feat: bump version to 0.3.31-fix3 (#1606)

Co-authored-by: jyong <jyong@dify.ai>

api/controllers/console/app/statistic.py

@@ -62,16 +62,15 @@ class DailyConversationStatistic(Resource):
 
         sql_query += ' GROUP BY date order by date'
 
-        with db.engine.begin() as conn:
-            rs = conn.execute(db.text(sql_query), arg_dict)
-
         response_data = []
 
-        for i in rs:
-            response_data.append({
-                'date': str(i.date),
-                'conversation_count': i.conversation_count
-            })
+        with db.engine.begin() as conn:
+            rs = conn.execute(db.text(sql_query), arg_dict)
+            for i in rs:
+                response_data.append({
+                    'date': str(i.date),
+                    'conversation_count': i.conversation_count
+                })
 
         return jsonify({
             'data': response_data
@@ -124,16 +123,15 @@ class DailyTerminalsStatistic(Resource):
 
         sql_query += ' GROUP BY date order by date'
 
-        with db.engine.begin() as conn:
-            rs = conn.execute(db.text(sql_query), arg_dict)
-
         response_data = []
 
-        for i in rs:
-            response_data.append({
-                'date': str(i.date),
-                'terminal_count': i.terminal_count
-            })
+        with db.engine.begin() as conn:
+            rs = conn.execute(db.text(sql_query), arg_dict)            
+            for i in rs:
+                response_data.append({
+                    'date': str(i.date),
+                    'terminal_count': i.terminal_count
+                })
 
         return jsonify({
             'data': response_data
@@ -187,18 +185,17 @@ class DailyTokenCostStatistic(Resource):
 
         sql_query += ' GROUP BY date order by date'
 
-        with db.engine.begin() as conn:
-            rs = conn.execute(db.text(sql_query), arg_dict)
-
         response_data = []
 
-        for i in rs:
-            response_data.append({
-                'date': str(i.date),
-                'token_count': i.token_count,
-                'total_price': i.total_price,
-                'currency': 'USD'
-            })
+        with db.engine.begin() as conn:
+            rs = conn.execute(db.text(sql_query), arg_dict)
+            for i in rs:
+                response_data.append({
+                    'date': str(i.date),
+                    'token_count': i.token_count,
+                    'total_price': i.total_price,
+                    'currency': 'USD'
+                })
 
         return jsonify({
             'data': response_data
@@ -256,16 +253,15 @@ LEFT JOIN conversations c on c.id=subquery.conversation_id
 GROUP BY date
 ORDER BY date"""
 
+        response_data = []
+        
         with db.engine.begin() as conn:
             rs = conn.execute(db.text(sql_query), arg_dict)
-
-        response_data = []
-
-        for i in rs:
-            response_data.append({
-                'date': str(i.date),
-                'interactions': float(i.interactions.quantize(Decimal('0.01')))
-            })
+            for i in rs:
+                response_data.append({
+                    'date': str(i.date),
+                    'interactions': float(i.interactions.quantize(Decimal('0.01')))
+                })
 
         return jsonify({
             'data': response_data
@@ -320,20 +316,19 @@ class UserSatisfactionRateStatistic(Resource):
 
         sql_query += ' GROUP BY date order by date'
 
-        with db.engine.begin() as conn:
-            rs = conn.execute(db.text(sql_query), arg_dict)
-
         response_data = []
 
-        for i in rs:
-            response_data.append({
-                'date': str(i.date),
-                'rate': round((i.feedback_count * 1000 / i.message_count) if i.message_count > 0 else 0, 2),
-            })
+        with db.engine.begin() as conn:
+            rs = conn.execute(db.text(sql_query), arg_dict)
+            for i in rs:
+                response_data.append({
+                    'date': str(i.date),
+                    'rate': round((i.feedback_count * 1000 / i.message_count) if i.message_count > 0 else 0, 2),
+                })
 
         return jsonify({
-                'data': response_data
-            })
+            'data': response_data
+        })
 
 
 class AverageResponseTimeStatistic(Resource):
@@ -383,16 +378,15 @@ class AverageResponseTimeStatistic(Resource):
 
         sql_query += ' GROUP BY date order by date'
 
