xangma
commited on
Commit
•
572a6c9
1
Parent(s):
df62f91
latest
Browse files- app.py +65 -41
- chain.py +62 -10
- requirements.txt +2 -1
app.py
CHANGED
@@ -8,14 +8,12 @@ import string
|
|
8 |
import sys
|
9 |
from pathlib import Path
|
10 |
import numpy as np
|
11 |
-
|
12 |
import chromadb
|
13 |
import gradio as gr
|
14 |
from chromadb.config import Settings
|
15 |
from langchain.docstore.document import Document
|
16 |
from langchain.embeddings import HuggingFaceEmbeddings, OpenAIEmbeddings
|
17 |
from langchain.vectorstores import Chroma
|
18 |
-
from langchain.retrievers import SVMRetriever
|
19 |
from chain import get_new_chain1
|
20 |
from ingest import embedding_chooser, ingest_docs
|
21 |
logging.basicConfig(stream=sys.stdout, level=logging.INFO)
|
@@ -97,14 +95,18 @@ def merge_collections(collection_load_names, vs_state, k_textbox, search_type_se
|
|
97 |
# merged_vectorstore.append(f.readlines())
|
98 |
return merged_vectorstore
|
99 |
|
100 |
-
def set_chain_up(openai_api_key, model_selector, k_textbox, search_type_selector, max_tokens_textbox, vectorstore_radio, vectorstore, agent):
|
101 |
if not agent or type(agent) == str:
|
102 |
if vectorstore != None:
|
103 |
if model_selector in ["gpt-3.5-turbo", "gpt-4"]:
|
104 |
if openai_api_key:
|
105 |
os.environ["OPENAI_API_KEY"] = openai_api_key
|
|
|
|
|
106 |
qa_chain = get_new_chain1(vectorstore, vectorstore_radio, model_selector, k_textbox, search_type_selector, max_tokens_textbox)
|
107 |
os.environ["OPENAI_API_KEY"] = ""
|
|
|
|
|
108 |
return qa_chain
|
109 |
else:
|
110 |
return 'no_open_aikey'
|
@@ -197,39 +199,7 @@ with block:
|
|
197 |
with gr.Tabs() as tabs:
|
198 |
with gr.TabItem("Chat", id=0):
|
199 |
with gr.Row():
|
200 |
-
|
201 |
-
placeholder="Paste your OpenAI API key (sk-...)",
|
202 |
-
show_label=False,
|
203 |
-
lines=1,
|
204 |
-
type="password",
|
205 |
-
)
|
206 |
-
model_selector = gr.Dropdown(
|
207 |
-
choices=["gpt-3.5-turbo", "gpt-4", "other"],
|
208 |
-
label="Model",
|
209 |
-
show_label=True,
|
210 |
-
value = "gpt-3.5-turbo"
|
211 |
-
)
|
212 |
-
k_textbox = gr.Textbox(
|
213 |
-
placeholder="k: Number of search results to consider",
|
214 |
-
label="Search Results k:",
|
215 |
-
show_label=True,
|
216 |
-
lines=1,
|
217 |
-
value="20",
|
218 |
-
)
|
219 |
-
search_type_selector = gr.Dropdown(
|
220 |
-
choices=["similarity", "mmr", "svm"],
|
221 |
-
label="Search Type",
|
222 |
-
show_label=True,
|
223 |
-
value = "similarity"
|
224 |
-
)
|
225 |
-
max_tokens_textbox = gr.Textbox(
|
226 |
-
placeholder="max_tokens: Maximum number of tokens to generate",
|
227 |
-
label="max_tokens",
|
228 |
-
show_label=True,
|
229 |
-
lines=1,
|
230 |
-
value="1000",
|
231 |
-
)
|
232 |
-
chatbot = gr.Chatbot()
|
233 |
with gr.Row():
|
234 |
clear_btn = gr.Button("Clear Chat", variant="secondary").style(full_width=False)
|
235 |
message = gr.Textbox(
|
@@ -240,12 +210,66 @@ with block:
|
|
240 |
submit = gr.Button(value="Send").style(full_width=False)
|
241 |
gr.Examples(
|
242 |
examples=[
|
243 |
-
"What does this code do?",
|
244 |
"I want to change the chat-pykg app to have a log viewer, where the user can see what python is doing in the background. How could I do that?",
