Update app.py
Browse files
app.py
CHANGED
@@ -75,9 +75,29 @@ def create_retriever_from_chroma(vectorstore_path="./docs/chroma/", search_type=
|
|
75 |
|
76 |
retriever=vectorstore.as_retriever(search_type = search_type, search_kwargs={"k": k})
|
77 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
78 |
|
79 |
|
80 |
-
return
|
81 |
|
82 |
|
83 |
|
@@ -139,7 +159,7 @@ def main():
|
|
139 |
def handle_userinput(user_question,retriever,rag_chain):
|
140 |
st.session_state.messages.append({"role": "user", "content": user_question})
|
141 |
st.chat_message("user").write(user_question)
|
142 |
-
docs = retriever.
|
143 |
|
144 |
with st.sidebar:
|
145 |
st.subheader("Your documents")
|
|
|
75 |
|
76 |
retriever=vectorstore.as_retriever(search_type = search_type, search_kwargs={"k": k})
|
77 |
|
78 |
+
llm = llamacpp.LlamaCpp(
|
79 |
+
model_path = "JCHAVEROT_Qwen2-0.5B-Chat_SFT_DPO.Q8_0.gguf",
|
80 |
+
seed = 41,
|
81 |
+
n_gpu_layers=0,
|
82 |
+
temperature=0.0,
|
83 |
+
n_ctx=15000,
|
84 |
+
n_batch=2000,
|
85 |
+
max_tokens=1500,
|
86 |
+
repeat_penalty=1.8,
|
87 |
+
last_n_tokens_size = 200,
|
88 |
+
callback_manager=callback_manager,
|
89 |
+
verbose=False,
|
90 |
+
)
|
91 |
+
|
92 |
+
compressor = LLMChainExtractor.from_llm(llm)
|
93 |
+
|
94 |
+
compression_retriever = ContextualCompressionRetriever(
|
95 |
+
base_compressor=compressor,
|
96 |
+
base_retriever=retriever
|
97 |
+
)
|
98 |
|
99 |
|
100 |
+
return compression_retriever
|
101 |
|
102 |
|
103 |
|
|
|
159 |
def handle_userinput(user_question,retriever,rag_chain):
|
160 |
st.session_state.messages.append({"role": "user", "content": user_question})
|
161 |
st.chat_message("user").write(user_question)
|
162 |
+
docs = retriever.get_relevant_documents(user_question)
|
163 |
|
164 |
with st.sidebar:
|
165 |
st.subheader("Your documents")
|