-        with db.engine.begin() as conn:
-            rs = conn.execute(db.text(sql_query), arg_dict)
-
         response_data = []
 
-        for i in rs:
-            response_data.append({
-                'date': str(i.date),
-                'latency': round(i.latency * 1000, 4)
-            })
+        with db.engine.begin() as conn:
+            rs = conn.execute(db.text(sql_query), arg_dict)            
+            for i in rs:
+                response_data.append({
+                    'date': str(i.date),
+                    'latency': round(i.latency * 1000, 4)
+                })
 
         return jsonify({
             'data': response_data
@@ -447,16 +441,15 @@ WHERE app_id = :app_id'''
 
         sql_query += ' GROUP BY date order by date'
 
-        with db.engine.begin() as conn:
-            rs = conn.execute(db.text(sql_query), arg_dict)
-
         response_data = []
 
-        for i in rs:
-            response_data.append({
-                'date': str(i.date),
-                'tps': round(i.tokens_per_second, 4)
-            })
+        with db.engine.begin() as conn:
+            rs = conn.execute(db.text(sql_query), arg_dict)
+            for i in rs:
+                response_data.append({
+                    'date': str(i.date),
+                    'tps': round(i.tokens_per_second, 4)
+                })
 
         return jsonify({
             'data': response_data

api/core/index/vector_index/weaviate_vector_index.py

@@ -111,7 +111,7 @@ class WeaviateVectorIndex(BaseVectorIndex):
         if self._vector_store:
             return self._vector_store
 
-        attributes = ['doc_id', 'dataset_id', 'document_id']
+        attributes = ['doc_id', 'dataset_id', 'document_id', 'doc_hash']
         if self._is_origin():
             attributes = ['doc_id']
 

api/core/model_providers/providers/anthropic_provider.py

@@ -32,9 +32,12 @@ class AnthropicProvider(BaseModelProvider):
         if model_type == ModelType.TEXT_GENERATION:
             return [
                 {
-                    'id': 'claude-instant-1',
-                    'name': 'claude-instant-1',
+                    'id': 'claude-2.1',
+                    'name': 'claude-2.1',
                     'mode': ModelMode.CHAT.value,
+                    'features': [
+                        ModelFeature.AGENT_THOUGHT.value
+                    ]
                 },
                 {
                     'id': 'claude-2',
@@ -44,6 +47,11 @@ class AnthropicProvider(BaseModelProvider):
                         ModelFeature.AGENT_THOUGHT.value
                     ]
                 },
+                {
+                    'id': 'claude-instant-1',
+                    'name': 'claude-instant-1',
+                    'mode': ModelMode.CHAT.value,
+                },
             ]
         else:
             return []
@@ -73,12 +81,18 @@ class AnthropicProvider(BaseModelProvider):
         :param model_type:
         :return:
         """
+        model_max_tokens = {
+            'claude-instant-1': 100000,
+            'claude-2': 100000,
+            'claude-2.1': 200000,
+        }
+
         return ModelKwargsRules(
             temperature=KwargRule[float](min=0, max=1, default=1, precision=2),
             top_p=KwargRule[float](min=0, max=1, default=0.7, precision=2),
             presence_penalty=KwargRule[float](enabled=False),
             frequency_penalty=KwargRule[float](enabled=False),
-            max_tokens=KwargRule[int](alias="max_tokens_to_sample", min=10, max=100000, default=256, precision=0),
+            max_tokens=KwargRule[int](alias="max_tokens_to_sample", min=10, max=model_max_tokens.get(model_name, 100000), default=256, precision=0),
         )
 