|
245 |
-
"Hello, I want to allow chat-pykg to search
|
|
|
246 |
],
|
247 |
inputs=message,
|
248 |
)
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
249 |
|
250 |
gr.HTML(
|
251 |
"""
|
@@ -318,8 +342,8 @@ with block:
|
|
318 |
debug_state.value = False
|
319 |
radio_state = gr.State()
|
320 |
|
321 |
-
submit.click(set_chain_up, inputs=[openai_api_key_textbox, model_selector, k_textbox, search_type_selector, max_tokens_textbox, select_vectorstore_radio, vs_state, agent_state], outputs=[agent_state]).then(chat, inputs=[message, history_state, agent_state], outputs=[chatbot, history_state])
|
322 |
-
message.submit(set_chain_up, inputs=[openai_api_key_textbox, model_selector, k_textbox, search_type_selector, max_tokens_textbox, select_vectorstore_radio, vs_state, agent_state], outputs=[agent_state]).then(chat, inputs=[message, history_state, agent_state], outputs=[chatbot, history_state])
|
323 |
|
324 |
load_collections_button.click(merge_collections, inputs=[collections_viewer, vs_state, k_textbox, search_type_selector, select_vectorstore_radio, select_embedding_radio], outputs=[vs_state])#.then(change_tab, None, tabs) #.then(set_chain_up, inputs=[openai_api_key_textbox, model_selector, k_textbox, max_tokens_textbox, vs_state, agent_state], outputs=[agent_state])
|
325 |
make_collections_button.click(ingest_docs, inputs=[all_collections_state, all_collections_to_get, chunk_size_textbox, chunk_overlap_textbox, select_vectorstore_radio, select_embedding_radio, debug_state], outputs=[all_collections_state, all_collections_to_get], show_progress=True).then(update_checkboxgroup, inputs = [all_collections_state], outputs = [collections_viewer])
|
@@ -334,7 +358,7 @@ with block:
|
|
334 |
select_vectorstore_radio.change(update_radio, inputs = select_vectorstore_radio, outputs = make_vectorstore_radio)
|
335 |
|
336 |
# Whenever chain parameters change, destroy the agent.
|
337 |
-
input_list = [openai_api_key_textbox, model_selector, k_textbox, max_tokens_textbox, select_vectorstore_radio, make_embedding_radio]
|
338 |
output_list = [agent_state]
|
339 |
for input_item in input_list:
|
340 |
input_item.change(
|
|
|
8 |
import sys
|
9 |
from pathlib import Path
|
10 |
import numpy as np
|
|
|
11 |
import chromadb
|
12 |
import gradio as gr
|
13 |
from chromadb.config import Settings
|
14 |
from langchain.docstore.document import Document
|
15 |
from langchain.embeddings import HuggingFaceEmbeddings, OpenAIEmbeddings
|
16 |
from langchain.vectorstores import Chroma
|
|
|
17 |
from chain import get_new_chain1
|
18 |
from ingest import embedding_chooser, ingest_docs
|
19 |
logging.basicConfig(stream=sys.stdout, level=logging.INFO)
|
|
|
95 |
# merged_vectorstore.append(f.readlines())
|
96 |
return merged_vectorstore
|
97 |
|
98 |
+
def set_chain_up(openai_api_key, google_api_key, google_cse_id, model_selector, k_textbox, search_type_selector, max_tokens_textbox, vectorstore_radio, vectorstore, agent):
|
99 |
if not agent or type(agent) == str:
|
100 |
if vectorstore != None:
|
101 |
if model_selector in ["gpt-3.5-turbo", "gpt-4"]:
|
102 |
if openai_api_key:
|
103 |
os.environ["OPENAI_API_KEY"] = openai_api_key
|
104 |
+
