Update app.py

app.py

@@ -1,71 +1,82 @@
-import chromadb
-from chromadb.api.types import Documents, EmbeddingFunction, Embeddings
+
 import json
-from ast import literal_eval
-from chromadb.config import Settings
-from paddlenlp import Taskflow
 import requests
-from io import BytesIO
-from PIL import Image
 import gradio as gr
+import typesense
+from urllib.parse import quote
+import os
 
-vision_language=Taskflow("feature_extraction", model='PaddlePaddle/ernie_vil-2.0-base-zh')
-
-def getImageTestFeture(content):
-    if content.startswith("http"):
-        response = requests.get(content)
-        x=BytesIO(response.content)
-        f_embeds = vision_language(Image.open(x))
-    else:
-        f_embeds = vision_language(content)
-    text_features = f_embeds["features"][0]
-    return text_features
 
-class MyEmbeddingFunction(EmbeddingFunction):
-    def __call__(self, texts: Documents) -> Embeddings:
-        qr=[]
-        for doc in texts:
-            text_embeds = getImageTestFeture(doc)
-            #print(len(text_features))
-            bedx=text_embeds.tolist()
-            qr.append(bedx)
-        return qr
 
-client = chromadb.Client(Settings(
-    chroma_db_impl="duckdb+parquet",
-    persist_directory="x/" # Optional, defaults to .chromadb/ in the current directory
-))
+def getWordVec(content):
+    content=quote(content,'utf-8')
+    #print(content)
+    xurl=os.getenv("emburl")
+    print(xurl)
+    url=xurl +content
+    #print(url)
+    response = requests.get(url)
+    jsonar=json.loads(response.text).get("embed")
+    #print(len(jsonar))
+    return jsonar
 
-collection = client.get_or_create_collection(name="pics", metadata={"hnsw:space": "cosine"}, embedding_function=MyEmbeddingFunction())
+typesenseserver=os.getenv("typesenseserver")
+typesenseport=os.getenv("typesenseport")
+typesensekey=os.getenv("typesensekey")
+typesensecolname=os.getenv("typesensecolname")
 
+confignode={}
+confignode['host']=typesenseserver
+confignode['port']=typesenseport
+confignode['protocol']='http'
+nodes=[]
+nodes.append(confignode)
+nodeconfig={}
+nodeconfig["nodes"]=nodes
+nodeconfig["api_key"]=typesensekey
+print(nodeconfig)
+client = typesense.Client(nodeconfig)
 def queryimgage(text):
     html="<table border='1'>\
     <tr>\
     <th>img</th>\
     <th>score</th>\
   </tr>"
-    atext=[]
-    atext.append(text)
-    results = collection.query(
-        query_texts=atext,
-        n_results=20,
-    )
-    ids=results['ids'][0]
-    documents=results['documents'][0]
-    distances=results['distances'][0]
-    xcount=len(ids)
-    for i in range(xcount):
-        #print("id:%s,url:%s,score:%s"%(ids[i],documents[i],distances[i]))
+    info=getWordVec(text)
+    search_requests = {
+  'searches': [
+    {
+      'collection': typesensecolname,
+      'q' : '*',
+      'per_page': 20,
+      'exclude_fields' : 'my_vector',
+      'vector_query': 'my_vector:(['+",".join(str(x) for x in info )+'], k:1000)'
+    }
+  ]
+}
+    common_search_params =  {}
+    res=client.multi_search.perform(search_requests, common_search_params)
+    result=res['results'][0]['hits']
+    for resultinfo in result:
+         documents=resultinfo['document']
+         score=(2-resultinfo['vector_distance'])/2
          html=html +"<tr>\
-        <td><img src='"+documents[xcount-1-i]+"' width=640></td>\
-        <td>"+ str(distances[xcount-1-i])+"</td>"
+        <td><img src='"+documents['imageurl']+"' width=640 height=600></td>\
+        <td>"+ str(score)+"</td>"
     html=html+"</table>"
     return  html
 
+def getNumtip():
+    num=client.collections[typesensecolname].retrieve()["num_documents"]
+    numtip="图片数:" + str(num)
+    return numtip
+
+
 demo = gr.Interface(
-    queryimgage,
-    gr.Textbox(placeholder="请输入文本"),
-    [ "html"]
+    fn=queryimgage,
+    inputs=gr.Textbox(placeholder="请输入文本"),
+    outputs=[ "html"],
+    article=getNumtip()
 )
 
 demo.launch()