     @classmethod

api/core/model_providers/rules/anthropic.json

@@ -23,8 +23,14 @@
             "currency": "USD"
         },
         "claude-2": {
-            "prompt": "11.02",
-            "completion": "32.68",
+            "prompt": "8.00",
+            "completion": "24.00",
+            "unit": "0.000001",
+            "currency": "USD"
+        },
+        "claude-2.1": {
+            "prompt": "8.00",
+            "completion": "24.00",
             "unit": "0.000001",
             "currency": "USD"
         }

api/core/orchestrator_rule_parser.py

@@ -207,10 +207,10 @@ class OrchestratorRuleParser:
             ).first()
 
             if not dataset:
-                return None
+                continue
 
             if dataset and dataset.available_document_count == 0 and dataset.available_document_count == 0:
-                return None
+                continue
             dataset_ids.append(dataset.id)
             if retrieval_model == 'single':
                 retrieval_model = dataset.retrieval_model if dataset.retrieval_model else default_retrieval_model

api/core/third_party/langchain/llms/anthropic_llm.py

@@ -1,7 +1,7 @@
 from typing import Dict
 
-from httpx import Limits
 from langchain.chat_models import ChatAnthropic
+from langchain.schema import ChatMessage, BaseMessage, HumanMessage, AIMessage, SystemMessage
 from langchain.utils import get_from_dict_or_env, check_package_version
 from pydantic import root_validator
 
@@ -29,8 +29,7 @@ class AnthropicLLM(ChatAnthropic):
                 base_url=values["anthropic_api_url"],
                 api_key=values["anthropic_api_key"],
                 timeout=values["default_request_timeout"],
-                max_retries=0,
-                connection_pool_limits=Limits(max_connections=200, max_keepalive_connections=100),
+                max_retries=0
             )
             values["async_client"] = anthropic.AsyncAnthropic(
                 base_url=values["anthropic_api_url"],
@@ -46,3 +45,16 @@ class AnthropicLLM(ChatAnthropic):
                 "Please it install it with `pip install anthropic`."
             )
         return values
+
+    def _convert_one_message_to_text(self, message: BaseMessage) -> str:
+        if isinstance(message, ChatMessage):
+            message_text = f"\n\n{message.role.capitalize()}: {message.content}"
+        elif isinstance(message, HumanMessage):
+            message_text = f"{self.HUMAN_PROMPT} {message.content}"
+        elif isinstance(message, AIMessage):
+            message_text = f"{self.AI_PROMPT} {message.content}"
+        elif isinstance(message, SystemMessage):
+            message_text = f"{message.content}"
+        else:
+            raise ValueError(f"Got unknown type {message}")
+        return message_text

api/core/tool/dataset_multi_retriever_tool.py

@@ -192,7 +192,7 @@ class DatasetMultiRetrieverTool(BaseTool):
                         'search_method'] == 'hybrid_search':
                         embedding_thread = threading.Thread(target=RetrievalService.embedding_search, kwargs={
                             'flask_app': current_app._get_current_object(),
-                            'dataset': dataset,
+                            'dataset_id': str(dataset.id),
                             'query': query,
                             'top_k': self.top_k,
                             'score_threshold': self.score_threshold,
@@ -210,7 +210,7 @@ class DatasetMultiRetrieverTool(BaseTool):
                         full_text_index_thread = threading.Thread(target=RetrievalService.full_text_index_search,
                                                                   kwargs={
                                                                       'flask_app': current_app._get_current_object(),
-                                                                      'dataset': dataset,
+                                                                      'dataset_id': str(dataset.id),
                                                                       'query': query,
                                                                       'search_method': 'hybrid_search',
                                                                       'embeddings': embeddings,

