Spaces:
Runtime error
Runtime error
Update app.py
Browse files
app.py
CHANGED
@@ -22,62 +22,12 @@ rate_limiter = InMemoryRateLimiter(
|
|
22 |
check_every_n_seconds=0.01, # Wake up every 100 ms to check whether allowed to make a request,
|
23 |
max_bucket_size=10, # Controls the maximum burst size.
|
24 |
)
|
25 |
-
"""
|
26 |
-
# get data
|
27 |
-
urlsfile = open("urls.txt")
|
28 |
-
urls = urlsfile.readlines()
|
29 |
-
urls = [url.replace("\n","") for url in urls]
|
30 |
-
urlsfile.close()
|
31 |
-
|
32 |
-
# Load, chunk and index the contents of the blog.
|
33 |
-
loader = WebBaseLoader(urls)
|
34 |
-
docs = loader.load()
|
35 |
-
|
36 |
-
# load arxiv papers
|
37 |
-
arxivfile = open("arxiv.txt")
|
38 |
-
arxivs = arxivfile.readlines()
|
39 |
-
arxivs = [arxiv.replace("\n","") for arxiv in arxivs]
|
40 |
-
arxivfile.close()
|
41 |
|
42 |
retriever = ArxivRetriever(
|
43 |
load_max_docs=2,
|
44 |
get_ful_documents=True,
|
45 |
)
|
46 |
|
47 |
-
for arxiv in arxivs:
|
48 |
-
doc = retriever.invoke(arxiv)
|
49 |
-
doc[0].metadata['Published'] = str(doc[0].metadata['Published'])
|
50 |
-
docs.append(doc[0])
|
51 |
-
|
52 |
-
|
53 |
-
def format_docs(docs):
|
54 |
-
return "\n\n".join(doc.page_content for doc in docs)
|
55 |
-
|
56 |
-
def RAG(llm, docs, embeddings):
|
57 |
-
|
58 |
-
# Split text
|
59 |
-
text_splitter = RecursiveCharacterTextSplitter(chunk_size=1000, chunk_overlap=200)
|
60 |
-
splits = text_splitter.split_documents(docs)
|
61 |
-
|
62 |
-
# Create vector store
|
63 |
-
vectorstore = Chroma.from_documents(documents=splits, embedding=embeddings)
|
64 |
-
|
65 |
-
# Retrieve and generate using the relevant snippets of the documents
|
66 |
-
retriever = vectorstore.as_retriever()
|
67 |
-
|
68 |
-
# Prompt basis example for RAG systems
|
69 |
-
prompt = hub.pull("rlm/rag-prompt")
|
70 |
-
|
71 |
-
# Create the chain
|
72 |
-
rag_chain = (
|
73 |
-
{"context": retriever | format_docs, "question": RunnablePassthrough()}
|
74 |
-
| prompt
|
75 |
-
| llm
|
76 |
-
| StrOutputParser()
|
77 |
-
)
|
78 |
-
|
79 |
-
return rag_chain
|
80 |
-
|
81 |
# LLM model
|
82 |
llm = ChatMistralAI(model="mistral-large-latest", rate_limiter=rate_limiter)
|
83 |
|
@@ -87,10 +37,48 @@ embed_model = "sentence-transformers/multi-qa-distilbert-cos-v1"
|
|
87 |
embeddings = HuggingFaceInstructEmbeddings(model_name=embed_model)
|
88 |
# embeddings = MistralAIEmbeddings()
|
89 |
|
90 |
-
|
91 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
92 |
|
93 |
-
def handle_prompt(message, history):
|
|
|
|
|
|
|
|
|
|
|
94 |
try:
|
95 |
# Stream output
|
96 |
out=""
|
@@ -99,17 +87,19 @@ def handle_prompt(message, history):
|
|
99 |
yield out
|
100 |
except:
|
101 |
raise gr.Error("Requests rate limit exceeded")
|
102 |
-
"""
|
103 |
|
104 |
-
def handle_prompt(message, history, input1):
|
105 |
-
return f"arxiv code: {input1}, {message}"
|
106 |
|
|
|
107 |
|
108 |
-
|
109 |
|
110 |
-
|
111 |
-
|
112 |
-
|
113 |
-
|
|
|
|
|
|
|
114 |
|
115 |
-
|
|
|
|
22 |
check_every_n_seconds=0.01, # Wake up every 100 ms to check whether allowed to make a request,
|
23 |
max_bucket_size=10, # Controls the maximum burst size.