os.environ["GOOGLE_API_KEY"] = google_api_key
|
105 |
+
os.environ["GOOGLE_CSE_ID"] = google_cse_id
|
106 |
qa_chain = get_new_chain1(vectorstore, vectorstore_radio, model_selector, k_textbox, search_type_selector, max_tokens_textbox)
|
107 |
os.environ["OPENAI_API_KEY"] = ""
|
108 |
+
os.environ["GOOGLE_API_KEY"] = ""
|
109 |
+
os.environ["GOOGLE_CSE_ID"] = ""
|
110 |
return qa_chain
|
111 |
else:
|
112 |
return 'no_open_aikey'
|
|
|
199 |
with gr.Tabs() as tabs:
|
200 |
with gr.TabItem("Chat", id=0):
|
201 |
with gr.Row():
|
202 |
+
chatbot = gr.Chatbot()
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
203 |
with gr.Row():
|
204 |
clear_btn = gr.Button("Clear Chat", variant="secondary").style(full_width=False)
|
205 |
message = gr.Textbox(
|
|
|
210 |
submit = gr.Button(value="Send").style(full_width=False)
|
211 |
gr.Examples(
|
212 |
examples=[
|
|
|
213 |
"I want to change the chat-pykg app to have a log viewer, where the user can see what python is doing in the background. How could I do that?",
|
214 |
+
"Hello, I want to allow chat-pykg to search google before answering. In the langchain docs it says you can use a tool to do this: from langchain.agents import load_tools\ntools = load_tools([“google-search”]). How would I need to change get_new_chain1 function to use tools when it needs to as well as searching the vectorstore? Thanks!",
|
215 |
+
"Great, thanks. What if I want to add other tools in the future? Can you please change get_new_chain1 function to do that?"
|
216 |
],
|
217 |
inputs=message,
|
218 |
)
|
219 |
+
with gr.Row():
|
220 |
+
with gr.Column(scale=1):
|
221 |
+
model_selector = gr.Dropdown(
|
222 |
+
choices=["gpt-3.5-turbo", "gpt-4", "other"],
|
223 |
+
label="Model",
|
224 |
+
show_label=True,
|
225 |
+
value = "gpt-4"
|
226 |
+
)
|
227 |
+
k_textbox = gr.Textbox(
|
228 |
+
placeholder="k: Number of search results to consider",
|
229 |
+
label="Search Results k:",
|
230 |
+
show_label=True,
|
231 |
+
lines=1,
|
232 |
+
value="20",
|
233 |
+
)
|
234 |
+
search_type_selector = gr.Dropdown(
|
235 |
+
choices=["similarity", "mmr", "svm"],
|
236 |
+
label="Search Type",
|
237 |
+
show_label=True,
|
238 |
+
value = "similarity"
|
239 |
+
)
|
240 |
+
max_tokens_textbox = gr.Textbox(
|
241 |
+
placeholder="max_tokens: Maximum number of tokens to generate",
|
242 |
+
label="max_tokens",
|
243 |
+
show_label=True,
|
244 |
+
lines=1,
|
245 |
+
value="500",
|
246 |
+
)
|
247 |
+
with gr.Column(scale=1):
|
248 |
+
gr.HTML("")
|
249 |
+
with gr.Column(scale=1):
|
250 |
+
gr.HTML("")
|
251 |
+
with gr.Column(scale=1):
|
252 |
+
openai_api_key_textbox = gr.Textbox(
|
253 |
+
placeholder="Paste your OpenAI API key (sk-...)",
|
254 |
+
show_label=True,
|
255 |
+
lines=1,
|
256 |
+
type="password",
|
257 |
+
label="OpenAI API Key",
|
258 |
+
)
|
259 |
+
google_api_key_textbox = gr.Textbox(
|
260 |
+
placeholder="Paste your Google API key (AIza...)",
|
261 |
+
show_label=True,
|
262 |
+
lines=1,
|
263 |
+
type="password",
|
264 |
+
label="Google API Key",
|
265 |
+
)
|
266 |
+
google_cse_id_textbox = gr.Textbox(
|
267 |
+
placeholder="Paste your Google CSE ID (0123...)",
|
268 |
+
show_label=True,
|
269 |
+
lines=1,
|
270 |
+