api/core/tool/dataset_retriever_tool.py

@@ -106,7 +106,7 @@ class DatasetRetrieverTool(BaseTool):
                 if retrieval_model['search_method'] == 'semantic_search' or retrieval_model['search_method'] == 'hybrid_search':
                     embedding_thread = threading.Thread(target=RetrievalService.embedding_search, kwargs={
                         'flask_app': current_app._get_current_object(),
-                        'dataset': dataset,
+                        'dataset_id': str(dataset.id),
                         'query': query,
                         'top_k': self.top_k,
                         'score_threshold': retrieval_model['score_threshold'] if retrieval_model[
@@ -124,7 +124,7 @@ class DatasetRetrieverTool(BaseTool):
                 if retrieval_model['search_method'] == 'full_text_search' or retrieval_model['search_method'] == 'hybrid_search':
                     full_text_index_thread = threading.Thread(target=RetrievalService.full_text_index_search, kwargs={
                         'flask_app': current_app._get_current_object(),
-                        'dataset': dataset,
+                        'dataset_id': str(dataset.id),
                         'query': query,
                         'search_method': retrieval_model['search_method'],
                         'embeddings': embeddings,

api/core/vector_store/vector/weaviate.py

@@ -60,7 +60,7 @@ def _create_weaviate_client(**kwargs: Any) -> Any:
 
 
 def _default_score_normalizer(val: float) -> float:
-    return 1 - 1 / (1 + np.exp(val))
+    return 1 - val
 
 
 def _json_serializable(value: Any) -> Any:
@@ -243,7 +243,8 @@ class Weaviate(VectorStore):
             query_obj = query_obj.with_where(kwargs.get("where_filter"))
         if kwargs.get("additional"):
             query_obj = query_obj.with_additional(kwargs.get("additional"))
-        result = query_obj.with_bm25(query=content).with_limit(k).do()
+        properties = ['text', 'dataset_id', 'doc_hash', 'doc_id', 'document_id']
+        result = query_obj.with_bm25(query=query, properties=properties).with_limit(k).do()
         if "errors" in result:
             raise ValueError(f"Error during query: {result['errors']}")
         docs = []
@@ -380,14 +381,14 @@ class Weaviate(VectorStore):
             result = (
                 query_obj.with_near_vector(vector)
                 .with_limit(k)
-                .with_additional("vector")
+                .with_additional(["vector", "distance"])
                 .do()
             )
         else:
             result = (
                 query_obj.with_near_text(content)
                 .with_limit(k)
-                .with_additional("vector")
+                .with_additional(["vector", "distance"])
                 .do()
             )
 
@@ -397,7 +398,7 @@ class Weaviate(VectorStore):
         docs_and_scores = []
         for res in result["data"]["Get"][self._index_name]:
             text = res.pop(self._text_key)
-            score = np.dot(res["_additional"]["vector"], embedded_query)
+            score = res["_additional"]["distance"]
             docs_and_scores.append((Document(page_content=text, metadata=res), score))
         return docs_and_scores
 

api/core/vector_store/weaviate_vector_store.py

@@ -1,4 +1,4 @@
-from langchain.vectorstores import Weaviate
+from core.vector_store.vector.weaviate import Weaviate
 
 
 class WeaviateVectorStore(Weaviate):

api/requirements.txt

@@ -35,7 +35,7 @@ docx2txt==0.8
 pypdfium2==4.16.0
 resend~=0.5.1
 pyjwt~=2.6.0
-anthropic~=0.3.4
+anthropic~=0.7.2
 newspaper3k==0.2.8
 google-api-python-client==2.90.0
 wikipedia==1.4.0

api/services/completion_service.py

@@ -232,7 +232,7 @@ class CompletionService:
                 logging.exception("Unknown Error in completion")
                 PubHandler.pub_error(user, generate_task_id, e)
             finally:
-                db.session.commit()
+                db.session.remove()
 
     @classmethod
     def countdown_and_close(cls, flask_app: Flask, worker_thread, pubsub, detached_user,
@@ -242,22 +242,25 @@ class CompletionService:
 
         def close_pubsub():
             with flask_app.app_context():
-                user = db.session.merge(detached_user)
+                try:
+                    user = db.session.merge(detached_user)
 
