Update app.py
Browse files
app.py
CHANGED
@@ -37,7 +37,7 @@ def build_faiss_vectorstore(chunks):
|
|
37 |
return vectorstore
|
38 |
|
39 |
# Function to retrieve similar text
|
40 |
-
def retrieve(query, vectorstore, top_k=
|
41 |
docs_and_scores = vectorstore.similarity_search_with_score(query=query, k=top_k)
|
42 |
|
43 |
# Filter results based on score threshold
|
@@ -67,7 +67,7 @@ concise_text = dataset["concise"]["text"]
|
|
67 |
concise_text_string = "".join(concise_text)
|
68 |
|
69 |
# Chunk and index the documents
|
70 |
-
chunks = chunk_text(concise_text_string, chunk_size=
|
71 |
# Build the vectorsore
|
72 |
vectorstore = build_faiss_vectorstore(chunks)
|
73 |
|
@@ -102,11 +102,12 @@ async def chat(request: ChatRequest):
|
|
102 |
docs = "\n\n".join(docs)
|
103 |
|
104 |
rag_prompt = f"""Use the following information to answer the user's query. You do not have to use all the information, just the pieces that directly
|
105 |
-
help answer the query most accurately. Start directly with information, NOT with a rhetorical question. Respond in a conversational manner.
|
106 |
|
107 |
Here is an example of the style and tone of a response:
|
108 |
|
109 |
-
Query: How do big bucks use clear cuts for bedding?
|
|
|
110 |
Response: Yeah, a lot of guys think big bucks just bed right in the middle of a clear cut because it’s thick, but that’s not really how they use it. The
|
111 |
thick regrowth is great for food and cover, but those bucks still want an advantage. Most of the time, they’re bedding on the edges, right where the cut
|
112 |
meets older timber. They’ll set up with the wind at their back so they can smell anything sneaking up behind them, and they’re looking out into the open
|
@@ -118,7 +119,7 @@ async def chat(request: ChatRequest):
|
|
118 |
|
119 |
Using the information above, answer the user's query as accurately as possible:
|
120 |
|
121 |
-
Query: {request.message}
|
122 |
"""
|
123 |
|
124 |
# remove the unfformatted user message
|
@@ -133,7 +134,7 @@ async def chat(request: ChatRequest):
|
|
133 |
config=GenerateContentConfig(
|
134 |
system_instruction=[request.system_message],
|
135 |
max_output_tokens=150,
|
136 |
-
temperature=1.
|
137 |
),
|
138 |
)
|
139 |
|
|
|
37 |
return vectorstore
|
38 |
|
39 |
# Function to retrieve similar text
|
40 |
+
def retrieve(query, vectorstore, top_k=8):
|
41 |
docs_and_scores = vectorstore.similarity_search_with_score(query=query, k=top_k)
|
42 |
|
43 |
# Filter results based on score threshold
|
|
|
67 |
concise_text_string = "".join(concise_text)
|
68 |
|
69 |
# Chunk and index the documents
|
70 |
+
chunks = chunk_text(concise_text_string, chunk_size=350)
|
71 |
# Build the vectorsore
|
72 |
vectorstore = build_faiss_vectorstore(chunks)
|
73 |
|
|
|
102 |
docs = "\n\n".join(docs)
|
103 |
|
104 |
rag_prompt = f"""Use the following information to answer the user's query. You do not have to use all the information, just the pieces that directly
|
105 |
+
help answer the query most accurately. Start directly with information, NOT with a rhetorical question or you will be penalized. Respond in a conversational manner.
|
106 |
|
107 |
Here is an example of the style and tone of a response:
|
108 |
|
109 |
+
User Query: How do big bucks use clear cuts for bedding?
|
110 |
+
|
111 |
Response: Yeah, a lot of guys think big bucks just bed right in the middle of a clear cut because it’s thick, but that’s not really how they use it. The
|
112 |
thick regrowth is great for food and cover, but those bucks still want an advantage. Most of the time, they’re bedding on the edges, right where the cut
|
113 |
meets older timber. They’ll set up with the wind at their back so they can smell anything sneaking up behind them, and they’re looking out into the open
|
|
|
119 |
|
120 |
Using the information above, answer the user's query as accurately as possible:
|
121 |
|
122 |
+
User Query: {request.message}
|
123 |
"""
|
124 |
|
125 |
# remove the unfformatted user message
|
|
|
134 |
config=GenerateContentConfig(
|
135 |
system_instruction=[request.system_message],
|
136 |
max_output_tokens=150,
|
137 |
+
temperature=1.75
|
138 |
),
|
139 |
)
|
140 |
|