type="password",
|
271 |
+
label="Google CSE ID",
|
272 |
+
)
|
273 |
|
274 |
gr.HTML(
|
275 |
"""
|
|
|
342 |
debug_state.value = False
|
343 |
radio_state = gr.State()
|
344 |
|
345 |
+
submit.click(set_chain_up, inputs=[openai_api_key_textbox, google_api_key_textbox, google_cse_id_textbox, model_selector, k_textbox, search_type_selector, max_tokens_textbox, select_vectorstore_radio, vs_state, agent_state], outputs=[agent_state]).then(chat, inputs=[message, history_state, agent_state], outputs=[chatbot, history_state])
|
346 |
+
message.submit(set_chain_up, inputs=[openai_api_key_textbox, google_api_key_textbox, google_cse_id_textbox, model_selector, k_textbox, search_type_selector, max_tokens_textbox, select_vectorstore_radio, vs_state, agent_state], outputs=[agent_state]).then(chat, inputs=[message, history_state, agent_state], outputs=[chatbot, history_state])
|
347 |
|
348 |
load_collections_button.click(merge_collections, inputs=[collections_viewer, vs_state, k_textbox, search_type_selector, select_vectorstore_radio, select_embedding_radio], outputs=[vs_state])#.then(change_tab, None, tabs) #.then(set_chain_up, inputs=[openai_api_key_textbox, model_selector, k_textbox, max_tokens_textbox, vs_state, agent_state], outputs=[agent_state])
|
349 |
make_collections_button.click(ingest_docs, inputs=[all_collections_state, all_collections_to_get, chunk_size_textbox, chunk_overlap_textbox, select_vectorstore_radio, select_embedding_radio, debug_state], outputs=[all_collections_state, all_collections_to_get], show_progress=True).then(update_checkboxgroup, inputs = [all_collections_state], outputs = [collections_viewer])
|
|
|
358 |
select_vectorstore_radio.change(update_radio, inputs = select_vectorstore_radio, outputs = make_vectorstore_radio)
|
359 |
|
360 |
# Whenever chain parameters change, destroy the agent.
|
361 |
+
input_list = [openai_api_key_textbox, model_selector, k_textbox, search_type_selector, max_tokens_textbox, select_vectorstore_radio, make_embedding_radio]
|
362 |
output_list = [agent_state]
|
363 |
for input_item in input_list:
|
364 |
input_item.change(
|
chain.py
CHANGED
@@ -15,28 +15,70 @@ from langchain.chains.combine_documents.base import BaseCombineDocumentsChain
|
|
15 |
from langchain.chains.llm import LLMChain
|
16 |
from langchain.schema import BaseLanguageModel, BaseRetriever, Document
|
17 |
from langchain.prompts.prompt import PromptTemplate
|
|
|
|
|
|
|
|
|
18 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
19 |
|
20 |
def get_new_chain1(vectorstore, vectorstore_radio, model_selector, k_textbox, search_type_selector, max_tokens_textbox) -> Chain:
|
21 |
retriever = None
|
22 |
if vectorstore_radio == 'Chroma':
|
23 |
retriever = vectorstore.as_retriever(search_type=search_type_selector)
|
24 |
retriever.search_kwargs = {"k":int(k_textbox)}
|
|
|
|
|
25 |
if vectorstore_radio == 'raw':
|
26 |
if search_type_selector == 'svm':
|
27 |
retriever = SVMRetriever.from_texts(merged_vectorstore, embedding_function)
|
28 |
retriever.k = int(k_textbox)
|
29 |
|
30 |
-
|
31 |
-
You are given the
|
32 |
-
|
|
|
|
|
|
|
33 |
If you don't know the answer, just say that you don't know, don't try to make up an answer.