-                sleep_iterations = 0
-                while sleep_iterations < timeout and worker_thread.is_alive():
-                    if sleep_iterations > 0 and sleep_iterations % 10 == 0:
-                        PubHandler.ping(user, generate_task_id)
+                    sleep_iterations = 0
+                    while sleep_iterations < timeout and worker_thread.is_alive():
+                        if sleep_iterations > 0 and sleep_iterations % 10 == 0:
+                            PubHandler.ping(user, generate_task_id)
 
-                    time.sleep(1)
-                    sleep_iterations += 1
+                        time.sleep(1)
+                        sleep_iterations += 1
 
-                if worker_thread.is_alive():
-                    PubHandler.stop(user, generate_task_id)
-                    try:
-                        pubsub.close()
-                    except Exception:
-                        pass
+                    if worker_thread.is_alive():
+                        PubHandler.stop(user, generate_task_id)
+                        try:
+                            pubsub.close()
+                        except Exception:
+                            pass
+                finally:
+                    db.session.remove()
 
         countdown_thread = threading.Thread(target=close_pubsub)
         countdown_thread.start()
@@ -394,7 +397,7 @@ class CompletionService:
                     logging.exception(e)
                     raise
             finally:
-                db.session.commit()
+                db.session.remove()
 
                 try:
                     pubsub.unsubscribe(generate_channel)
@@ -436,7 +439,7 @@ class CompletionService:
                         logging.exception(e)
                         raise
                 finally:
-                    db.session.commit()
+                    db.session.remove()
 
                     try:
                         pubsub.unsubscribe(generate_channel)

api/services/hit_testing_service.py

@@ -61,7 +61,7 @@ class HitTestingService:
         if retrieval_model['search_method'] == 'semantic_search' or retrieval_model['search_method'] == 'hybrid_search':
             embedding_thread = threading.Thread(target=RetrievalService.embedding_search, kwargs={
                 'flask_app': current_app._get_current_object(),
-                'dataset': dataset,
+                'dataset_id': str(dataset.id),
                 'query': query,
                 'top_k': retrieval_model['top_k'],
                 'score_threshold': retrieval_model['score_threshold'] if retrieval_model['score_threshold_enable'] else None,
@@ -77,7 +77,7 @@ class HitTestingService:
         if retrieval_model['search_method'] == 'full_text_search' or retrieval_model['search_method'] == 'hybrid_search':
             full_text_index_thread = threading.Thread(target=RetrievalService.full_text_index_search, kwargs={
                 'flask_app': current_app._get_current_object(),
-                'dataset': dataset,
+                'dataset_id': str(dataset.id),
                 'query': query,
                 'search_method': retrieval_model['search_method'],
                 'embeddings': embeddings,

api/services/retrieval_service.py

@@ -4,6 +4,7 @@ from flask import current_app, Flask
 from langchain.embeddings.base import Embeddings
 from core.index.vector_index.vector_index import VectorIndex
 from core.model_providers.model_factory import ModelFactory
+from extensions.ext_database import db
 from models.dataset import Dataset
 
 default_retrieval_model = {
@@ -21,10 +22,13 @@ default_retrieval_model = {
 class RetrievalService:
 
     @classmethod
-    def embedding_search(cls, flask_app: Flask, dataset: Dataset, query: str,
+    def embedding_search(cls, flask_app: Flask, dataset_id: str, query: str,
                          top_k: int, score_threshold: Optional[float], reranking_model: Optional[dict],
                          all_documents: list, search_method: str, embeddings: Embeddings):
         with flask_app.app_context():
+            dataset = db.session.query(Dataset).filter(
+                Dataset.id == dataset_id
+            ).first()
 
             vector_index = VectorIndex(
                 dataset=dataset,
@@ -56,10 +60,13 @@ class RetrievalService:
                     all_documents.extend(documents)
 
