Spaces:
Sleeping
Sleeping
chatbot infrastructure + basic rag outline
Browse files
model.py
CHANGED
|
@@ -1,4 +1,72 @@
|
|
| 1 |
import gradio as gr
|
| 2 |
from sentence_transformers import SentenceTransformer
|
| 3 |
import torch
|
| 4 |
-
from huggingface_hub import InferenceClient
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
import gradio as gr
|
| 2 |
from sentence_transformers import SentenceTransformer
|
| 3 |
import torch
|
| 4 |
+
from huggingface_hub import InferenceClient
|
| 5 |
+
|
| 6 |
+
# opening the file when it's ready should go here
|
| 7 |
+
# finlitText = file.read()
|
| 8 |
+
|
| 9 |
+
model = SentenceTransformer('all-MiniLM-L6-v2')
|
| 10 |
+
|
| 11 |
+
def preprocessText(text):
|
| 12 |
+
cleanedText = text.strip()
|
| 13 |
+
chunks = cleanedText.strip("\n")
|
| 14 |
+
cleanedChunks = [chunk.strip() for chunk in chunks if chunk is not None]
|
| 15 |
+
return cleanedChunks
|
| 16 |
+
|
| 17 |
+
def createEmbeddings(textChunks):
|
| 18 |
+
chunkEmbeddings = model.encode(cleanedChunks, convert_to_tensor = True)
|
| 19 |
+
return chunkEmbeddings
|
| 20 |
+
|
| 21 |
+
def getTopChunks(query, chunkEmbeddings, textChunks):
|
| 22 |
+
queryEmbedding = model.encode(query, convert_to_tensor = True)
|
| 23 |
+
queryEmbeddingNormalized = queryEmbedding / queryEmbedding.norm()
|
| 24 |
+
chunkEmbeddingsNormalized = chunkEmbeddings / chunkEmbeddings.norm(dim = 1, keepdim = True)
|
| 25 |
+
similarities = torch.matmul(chunkEmbeddingsNormalized, queryEmbeddingNormalized)
|
| 26 |
+
topIndices = torch.topk(similarities, k=3).indices
|
| 27 |
+
topChunks = [textChunks[i] for i in topIndices]
|
| 28 |
+
return topChunks
|
| 29 |
+
|
| 30 |
+
# cleanedChunks = preprocessText(finlitText)
|
| 31 |
+
# chunkEmbeddings = createEmbeddings(cleanedChunks)
|
| 32 |
+
|
| 33 |
+
client = InferenceClient("???")
|
| 34 |
+
|
| 35 |
+
def respond(message, history):
|
| 36 |
+
messages = [{"role": "system",
|
| 37 |
+
"content": "You are a friendly, approachable AI assistant whose main goal is to help high school and college students to learn more about productivity, setting goals for their education, and financial literacy."
|
| 38 |
+
"Keep responses between 200-300 words unless asked for more detail about your suggestions from the user."
|
| 39 |
+
"Explanations should be clear with examples and always include actionable steps, following this format:"
|
| 40 |
+
"User: What is the 50/30/20 rule when it comes to budgeting?"
|
| 41 |
+
"AI: Great question! The 50/30/20 rule states that you should spend 50% of your income on needs, 30% of your income on wants, and 20% of your income on investing. For example, if you make $4,000 each month, you should spend $2,000 on your needs, $1,200 on your wants, and $800 on investing. That way, you can set aside money to take care of yourself while still making progress towards saving up for the things that aren't as essential."}]
|
| 42 |
+
|
| 43 |
+
if history:
|
| 44 |
+
messages.extend(history)
|
| 45 |
+
|
| 46 |
+
# topResults = getTopChunks(message, chunkEmbeddings, cleanedChunks)
|
| 47 |
+
# context = "\n\n".join(topResults)
|
| 48 |
+
|
| 49 |
+
# messages.append({"role": "system",
|
| 50 |
+
# "content": context})
|
| 51 |
+
|
| 52 |
+
messages.append({"role": "user",
|
| 53 |
+
"content": message})
|
| 54 |
+
|
| 55 |
+
response = ""
|
| 56 |
+
|
| 57 |
+
responseStream = client.chat_completion(
|
| 58 |
+
messages, stream = True, max_tokens = 1024, temperature = 0.4
|
| 59 |
+
)
|
| 60 |
+
|
| 61 |
+
for segment in responseStream:
|
| 62 |
+
token = segment.choices[0].delta.content
|
| 63 |
+
if token is not None:
|
| 64 |
+
response += token
|
| 65 |
+
yield response
|
| 66 |
+
|
| 67 |
+
chatbot = gr.ChatInterface(response, title = "Student Formula Bot 🔬",
|
| 68 |
+
description = "Welcome to the core component of our project, The Student Formula: the RAG chatbot! With the ability to act as a finance tutor, accountability buddy, and goal-setting partner all in one, it's designed to best suit your needs on the way to productivity and success.")
|
| 69 |
+
|
| 70 |
+
chatbot.launch()
|
| 71 |
+
|
| 72 |
+
|