|
34 |
-
|
35 |
=========
|
36 |
-
{context}
|
37 |
=========
|
38 |
-
|
39 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
40 |
if model_selector in ['gpt-4', 'gpt-3.5-turbo']:
|
41 |
llm = ChatOpenAI(client = None, temperature=0.7, model_name=model_selector)
|
42 |
doc_chain_llm = ChatOpenAI(client = None, streaming=True, callback_manager=CallbackManager([StreamingStdOutCallbackHandler()]), verbose=True, temperature=0.7, model_name=model_selector, max_tokens=int(max_tokens_textbox))
|
@@ -49,8 +91,18 @@ def get_new_chain1(vectorstore, vectorstore_radio, model_selector, k_textbox, se
|
|
49 |
# memory = ConversationKGMemory(llm=llm, input_key="question", output_key="answer")
|
50 |
memory = ConversationBufferWindowMemory(input_key="question", output_key="answer", k=5)
|
51 |
|
52 |
-
|
53 |
-
retriever=retriever, memory=memory, combine_docs_chain=doc_chain, question_generator=question_generator, verbose=True, callback_manager=CallbackManager([StreamingStdOutCallbackHandler()]))
|
54 |
-
# qa._get_docs = _get_docs.__get__(qa, ConversationalRetrievalChain)
|
55 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
56 |
return qa
|
|
|
15 |
from langchain.chains.llm import LLMChain
|
16 |
from langchain.schema import BaseLanguageModel, BaseRetriever, Document
|
17 |
from langchain.prompts.prompt import PromptTemplate
|
18 |
+
from langchain.utilities.google_serper import GoogleSerperAPIWrapper
|
19 |
+
from langchain.utilities.google_search import GoogleSearchAPIWrapper
|
20 |
+
from langchain.agents.self_ask_with_search.base import SelfAskWithSearchChain
|
21 |
+
from langchain.agents.self_ask_with_search.prompt import PROMPT
|
22 |
|
23 |
+
class ConversationalRetrievalChainWithGoogleSearch(ConversationalRetrievalChain):
|
24 |
+
google_search_tool: GoogleSearchAPIWrapper
|
25 |
+
|
26 |
+
def _get_docs(self, question: str, inputs: Dict[str, Any]) -> List[Document]:
|
27 |
+
# Get documents from the retriever
|
28 |
+
docs_from_retriever = self.retriever.get_relevant_documents(question)
|
29 |
+
|
30 |
+
# Get search results from Google Search
|
31 |
+
search_results = self.google_search_tool.results(question, num_results=self.google_search_tool.k)
|
32 |
+
|
33 |
+
# Create documents from the search results
|
34 |
+
docs_from_search = []
|
35 |
+
for result in search_results:
|
36 |
+
content = result.get("snippet", "")
|
37 |
+
metadata = {"title": result["title"], "link": result["link"]}
|
38 |
+
docs_from_search.append(Document(page_content=content, metadata=metadata))
|
39 |
+
|
40 |
+
# Combine both lists of documents
|
41 |
+
docs = docs_from_retriever + docs_from_search
|
42 |
+
|
43 |
+
return self._reduce_tokens_below_limit(docs)
|
44 |
|
45 |
def get_new_chain1(vectorstore, vectorstore_radio, model_selector, k_textbox, search_type_selector, max_tokens_textbox) -> Chain:
|
46 |
retriever = None
|
47 |
if vectorstore_radio == 'Chroma':
|
48 |
retriever = vectorstore.as_retriever(search_type=search_type_selector)
|
49 |
retriever.search_kwargs = {"k":int(k_textbox)}
|
50 |
+
if search_type_selector == 'mmr':
|
51 |
+
retriever.search_kwargs = {"k":int(k_textbox), "fetch_k":4*int(k_textbox)}
|
52 |
if vectorstore_radio == 'raw':
|
53 |
if search_type_selector == 'svm':
|
54 |
retriever = SVMRetriever.from_texts(merged_vectorstore, embedding_function)
|
55 |
retriever.k = int(k_textbox)
|
56 |
|
57 |
+
qa_template = """You are called chat-pykg and are an AI assistant coded in python using langchain and gradio. You are very helpful for answering questions about programming with various open source packages and libraries.