|
24 |
)
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
25 |
|
26 |
retriever = ArxivRetriever(
|
27 |
load_max_docs=2,
|
28 |
get_ful_documents=True,
|
29 |
)
|
30 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
31 |
# LLM model
|
32 |
llm = ChatMistralAI(model="mistral-large-latest", rate_limiter=rate_limiter)
|
33 |
|
|
|
37 |
embeddings = HuggingFaceInstructEmbeddings(model_name=embed_model)
|
38 |
# embeddings = MistralAIEmbeddings()
|
39 |
|
40 |
+
def initialize(arxivcode):
|
41 |
+
docs = retriever.invoke(arxiv)
|
42 |
+
docs[0].metadata['Published'] = str(doc[0].metadata['Published'])
|
43 |
+
|
44 |
+
def format_docs(docs):
|
45 |
+
return "\n\n".join(doc.page_content for doc in docs)
|
46 |
+
|
47 |
+
def RAG(llm, docs, embeddings):
|
48 |
+
|
49 |
+
# Split text
|
50 |
+
text_splitter = RecursiveCharacterTextSplitter(chunk_size=1000, chunk_overlap=200)
|
51 |
+
splits = text_splitter.split_documents(docs)
|
52 |
+
|
53 |
+
# Create vector store
|
54 |
+
vectorstore = Chroma.from_documents(documents=splits, embedding=embeddings)
|
55 |
+
|
56 |
+
# Retrieve and generate using the relevant snippets of the documents
|
57 |
+
retriever = vectorstore.as_retriever()
|
58 |
+
|
59 |
+
# Prompt basis example for RAG systems
|
60 |
+
prompt = hub.pull("rlm/rag-prompt")
|
61 |
+
|
62 |
+
# Create the chain
|
63 |
+
rag_chain = (
|
64 |
+
{"context": retriever | format_docs, "question": RunnablePassthrough()}
|
65 |
+
| prompt
|
66 |
+
| llm
|
67 |
+
| StrOutputParser()
|
68 |
+
)
|
69 |
+
|
70 |
+
return rag_chain
|
71 |
+
|
72 |
+
return RAG(llm, docs, embeddings)
|
73 |
+
|
74 |
+
rag_chain = None
|
75 |
|
76 |
+
def handle_prompt(message, history, arxivcode):
|
77 |
+
if rag_chain is None:
|
78 |
+
# initialize RAG chain
|
79 |
+
# RAG chain
|
80 |
+
rag_chain = initialize(arxivcode)
|
81 |
+
|
82 |
try:
|
83 |
# Stream output
|
84 |
out=""
|
|
|
87 |
yield out
|
88 |
except:
|
89 |
raise gr.Error("Requests rate limit exceeded")
|
|
|
90 |
|
|
|
|
|
91 |
|
92 |
+
greetingsmessage = "Hi, I'm your personal arXiv reader. Ask me questions about the arXiv paper above"
|
93 |
|
94 |
+
with gr.Blocks() as demo:
|
95 |
|
96 |
+
arxiv_code = gr.Textbox("", label="arxiv.number")
|
97 |
+
|
98 |
+
gr.ChatInterface(handle_prompt, type="messages", theme=gr.themes.Soft(),
|
99 |
+
description=greetingsmessage,
|
100 |
+
additional_inputs=[arxiv_code]
|
101 |
+
)
|
102 |
+
|
103 |
|
104 |
+
if __name__ == "__main__":
|
105 |
+
demo.launch()
|