Spaces:
Sleeping
Sleeping
Update app.py
Browse files
app.py
CHANGED
|
@@ -8,61 +8,59 @@ from groq import Groq
|
|
| 8 |
from tqdm.auto import tqdm
|
| 9 |
import streamlit as st
|
| 10 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 11 |
# Constants (hardcoded)
|
| 12 |
FILE_PATH = "anjibot_chunks.json"
|
| 13 |
BATCH_SIZE = 384
|
| 14 |
INDEX_NAME = "groq-llama-3-rag"
|
| 15 |
-
PINECONE_API_KEY = os.getenv("PINECONE_API_KEY")
|
| 16 |
-
GROQ_API_KEY = os.getenv("GROQ_API_KEY")
|
| 17 |
-
|
| 18 |
-
|
| 19 |
-
# Load data once at the start
|
| 20 |
-
data = load_data(FILE_PATH)
|
| 21 |
-
|
| 22 |
-
# Initialize Pinecone and SentenceTransformer once
|
| 23 |
-
index = initialize_pinecone(PINECONE_API_KEY, INDEX_NAME, DIMENSIONS)
|
| 24 |
encoder = SentenceTransformer('dwzhu/e5-base-4k')
|
| 25 |
|
| 26 |
-
|
| 27 |
-
|
| 28 |
-
return json.load(file)
|
| 29 |
|
| 30 |
-
|
| 31 |
-
|
| 32 |
-
|
|
|
|
|
|
|
|
|
|
| 33 |
|
| 34 |
-
|
|
|
|
|
|
|
| 35 |
|
| 36 |
-
|
| 37 |
-
if index_name not in existing_indexes:
|
| 38 |
-
pc.create_index(index_name, dimension=dims, metric='cosine', spec=spec)
|
| 39 |
|
| 40 |
-
|
| 41 |
-
|
| 42 |
-
|
| 43 |
|
| 44 |
-
|
|
|
|
| 45 |
|
| 46 |
-
|
| 47 |
-
for
|
| 48 |
-
|
| 49 |
-
i_end = min(len(data['id']), i + BATCH_SIZE)
|
| 50 |
-
|
| 51 |
-
# Create batch
|
| 52 |
-
batch = {k: v[i:i_end] for k, v in data.items()}
|
| 53 |
-
|
| 54 |
-
# Create embeddings
|
| 55 |
-
chunks = [f'{x["title"]}: {x["content"]}' for x in batch["metadata"]]
|
| 56 |
-
embeds = encoder.encode(chunks)
|
| 57 |
|
| 58 |
-
|
| 59 |
-
|
| 60 |
|
| 61 |
-
|
| 62 |
-
|
| 63 |
-
|
| 64 |
|
| 65 |
-
def get_docs(query: str,
|
| 66 |
xq = encoder.encode(query)
|
| 67 |
res = index.query(vector=xq.tolist(), top_k=top_k, include_metadata=True)
|
| 68 |
return [x["metadata"]['content'] for x in res["matches"]]
|
|
@@ -88,20 +86,22 @@ def get_response(query: str, docs: list[str], groq_client: any) -> str:
|
|
| 88 |
)
|
| 89 |
return chat_response.choices[0].message.content
|
| 90 |
|
|
|
|
|
|
|
| 91 |
def handle_query(user_query: str):
|
| 92 |
-
# Upsert data into Pinecone (if necessary)
|
| 93 |
-
upsert_data_to_pinecone(index, data)
|
| 94 |
|
| 95 |
# Initialize Groq client
|
| 96 |
groq_client = Groq(api_key=GROQ_API_KEY)
|
| 97 |
|
| 98 |
# Get relevant documents
|
| 99 |
-
docs = get_docs(user_query,
|
| 100 |
|
| 101 |
# Generate and return response
|
| 102 |
response = get_response(user_query, docs, groq_client)
|
| 103 |
|
| 104 |
-
|
|
|
|
|
|
|
| 105 |
|
| 106 |
def main():
|
| 107 |
st.title("Ask Anjibot 2.0")
|
|
|
|
| 8 |
from tqdm.auto import tqdm
|
| 9 |
import streamlit as st
|
| 10 |
|
| 11 |
+
# Required imports
|
| 12 |
+
import json
|
| 13 |
+
import time
|
| 14 |
+
import os
|
| 15 |
+
from sentence_transformers import SentenceTransformer
|
| 16 |
+