     @classmethod
-    def full_text_index_search(cls, flask_app: Flask, dataset: Dataset, query: str,
+    def full_text_index_search(cls, flask_app: Flask, dataset_id: str, query: str,
                                top_k: int, score_threshold: Optional[float], reranking_model: Optional[dict],
                                all_documents: list, search_method: str, embeddings: Embeddings):
         with flask_app.app_context():
+            dataset = db.session.query(Dataset).filter(
+                Dataset.id == dataset_id
+            ).first()
 
             vector_index = VectorIndex(
                 dataset=dataset,

api/tests/unit_tests/model_providers/test_anthropic_provider.py

@@ -31,12 +31,12 @@ def mock_chat_generate_invalid(messages: List[BaseMessage],
                                run_manager: Optional[CallbackManagerForLLMRun] = None,
                                **kwargs: Any):
     raise anthropic.APIStatusError('Invalid credentials',
-                                   request=httpx._models.Request(
-                                       method='POST',
-                                       url='https://api.anthropic.com/v1/completions',
-                                   ),
                                    response=httpx._models.Response(
                                        status_code=401,
+                                       request=httpx._models.Request(
+                                           method='POST',
+                                           url='https://api.anthropic.com/v1/completions',
+                                       )
                                    ),
                                    body=None
                                 )

docker/docker-compose.middleware.yaml

@@ -30,7 +30,7 @@ services:
 
   # The Weaviate vector store.
   weaviate:
-    image: semitechnologies/weaviate:1.18.4
+    image: semitechnologies/weaviate:1.19.0
     restart: always
     volumes:
       # Mount the Weaviate data directory to the container.
@@ -63,4 +63,4 @@ services:
 #    environment:
 #      QDRANT__API_KEY: 'difyai123456'
 #    ports:
-#      - "6333:6333"
+#      - "6333:6333"

docker/docker-compose.yaml

@@ -2,7 +2,7 @@ version: '3.1'
 services:
   # API service
   api:
-    image: langgenius/dify-api:0.3.31-fix1
+    image: langgenius/dify-api:0.3.31-fix3
     restart: always
     environment:
       # Startup mode, 'api' starts the API server.
@@ -128,7 +128,7 @@ services:
   # worker service
   # The Celery worker for processing the queue.
   worker:
-    image: langgenius/dify-api:0.3.31-fix1
+    image: langgenius/dify-api:0.3.31-fix3
     restart: always
     environment:
       # Startup mode, 'worker' starts the Celery worker for processing the queue.
@@ -196,7 +196,7 @@ services:
 
   # Frontend web application.
   web:
-    image: langgenius/dify-web:0.3.31-fix1
+    image: langgenius/dify-web:0.3.31-fix3
     restart: always
     environment:
       EDITION: SELF_HOSTED
@@ -253,7 +253,7 @@ services:
 
   # The Weaviate vector store.
   weaviate:
-    image: semitechnologies/weaviate:1.18.4
+    image: semitechnologies/weaviate:1.19.0
     restart: always
     volumes:
       # Mount the Weaviate data directory to the container.

web/app/components/app/chat/answer/index.tsx

@@ -194,7 +194,7 @@ const Answer: FC<IAnswerProps> = ({
             </div>
           )
         }
-        <div className={cn(s.answerWrapWrap, 'chat-answer-container')}>
+        <div className={cn(s.answerWrapWrap, 'chat-answer-container group')}>
           <div className={`${s.answerWrap} ${showEdit ? 'w-full' : ''}`}>
             <div className={`${s.answer} relative text-sm text-gray-900`}>
               <div className={'ml-2 py-3 px-4 bg-gray-100 rounded-tr-2xl rounded-b-2xl'}>
@@ -280,7 +280,7 @@ const Answer: FC<IAnswerProps> = ({
                 {!feedbackDisabled && renderFeedbackRating(feedback?.rating, !isHideFeedbackEdit, displayScene !== 'console')}
               </div>
             </div>
-            {more && <MoreInfo more={more} isQuestion={false} />}
+            {more && <MoreInfo className='hidden group-hover:block' more={more} isQuestion={false} />}
           </div>
         </div>
       </div>

web/app/components/app/chat/more-info/index.tsx

@@ -5,11 +5,15 @@ import { useTranslation } from 'react-i18next'
 import type { MessageMore } from '../type'
 import { formatNumber } from '@/utils/format'
 