|
58 |
+
You are given snippets of code and information in the Context below, as well as a Question to give a Helpful answer to.
|
59 |
+
Due to data size limitations, the snippets of code in the Context have been specifically filtered/selected for their relevance from a document store containing code from one or many packages and libraries.
|
60 |
+
Each of the code snippets is marked with '# source: package/filename' so you can attempt to establish where they are located in their package structure and gain more understanding of the code.
|
61 |
+
Please provide a helpful answer in markdown to the Question.
|
62 |
+
Do not make up any hyperlinks that are not in the Context.
|
63 |
If you don't know the answer, just say that you don't know, don't try to make up an answer.
|
64 |
+
|
65 |
=========
|
66 |
+
Context:{context}
|
67 |
=========
|
68 |
+
Question: {question}
|
69 |
+
Helpful answer:"""
|
70 |
+
QA_PROMPT.template = qa_template
|
71 |
+
|
72 |
+
condense_question_template = """Given the following conversation and a Follow Up Input, rephrase the Follow Up Input to be a Standalone question.
|
73 |
+
The Standalone question will be used for retrieving relevant source code and information from a document store, where each document is marked with '# source: package/filename'.
|
74 |
+
Therefore, in your Standalone question you must try to include references to related code or sources that have been mentioned in the Follow Up Input or Chat History.
|
75 |
+
=========
|
76 |
+
Chat History:
|
77 |
+
{chat_history}
|
78 |
+
=========
|
79 |
+
Follow Up Input: {question}
|
80 |
+
Standalone question in markdown:"""
|
81 |
+
CONDENSE_QUESTION_PROMPT.template = condense_question_template
|
82 |
if model_selector in ['gpt-4', 'gpt-3.5-turbo']:
|
83 |
llm = ChatOpenAI(client = None, temperature=0.7, model_name=model_selector)
|
84 |
doc_chain_llm = ChatOpenAI(client = None, streaming=True, callback_manager=CallbackManager([StreamingStdOutCallbackHandler()]), verbose=True, temperature=0.7, model_name=model_selector, max_tokens=int(max_tokens_textbox))
|
|
|
91 |
# memory = ConversationKGMemory(llm=llm, input_key="question", output_key="answer")
|
92 |
memory = ConversationBufferWindowMemory(input_key="question", output_key="answer", k=5)
|
93 |
|
94 |
+
google_search_tool = GoogleSearchAPIWrapper(search_engine = "google", k = int(int(k_textbox)/2))
|
|
|
|
|
95 |
|
96 |
+
qa_orig = ConversationalRetrievalChain(
|
97 |
+
retriever=retriever, memory=memory, combine_docs_chain=doc_chain, question_generator=question_generator, verbose=True, callback_manager=CallbackManager([StreamingStdOutCallbackHandler()]))
|
98 |
+
qa_with_google_search = ConversationalRetrievalChainWithGoogleSearch(
|
99 |
+
retriever=retriever,
|
100 |
+
memory=memory,
|
101 |
+
combine_docs_chain=doc_chain,
|
102 |
+
question_generator=question_generator,
|
103 |
+
google_search_tool=google_search_tool,
|
104 |
+
verbose=True,
|
105 |
+
callback_manager=CallbackManager([StreamingStdOutCallbackHandler()])
|
106 |
+
)
|
107 |
+
qa = qa_orig
|
108 |
return qa
|
requirements.txt
CHANGED
@@ -7,4 +7,5 @@ transformers
|
|
7 |
gradio
|
8 |
chromadb
|
9 |
sentence_transformers
|
10 |
-
python-magic
|
|
|
|
7 |
gradio
|
8 |
chromadb
|
9 |
sentence_transformers
|
10 |
+
python-magic
|
11 |
+
google-api-python-client
|