rasyosef commited on
Commit
c44aa23
1 Parent(s): 31662de

Create app.py

Browse files
Files changed (1) hide show
  1. app.py +76 -0
app.py ADDED
@@ -0,0 +1,76 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ import gradio as gr
2
+
3
+ import torch
4
+ from transformers import AutoTokenizer, AutoModelForCausalLM, TextIteratorStreamer, pipeline
5
+ from threading import Thread
6
+
7
+ # The huggingface model id for phi-1_5 instruct model
8
+ checkpoint = "rasyosef/Phi-1_5-Instruct-v0.1"
9
+
10
+ # Download and load model and tokenizer
11
+ tokenizer = AutoTokenizer.from_pretrained(checkpoint)
12
+ model = AutoModelForCausalLM.from_pretrained(checkpoint, torch_dtype=torch.float32, device_map="auto")
13
+
14
+ # Text generation pipeline
15
+ phi1_5 = pipeline(
16
+ "text-generation",
17
+ tokenizer=tokenizer,
18
+ model=model,
19
+ pad_token_id=tokenizer.eos_token_id,
20
+ eos_token_id=[tokenizer.eos_token_id],
21
+ device_map="cpu"
22
+ )
23
+
24
+ # Function that accepts a prompt and generates text using the phi2 pipeline
25
+ def generate(message, chat_history, max_new_tokens):
26
+
27
+ history = [
28
+ {"role": "system", "content": "You are Phi, a helpful AI assistant made by Microsoft and RasYosef. User will you give you a task. Your goal is to complete the task as faithfully as you can."}
29
+ ]
30
+
31
+ for sent, received in chat_history:
32
+ history.append({"role": "user", "content": sent})
33
+ history.append({"role": "assistant", "content": received})
34
+
35
+ history.append({"role": "user", "content": message})
36
+ #print(history)
37
+
38
+ if len(tokenizer.apply_chat_template(history)) > 512:
39
+ yield "chat history is too long"
40
+ else:
41
+ # Streamer
42
+ streamer = TextIteratorStreamer(tokenizer=tokenizer, skip_prompt=True, skip_special_tokens=True, timeout=300.0)
43
+ thread = Thread(target=phi1_5, kwargs={"text_inputs":history, "max_new_tokens":max_new_tokens, "streamer":streamer})
44
+ thread.start()
45
+
46
+ generated_text = ""
47
+ for word in streamer:
48
+ generated_text += word
49
+ response = generated_text.strip()
50
+
51
+ yield response
52
+
53
+ # Chat interface with gradio
54
+ with gr.Blocks() as demo:
55
+ gr.Markdown("""
56
+ # Phi-1_5 Chatbot Demo
57
+ This chatbot was created using a finetuned version of Microsoft's 1.4 billion parameter Phi 1.5 transformer model, [Phi-1_5-Instruct-v0.1](https://huggingface.co/rasyosef/Phi-1_5-Instruct-v0.1).
58
+ """)
59
+
60
+ tokens_slider = gr.Slider(8, 256, value=64, label="Maximum new tokens", info="A larger `max_new_tokens` parameter value gives you longer text responses but at the cost of a slower response time.")
61
+
62
+ chatbot = gr.ChatInterface(
63
+ chatbot=gr.Chatbot(height=400),
64
+ fn=generate,
65
+ additional_inputs=[tokens_slider],
66
+ stop_btn=None,
67
+ examples=[
68
+ ["Translate the word 'cat' to German."],
69
+ ["Recommend me three animated movies."],
70
+ ["Solve this quadratic equation.\n\nx^2 + 3x + 2 = 0\nwhat is the value of x?"],
71
+ ["Implement Euclid's GCD Algorithm in python"],
72
+ ["Molly and Abigail want to attend a beauty and modeling contest. They both want to buy new pairs of shoes and dresses. Molly buys a pair of shoes which costs $40 and a dress which costs $160. How much should Abigail budget if she wants to spend half of what Molly spent on the pair of shoes and dress?"],
73
+ ]
74
+ )
75
+
76
+ demo.queue().launch(debug=True)