-export type IMoreInfoProps = { more: MessageMore; isQuestion: boolean }
+export type IMoreInfoProps = {
+  more: MessageMore
+  isQuestion: boolean
+  className?: string
+}
 
-const MoreInfo: FC<IMoreInfoProps> = ({ more, isQuestion }) => {
+const MoreInfo: FC<IMoreInfoProps> = ({ more, isQuestion, className }) => {
   const { t } = useTranslation()
-  return (<div className={`mt-1 w-full text-xs text-gray-400 !text-right ${isQuestion ? 'mr-2 text-right ' : 'ml-2 text-left float-right'}`}>
+  return (<div className={`mt-1 w-full text-xs text-gray-400 ${isQuestion ? 'mr-2 text-right ' : 'pl-2 text-left float-right'} ${className}`}>
     <span>{`${t('appLog.detail.timeConsuming')} ${more.latency}${t('appLog.detail.second')}`}</span>
     <span>{`${t('appLog.detail.tokenCost')} ${formatNumber(more.tokens)}`}</span>
     <span>· </span>

web/global.d.ts

@@ -1,5 +1,3 @@
 declare module 'lamejs';
 declare module 'react-18-input-autosize';
-declare module 'fetch-readablestream' {
-  export default function fetchReadableStream(url: string, options?: RequestInit): Promise<Response>
-}
+

web/package.json

@@ -36,7 +36,6 @@
     "echarts": "^5.4.1",
     "echarts-for-react": "^3.0.2",
     "emoji-mart": "^5.5.2",
-    "fetch-readablestream": "^0.2.0",
     "i18next": "^22.4.13",
     "i18next-resources-to-backend": "^1.1.3",
     "immer": "^9.0.19",

web/service/base.ts

@@ -1,11 +1,8 @@
-import fetchStream from 'fetch-readablestream'
 import { API_PREFIX, IS_CE_EDITION, PUBLIC_API_PREFIX } from '@/config'
 import Toast from '@/app/components/base/toast'
 import type { MessageEnd, MessageReplace, ThoughtItem } from '@/app/components/app/chat/type'
-import { isSupportNativeFetchStream } from '@/utils/stream'
 
 const TIME_OUT = 100000
-const supportNativeFetchStream = isSupportNativeFetchStream()
 
 const ContentType = {
   json: 'application/json',
@@ -223,9 +220,6 @@ const baseFetch = <T>(
   if (body && bodyStringify)
     options.body = JSON.stringify(body)
 
-  // for those do not support native fetch stream, we use fetch-readablestream as polyfill
-  const doFetch = supportNativeFetchStream ? globalThis.fetch : fetchStream
-
   // Handle timeout
   return Promise.race([
     new Promise((resolve, reject) => {
@@ -234,7 +228,7 @@ const baseFetch = <T>(
       }, TIME_OUT)
     }),
     new Promise((resolve, reject) => {
-      doFetch(urlWithPrefix, options as RequestInit)
+      globalThis.fetch(urlWithPrefix, options as RequestInit)
         .then((res) => {
           const resClone = res.clone()
           // Error handler

web/utils/stream.ts

@@ -1,21 +0,0 @@
-// https://developer.chrome.com/articles/fetch-streaming-requests/#feature-detection
-export const isSupportNativeFetchStream = () => {
-  const supportsRequestStreams = (() => {
-    let duplexAccessed = false
-
-    const params = {
-      body: new ReadableStream(),
-      method: 'POST',
-      get duplex() {
-        duplexAccessed = true
-        return 'half'
-      },
-    }
-
-    const hasContentType = new Request('', params).headers.has('Content-Type')
-
-    return duplexAccessed && !hasContentType
-  })()
-
-  return supportsRequestStreams
-}