from fasthtml.common import *
from llama_index.core import SimpleDirectoryReader, Document,VectorStoreIndex
from llama_index.core.node_parser import SimpleNodeParser
from llama_index.core.text_splitter import TokenTextSplitter
from llama_index.vector_stores.chroma import ChromaVectorStore
from llama_index.core.storage.storage_context import StorageContext
from llama_index.embeddings.huggingface import HuggingFaceEmbedding
from llama_index.core import SummaryIndex
from llama_index.llms.groq import Groq
from chromadb import PersistentClient
from llama_index.core import Settings
from llama_index.embeddings.huggingface_api import (
HuggingFaceInferenceAPIEmbedding,
)
import chromadb
import os
import threading
import time
from llama_index.core.memory import ChatMemoryBuffer
os.environ["GROQ_API_KEY"] = "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"
llm_70b = Groq(model="llama-3.1-70b-versatile")
memory = ChatMemoryBuffer.from_defaults(token_limit=3900)
Settings.llm = llm_70b
app= FastHTML()
def listUsers():
with os.scandir("static") as files:
subdir = [file.name for file in files if file.is_dir()]
return subdir
def menuusers(users):
T=[]
n=0
for user in users:
T.append(Option(user, value=str(user)) )
return Form(
Select(*T,name="user"),
Button("Submit",type="submit",id="buttonMenuuser"),
hx_post="/checkInfoSources",hx_swap="innerHTML",hx_target="#files" ,id="menuuser")
@app.post("/checkInfoSources")
def checkInfoSources(user:str):
global userdata
with os.scandir("static/"+user) as files:
subdir = [Option(file.name,value="static/"+user+"/"+file.name) for file in files if file.is_file()]
userdata=user
print("Cambio",userdata)
return Form(
Select(
*subdir,name="data"),
Input(id="name-db", name="collection", placeholder="Enter a collection name"),
Button("Submit",type="submit"), hx_post="/createCollection",hx_swap="innerHTML",hx_target="#NewCollection" )
def create_or_load_db(path="./chroma_db",collection="init",Nodes=None,model="sentence-transformers/all-mpnet-base-v2"):
embed_model = HuggingFaceEmbedding(model_name="BAAI/bge-base-en-v1.5")
#embed_model = HuggingFaceInferenceAPIEmbedding(
#model_name="BAAI/bge-small-en-v1.5",
#token="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", # Optional
#)
db = chromadb.PersistentClient(path=path)
chroma_collection = db.get_or_create_collection(collection)
vector_store = ChromaVectorStore(chroma_collection=chroma_collection)
storage_context = StorageContext.from_defaults(vector_store=vector_store)
if Nodes:
index = VectorStoreIndex(
Nodes, storage_context=storage_context, embed_model=embed_model
)
else:
index = VectorStoreIndex.from_vector_store(
vector_store,
embed_model=embed_model,
)
return index
def post_process_documents(documents):
processed_documents = []
n=0
print(len(documents))
for doc in documents:
# 1. Text cleaning
n+=1
print(n)
text = doc.text.lower() # Convert to lowercase
# 2. Remove stopwords
stop_words = set("adssss")
tokens = text.split(" ")
filtered_text = ' '.join([word for word in tokens if word.lower() not in stop_words])
# 3. Custom metadata extraction (example)
metadata = doc.metadata.copy()
metadata['word_count'] = len(tokens)
# 4. Create a new document with processed text and updated metadata
processed_doc = Document(text=filtered_text, metadata=metadata)
processed_documents.append(processed_doc)
node_parser = SimpleNodeParser(chunk_size=200, chunk_overlap=30)
nodes = node_parser.get_nodes_from_documents(processed_documents)
return nodes
@app.get("/listmodelactives")
def listmodelactives():
try:
print(userdata)
except:
print("cambio")
return Div(id="options",hx_target="this",hx_swap="outerHTML",hx_get="/listmodelactives",hx_trigger="click from:#buttonMenuuser")
db = chromadb.PersistentClient(path="static/"+userdata+"/chroma_db")
files= db.list_collections()
collecs = [Option(file.name, value=file.name)for file in files]
return Form(
Select(
*collecs,name="data"),
Button("Submit",type="submit"),
hx_post="/loadCollection",hx_swap="innerHTML",hx_target="#Infomodel")
@app.post("/loadCollection")
def loadCollection(data:str):
global index
index=create_or_load_db(path="static/"+userdata+"/chroma_db",collection=data,model="BAAI/bge-m3")
return P("El usuario %s colleccion %s"%(userdata,data))
@app.post("/queryprompt")
def queryPrompt(question:str):
#index=load_create_db(collection="my_collection")
query_engine = index.as_query_engine()
response = query_engine.query(question)
return P(response)
@app.post("/chatData")
def questionChat(message:str):
chat_engine = index.as_chat_engine(
chat_mode="condense_plus_context",
memory=memory,
llm=llm_70b,
context_prompt=(
"You are a chatbot, able to have normal interactions, as well as talk"
" about an essay discussing IA and uses in leardeship."