from pinecone import Pinecone, ServerlessSpec
|
| 17 |
+
from groq import Groq
|
| 18 |
+
from tqdm.auto import tqdm
|
| 19 |
+
|
| 20 |
# Constants (hardcoded)
|
| 21 |
FILE_PATH = "anjibot_chunks.json"
|
| 22 |
BATCH_SIZE = 384
|
| 23 |
INDEX_NAME = "groq-llama-3-rag"
|
| 24 |
+
PINECONE_API_KEY = os.getenv("PINECONE_API_KEY") # Fixed syntax here
|
| 25 |
+
GROQ_API_KEY = os.getenv("GROQ_API_KEY") # Fixed s
|
| 26 |
+
DIMS = 768
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 27 |
encoder = SentenceTransformer('dwzhu/e5-base-4k')
|
| 28 |
|
| 29 |
+
with open(FILE_PATH, 'r') as file:
|
| 30 |
+
data= json.load(file)
|
|
|
|
| 31 |
|
| 32 |
+
pc = Pinecone(api_key=PINECONE_API_KEY)
|
| 33 |
+
spec = ServerlessSpec(cloud="aws", region='us-east-1')
|
| 34 |
+
existing_indexes = [index_info["name"] for index_info in pc.list_indexes()]
|
| 35 |
+
# Check if index already exists; if not, create it
|
| 36 |
+
if INDEX_NAME not in existing_indexes:
|
| 37 |
+
pc.create_index(INDEX_NAME, dimension=DIMS, metric='cosine', spec=spec)
|
| 38 |
|
| 39 |
+
# Wait for the index to be initialized
|
| 40 |
+
while not pc.describe_index(INDEX_NAME).status['ready']:
|
| 41 |
+
time.sleep(1)
|
| 42 |
|
| 43 |
+
index = pc.Index(INDEX_NAME)
|
|
|
|
|
|
|
| 44 |
|
| 45 |
+
for i in tqdm(range(0, len(data['id']), BATCH_SIZE)):
|
| 46 |
+
# Find end of batch
|
| 47 |
+
i_end = min(len(data['id']), i + BATCH_SIZE)
|
| 48 |
|
| 49 |
+
# Create batch
|
| 50 |
+
batch = {k: v[i:i_end] for k, v in data.items()}
|
| 51 |
|
| 52 |
+
# Create embeddings
|
| 53 |
+
chunks = [f'{x["title"]}: {x["content"]}' for x in batch["metadata"]]
|
| 54 |
+
embeds = encoder.encode(chunks)
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 55 |
|
| 56 |
+
# Ensure correct length
|
| 57 |
+
assert len(embeds) == (i_end - i)
|
| 58 |
|
| 59 |
+
# Upsert to Pinecone
|
| 60 |
+
to_upsert = list(zip(batch["id"], embeds, batch["metadata"]))
|
| 61 |
+
index.upsert(vectors=to_upsert)
|
| 62 |
|
| 63 |
+
def get_docs(query: str, top_k: int) -> list[str]:
|
| 64 |
xq = encoder.encode(query)
|
| 65 |
res = index.query(vector=xq.tolist(), top_k=top_k, include_metadata=True)
|
| 66 |
return [x["metadata"]['content'] for x in res["matches"]]
|
|
|
|
| 86 |
)
|
| 87 |
return chat_response.choices[0].message.content
|
| 88 |
|
| 89 |
+
|
| 90 |
+
|
| 91 |
def handle_query(user_query: str):
|
|
|
|
|
|
|
| 92 |
|
| 93 |
# Initialize Groq client
|
| 94 |
groq_client = Groq(api_key=GROQ_API_KEY)
|
| 95 |
|
| 96 |
# Get relevant documents
|
| 97 |
+
docs = get_docs(user_query, top_k=5)
|
| 98 |
|
| 99 |
# Generate and return response
|
| 100 |
response = get_response(user_query, docs, groq_client)
|
| 101 |
|
| 102 |
+
for word in response.split():
|
| 103 |
+
yield word + " "
|
| 104 |
+
time.sleep(0.05)
|
| 105 |
|
| 106 |
def main():
|
| 107 |
st.title("Ask Anjibot 2.0")
|