Spaces:
Sleeping
Sleeping
clementsan
commited on
Commit
•
1ef8d7c
1
Parent(s):
2239106
Include ephemeral client and collection_name for chromadb
Browse files
app.py
CHANGED
@@ -11,6 +11,9 @@ from langchain.chains import ConversationChain
|
|
11 |
from langchain.memory import ConversationBufferMemory
|
12 |
from langchain.llms import HuggingFaceHub
|
13 |
|
|
|
|
|
|
|
14 |
from transformers import AutoTokenizer
|
15 |
import transformers
|
16 |
import torch
|
@@ -50,11 +53,14 @@ def load_doc(list_file_path, chunk_size, chunk_overlap):
|
|
50 |
|
51 |
|
52 |
# Create vector database
|
53 |
-
def create_db(splits):
|
54 |
embedding = HuggingFaceEmbeddings()
|
|
|
55 |
vectordb = Chroma.from_documents(
|
56 |
documents=splits,
|
57 |
embedding=embedding,
|
|
|
|
|
58 |
# persist_directory=default_persist_directory
|
59 |
)
|
60 |
return vectordb
|
@@ -147,16 +153,18 @@ def initialize_database(list_file_obj, chunk_size, chunk_overlap, progress=gr.Pr
|
|
147 |
# Create list of documents (when valid)
|
148 |
#file_path = file_obj.name
|
149 |
list_file_path = [x.name for x in list_file_obj if x is not None]
|
150 |
-
|
|
|
|
|
151 |
progress(0.25, desc="Loading document...")
|
152 |
# Load document and create splits
|
153 |
doc_splits = load_doc(list_file_path, chunk_size, chunk_overlap)
|
154 |
# Create or load Vector database
|
155 |
progress(0.5, desc="Generating vector database...")
|
156 |
# global vector_db
|
157 |
-
vector_db = create_db(doc_splits)
|
158 |
progress(0.9, desc="Done!")
|
159 |
-
return vector_db, "Complete!"
|
160 |
|
161 |
|
162 |
def initialize_LLM(llm_option, llm_temperature, max_tokens, top_k, vector_db, progress=gr.Progress()):
|
@@ -211,6 +219,7 @@ def demo():
|
|
211 |
with gr.Blocks(theme="base") as demo:
|
212 |
vector_db = gr.State()
|
213 |
qa_chain = gr.State()
|
|
|
214 |
|
215 |
gr.Markdown(
|
216 |
"""<center><h2>PDF-based chatbot (powered by LangChain and open-source LLMs)</center></h2>
|
@@ -270,7 +279,7 @@ def demo():
|
|
270 |
#upload_btn.upload(upload_file, inputs=[upload_btn], outputs=[document])
|
271 |
db_btn.click(initialize_database, \
|
272 |
inputs=[document, slider_chunk_size, slider_chunk_overlap], \
|
273 |
-
outputs=[vector_db, db_progress])
|
274 |
qachain_btn.click(initialize_LLM, \
|
275 |
inputs=[llm_btn, slider_temperature, slider_maxtokens, slider_topk, vector_db], \
|
276 |
outputs=[qa_chain, llm_progress]).then(lambda:[None,"",0,"",0], \
|
|
|
11 |
from langchain.memory import ConversationBufferMemory
|
12 |
from langchain.llms import HuggingFaceHub
|
13 |
|
14 |
+
from pathlib import Path
|
15 |
+
import chromadb
|
16 |
+
|
17 |
from transformers import AutoTokenizer
|
18 |
import transformers
|
19 |
import torch
|
|
|
53 |
|
54 |
|
55 |
# Create vector database
|
56 |
+
def create_db(splits, collection_name):
|
57 |
embedding = HuggingFaceEmbeddings()
|
58 |
+
new_client = chromadb.EphemeralClient()
|
59 |
vectordb = Chroma.from_documents(
|
60 |
documents=splits,
|
61 |
embedding=embedding,
|
62 |
+
client=new_client,
|
63 |
+
collection_name=collection_name,
|
64 |
# persist_directory=default_persist_directory
|
65 |
)
|
66 |
return vectordb
|
|
|
153 |
# Create list of documents (when valid)
|
154 |
#file_path = file_obj.name
|
155 |
list_file_path = [x.name for x in list_file_obj if x is not None]
|
156 |
+
collection_name = Path(list_file_path[0]).stem
|
157 |
+
# print('list_file_path: ', list_file_path)
|
158 |
+
# print('Collection name: ', collection_name)
|
159 |
progress(0.25, desc="Loading document...")
|
160 |
# Load document and create splits
|
161 |
doc_splits = load_doc(list_file_path, chunk_size, chunk_overlap)
|
162 |
# Create or load Vector database
|
163 |
progress(0.5, desc="Generating vector database...")
|
164 |
# global vector_db
|
165 |
+
vector_db = create_db(doc_splits, collection_name)
|
166 |
progress(0.9, desc="Done!")
|
167 |
+
return vector_db, collection_name, "Complete!"
|
168 |
|
169 |
|
170 |
def initialize_LLM(llm_option, llm_temperature, max_tokens, top_k, vector_db, progress=gr.Progress()):
|
|
|
219 |
with gr.Blocks(theme="base") as demo:
|
220 |
vector_db = gr.State()
|
221 |
qa_chain = gr.State()
|
222 |
+
collection_name = gr.State()
|
223 |
|
224 |
gr.Markdown(
|
225 |
"""<center><h2>PDF-based chatbot (powered by LangChain and open-source LLMs)</center></h2>
|
|
|
279 |
#upload_btn.upload(upload_file, inputs=[upload_btn], outputs=[document])
|
280 |
db_btn.click(initialize_database, \
|
281 |
inputs=[document, slider_chunk_size, slider_chunk_overlap], \
|
282 |
+
outputs=[vector_db, collection_name, db_progress])
|
283 |
qachain_btn.click(initialize_LLM, \
|
284 |
inputs=[llm_btn, slider_temperature, slider_maxtokens, slider_topk, vector_db], \
|
285 |
outputs=[qa_chain, llm_progress]).then(lambda:[None,"",0,"",0], \
|