"Here are the relevant documents for the context:\n"
"{context_str}"
"\nInstruction: Use the previous chat history, or the context above, to interact and help the user but only about tematic of the essay"
),
verbose=False,
)
response = chat_engine.chat(message)
return P(message),P(response)
@app.get("/SummarySources")
def SummarySources():
with os.scandir("static/"+userdata) as files:
subdir = [Option(file.name,value="static/"+userdata+"/"+file.name) for file in files if file.is_file()]
return Form("Este es muy caro para documentos grandes y tarda mucho",
Select(
*subdir,name="data"),
Input( name="query", placeholder="Enter a query"),
Button("Submit",type="submit"), hx_post="/SummaryMake",hx_swap="innerHTML",hx_target="#summaryR" )
@app.post("/SummaryMake")
def SummaryMake(data:str,query:str):
print(data,query)
docs = SimpleDirectoryReader(
input_files=[data]
).load_data()
print("p1")
summary_index = SummaryIndex.from_documents(docs)
print("p2")
summary_engine = summary_index.as_query_engine()
print("p3")
response = summary_engine.query(
query
)
print("p4")
return P(response)
@app.post("/createCollection")
def createCollection(data:str,collection:str):
print("Reading")
docs = SimpleDirectoryReader(
input_files=[data]
).load_data()
print("Process Documents")
Nodes=post_process_documents(docs)
print("create DB")
class MyThread(threading.Thread):
def run(self):
print("Hilo")
create_or_load_db(path="static/"+data.split("/")[1]+"/chroma_db",collection=collection,Nodes=Nodes,model="BAAI/bge-m3")
# create and start the thread
global t
t = MyThread()
t.start()
global t_time
t_time=time.time()
return Div("Iniciando carga de datos")
@app.get("/is_busy")
def is_busy():
try:
Busy= t.is_alive()
except:
Busy=False
if not Busy:
return Busy
else:
return "Esta ocupados desde hace %s , este es un proceso largo"%(str(time.time()-t_time))
@app.get("/")
def home():
page = Title('Super tutor'),Main(
Div('Este es el sistema de super tutor, ',
menuusers(listUsers()),
#A('A link', href='https://example.com'),
#Img(src="https://placehold.co/200"),
Div("Archivos",id="files"),
Div(id="NewCollection"),
Div("Estado",id="status",hx_target="this",hx_swap="innerHTML",hx_get="/is_busy",hx_trigger="every 60000ms"),
Div(
Div(id="options",hx_target="this",hx_swap="outerHTML",hx_get="/listmodelactives",hx_trigger="click from:#buttonMenuuser delay:3s"),
Div(id="Infomodel"),
#Div("Resumen",Div(id="summary",hx_target="this",hx_swap="outerHTML",hx_get="/SummarySources",hx_trigger="click from:#buttonMenuuser"),Div(id="summaryR")),
Div(
Form(
Input(id="question", name="message", placeholder="Enter a message"),
Button("Submit",type="submit"), hx_post="/chatData",hx_swap="afterend",hx_target="#questionR" ),
Div(id="questionR")
,id="questions"),
Div(
Form(
Input(id="query", name="question", placeholder="Enter a query"),
Button("Submit",type="submit"), hx_post="/queryprompt",hx_swap="innerHTML",hx_target="#queryR" ),
Div(id="queryR"),
id="query"),
id="chatbot")
))
return page
app.mount("/static", StaticFiles(directory="static"), name="static")
serve()