Hugging Face's logo

Spaces:
WrapIt
/
capstonefinal
Sleeping

App Files Community
capstonefinal
File size: 4,859 Bytes 
706313e
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
652389b
 
 
 
706313e
 
 
 
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
 
import gradio as gr
import random
import os
from huggingface_hub import InferenceClient
from sentence_transformers import SentenceTransformer
import torch

with open("knowledge.txt", "r", encoding="utf-8") as file:
  recent = file.read()
  # opens the text, saves as "file"
  # reads the text and saves as water_cycle_text variable



cleaned_text = recent.strip()
# cleaning up the text
chunks = cleaned_text.split("\n")
# seperating the text into one sentence pieces
cleaned_chunks = []
# creating an empty list to put the cleaned chunks in

for chunk in chunks:
  stripped_chunk = chunk.strip()
  if stripped_chunk:
    cleaned_chunks.append(stripped_chunk)
    # loop through chunks and add not empty chunks to cleaned_chunks list


model = SentenceTransformer('all-MiniLM-L6-v2')

chunk_embeddings = model.encode(cleaned_chunks, convert_to_tensor=True)
# encode the model, pass through my cleaned chunks and convert to vector embeddings (not arrays)


def get_top_chunks(query):
  # create my function taking query as parameter
  query_embedding = model.encode(query, convert_to_tensor=True)
  # encode query to vector embedding for comparison
  query_embedding_normalized = query_embedding / query_embedding.norm()
  # normalize my query to 1; allows for comparison of meaning
  chunk_embeddings_normalized = chunk_embeddings / chunk_embeddings.norm(dim=1, keepdim=True)
  # normailizing chunks for comparison of meaning

  similarities = torch.matmul(chunk_embeddings_normalized, query_embedding_normalized)

  # using matmul (matrix multiplication method) to compare query to chunks
  top_indices = torch.topk(similarities, k=3).indices

  # get the indices of the chunks that are most similar to query

  top_chunks = []

  for i in top_indices:
    chunk = chunks[i]
    # for each index number in top_indices, get back the text
    top_chunks.append(chunk)
    # values of each index number is added to top_chunks
  return top_chunks
    
client = InferenceClient(
    model='Qwen/Qwen2.5-72B-Instruct',
    #token=os.getenv('HF_TOKEN')
)

#client is where you can change the LLM model!
def respond(message,history):
    if not message.strip():
        return "Hello!"
    gift_ideas = get_top_chunks(message)
    messages = [{'role': 'system', 'content': f'You give really good gift ideas and are super helpful! You also tell me the price of each item. You also offer to give a card stem to write thoughful cards to give with the gift. The card stems you give are related to the type of celebration the person is buying a gift for. You offer to give them a card template for a longer card if they want and if they say yes give them a card template. Give me 5 gift ideas if I ask. Use the following database for gift ideas: {gift_ideas}'}]
    
    if history:
        messages.extend(history)
    
    messages.append({"role": "user", "content": message})

    response = ""
    for message in client.chat_completion(
        #max_tokens controls how many words your responses is 
        messages,
        max_tokens = 500,
        stream = True,
        
        #temperature = 0.8, #code a decimal between 0-2
        #top_p = .65 #code a decimal between 0-1
    ):
        token = message.choices[0].delta.content
        response += token
        yield response
    #print(response["choices"][0]["message"]["content"].strip())
    
    #yield response["choices"][0]["message"]["content"].strip()
    


#with gr.Blocks(theme='hmb/amethyst') as demo:
 #   with gr.Row(equal_height=True):
  #      with gr.Column(scale=10):
   #            """
    #            # 🎁 Introducing WrapIT!
     #           **WrapIT** helps users find personalized gift ideas and craft thoughtful card messages  
      #          by inputting details like the recipient's interests, celebration type, and budget.
       #         
        #        ✨ *All you have to do is wrap it.*
         #       """
          #  )
           # gr.ChatInterface(respond, type='messages')

  
    
    #chatbot = gr.Chatbot()
    #msg = gr.Textbox(placeholder="Say hi to WrapIT here!", label="Message")
    #send = gr.Button("Send")

    #msg.submit(respond, [msg, chatbot], [msg, chatbot])
    #send.click(respond, [msg, chatbot], [msg, chatbot])
    
with gr.Blocks(theme='hmb/amethyst') as demo:
    # Top image
    gr.Image(value="wrap_it_top_image.png", show_label=False, elem_id="top-image")

    # Title and description
    gr.Markdown("## 🎁 Introducing WrapIT!")
    gr.Markdown("**WrapIT** helps users find personalized gift ideas and craft thoughtful card messages by inputting details like the recipient's interests, celebration type, and budget ✨ *All you have to do is wrap it.*")


    
    # Chat interface
    gr.ChatInterface(
    fn=respond,
    examples=["Best birthday gift?", "Romantic anniversary idea?", "Budget-friendly gifts?"]
)


demo.launch()