEmbedding(BaseEmbedding):
async def _aget_text_embeddings(self, texts: List[str]) -> List[List[float]]:
"""Asynchronously get text embeddings."""
+ if self.openai_api_type and self.openai_api_type == 'azure':
+ embeddings = []
+ for text in texts:
+ embeddings.append(await self._aget_text_embedding(text))
+
+ return embeddings
+
if self.deployment_name is not None:
engine = self.deployment_name
else:
diff --git a/api/core/index/index_builder.py b/api/core/index/index_builder.py
index 7f0486546e..05f08075d4 100644
--- a/api/core/index/index_builder.py
+++ b/api/core/index/index_builder.py
@@ -46,3 +46,15 @@ class IndexBuilder:
prompt_helper=prompt_helper,
embed_model=OpenAIEmbedding(**model_credentials),
)
+
+ @classmethod
+ def get_fake_llm_service_context(cls, tenant_id: str) -> ServiceContext:
+ llm = LLMBuilder.to_llm(
+ tenant_id=tenant_id,
+ model_name='fake'
+ )
+
+ return ServiceContext.from_defaults(
+ llm_predictor=LLMPredictor(llm=llm),
+ embed_model=OpenAIEmbedding()
+ )
diff --git a/api/core/index/vector_index.py b/api/core/index/vector_index.py
index f9d8542a8c..fa1c93cc06 100644
--- a/api/core/index/vector_index.py
+++ b/api/core/index/vector_index.py
@@ -83,7 +83,7 @@ class VectorIndex:
if not self._dataset.index_struct_dict:
return
- service_context = IndexBuilder.get_default_service_context(tenant_id=self._dataset.tenant_id)
+ service_context = IndexBuilder.get_fake_llm_service_context(tenant_id=self._dataset.tenant_id)
index = vector_store.get_index(
service_context=service_context,
@@ -101,7 +101,7 @@ class VectorIndex:
if not self._dataset.index_struct_dict:
return
- service_context = IndexBuilder.get_default_service_context(tenant_id=self._dataset.tenant_id)
+ service_context = IndexBuilder.get_fake_llm_service_context(tenant_id=self._dataset.tenant_id)
index = vector_store.get_index(
service_context=service_context,
diff --git a/api/core/llm/provider/azure_provider.py b/api/core/llm/provider/azure_provider.py
index d68ed3ccc4..649c64cf73 100644
--- a/api/core/llm/provider/azure_provider.py
+++ b/api/core/llm/provider/azure_provider.py
@@ -1,22 +1,24 @@
import json
+import logging
from typing import Optional, Union
import requests
from core.llm.provider.base import BaseProvider
+from core.llm.provider.errors import ValidateFailedError
from models.provider import ProviderName
class AzureProvider(BaseProvider):
- def get_models(self, model_id: Optional[str] = None) -> list[dict]:
- credentials = self.get_credentials(model_id)
+ def get_models(self, model_id: Optional[str] = None, credentials: Optional[dict] = None) -> list[dict]:
+ credentials = self.get_credentials(model_id) if not credentials else credentials
url = "{}/openai/deployments?api-version={}".format(
- credentials.get('openai_api_base'),
- credentials.get('openai_api_version')
+ str(credentials.get('openai_api_base')),
+ str(credentials.get('openai_api_version'))
)
headers = {
- "api-key": credentials.get('openai_api_key'),
+ "api-key": str(credentials.get('openai_api_key')),
"content-type": "application/json; charset=utf-8"
}
@@ -29,8 +31,10 @@ class AzureProvider(BaseProvider):
'name': '{} ({})'.format(deployment['id'], deployment['model'])
} for deployment in result['data'] if deployment['status'] == 'succeeded']
else:
- # TODO: optimize in future
- raise Exception('Failed to get deployments from Azure OpenAI. Status code: {}'.format(response.status_code))
+ if response.status_code == 401:
+ raise AzureAuthenticationError()
+ else:
+ raise AzureRequestFailedError('Failed to request Azure OpenAI. Status code: {}'.format(response.status_code))
def get_credentials(self, model_id: Optional[str] = None) -> dict:
"""
@@ -38,7 +42,7 @@ class AzureProvider(BaseProvider):
"""
config = self.get_provider_api_key(model_id=model_id)
config['openai_api_type'] = 'azure'
- config['deployment_name'] = model_id.replace('.', '')
+ config['deployment_name'] = model_id.replace('.', '') if model_id else None
return config
def get_provider_name(self):
@@ -54,7 +58,7 @@ class AzureProvider(BaseProvider):
config = {
'openai_api_type': 'azure',
'openai_api_version': '2023-03-15-preview',
- 'openai_api_base': 'https://