arad1367 commited on
Commit
d2a125e
β€’
1 Parent(s): b93f63b

Upload 2 files

Browse files
Files changed (2) hide show
  1. app.py +169 -0
  2. requirements.txt +6 -0
app.py ADDED
@@ -0,0 +1,169 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ # Starting with transformers >= 4.43.0 onward.and
2
+ # you can run conversational inference using the Transformers pipeline abstraction or by leveraging the Auto classes with the generate() function.
3
+ import os
4
+ import time
5
+ import spaces
6
+ import torch
7
+ from transformers import AutoModelForCausalLM, AutoTokenizer, TextIteratorStreamer, BitsAndBytesConfig
8
+ import gradio as gr
9
+ from threading import Thread
10
+
11
+ MODEL_LIST = ["meta-llama/Meta-Llama-3.1-8B-Instruct"]
12
+ HF_TOKEN = os.environ.get("HF_TOKEN", None)
13
+ MODEL = os.environ.get("MODEL_ID")
14
+
15
+ TITLE = "<h1><center>Meta-Llama3.1-8B Chatbot</center></h1>"
16
+
17
+ PLACEHOLDER = """
18
+ <center>
19
+ <p>Hi! I'm your assistant. Feel free to ask your questions</p>
20
+ </center>
21
+ """
22
+
23
+
24
+ CSS = """
25
+ .duplicate-button {
26
+ margin: auto !important;
27
+ color: white !important;
28
+ background: black !important;
29
+ border-radius: 100vh !important;
30
+ }
31
+ h3 {
32
+ text-align: center;
33
+ }
34
+ """
35
+
36
+ device = "cuda" # for GPU usage or "cpu" for CPU usage
37
+
38
+ quantization_config = BitsAndBytesConfig(
39
+ load_in_4bit=True,
40
+ bnb_4bit_compute_dtype=torch.bfloat16,
41
+ bnb_4bit_use_double_quant=True,
42
+ bnb_4bit_quant_type= "nf4")
43
+
44
+ tokenizer = AutoTokenizer.from_pretrained(MODEL)
45
+ model = AutoModelForCausalLM.from_pretrained(
46
+ MODEL,
47
+ torch_dtype=torch.bfloat16,
48
+ device_map="auto",
49
+ quantization_config=quantization_config)
50
+
51
+ @spaces.GPU()
52
+ def stream_chat(
53
+ message: str,
54
+ history: list,
55
+ system_prompt: str,
56
+ temperature: float = 0.8,
57
+ max_new_tokens: int = 1024,
58
+ top_p: float = 1.0,
59
+ top_k: int = 20,
60
+ penalty: float = 1.2,
61
+ ):
62
+ print(f'message: {message}')
63
+ print(f'history: {history}')
64
+
65
+ conversation = [
66
+ {"role": "system", "content": system_prompt}
67
+ ]
68
+ for prompt, answer in history:
69
+ conversation.extend([
70
+ {"role": "user", "content": prompt},
71
+ {"role": "assistant", "content": answer},
72
+ ])
73
+
74
+ conversation.append({"role": "user", "content": message})
75
+
76
+ input_ids = tokenizer.apply_chat_template(conversation, add_generation_prompt=True, return_tensors="pt").to(model.device)
77
+
78
+ streamer = TextIteratorStreamer(tokenizer, timeout=60.0, skip_prompt=True, skip_special_tokens=True)
79
+
80
+ generate_kwargs = dict(
81
+ input_ids=input_ids,
82
+ max_new_tokens = max_new_tokens,
83
+ do_sample = False if temperature == 0 else True,
84
+ top_p = top_p,
85
+ top_k = top_k,
86
+ temperature = temperature,
87
+ eos_token_id=[128001,128008,128009],
88
+ streamer=streamer,
89
+ )
90
+
91
+ with torch.no_grad():
92
+ thread = Thread(target=model.generate, kwargs=generate_kwargs)
93
+ thread.start()
94
+
95
+ buffer = ""
96
+ for new_text in streamer:
97
+ buffer += new_text
98
+ yield buffer
99
+
100
+
101
+ chatbot = gr.Chatbot(height=600, placeholder=PLACEHOLDER)
102
+
103
+ with gr.Blocks(css=CSS, theme="small_and_pretty") as demo:
104
+ gr.HTML(TITLE)
105
+ gr.DuplicateButton(value="Duplicate Space for private use", elem_classes="duplicate-button")
106
+ gr.ChatInterface(
107
+ fn=stream_chat,
108
+ chatbot=chatbot,
109
+ fill_height=True,
110
+ additional_inputs_accordion=gr.Accordion(label="βš™οΈ Parameters", open=False, render=False),
111
+ additional_inputs=[
112
+ gr.Textbox(
113
+ value="You are a helpful assistant",
114
+ label="System Prompt",
115
+ render=False,
116
+ ),
117
+ gr.Slider(
118
+ minimum=0,
119
+ maximum=1,
120
+ step=0.1,
121
+ value=0.8,
122
+ label="Temperature",
123
+ render=False,
124
+ ),
125
+ gr.Slider(
126
+ minimum=128,
127
+ maximum=8192,
128
+ step=1,
129
+ value=1024,
130
+ label="Max new tokens",
131
+ render=False,
132
+ ),
133
+ gr.Slider(
134
+ minimum=0.0,
135
+ maximum=1.0,
136
+ step=0.1,
137
+ value=1.0,
138
+ label="top_p",
139
+ render=False,
140
+ ),
141
+ gr.Slider(
142
+ minimum=1,
143
+ maximum=20,
144
+ step=1,
145
+ value=20,
146
+ label="top_k",
147
+ render=False,
148
+ ),
149
+ gr.Slider(
150
+ minimum=0.0,
151
+ maximum=2.0,
152
+ step=0.1,
153
+ value=1.2,
154
+ label="Repetition penalty",
155
+ render=False,
156
+ ),
157
+ ],
158
+ examples=[
159
+ ["How to make a self-driving car?"],
160
+ ["Give me creative idea to establish a startup"],
161
+ ["How can I improve my programming skills?"],
162
+ ["Show me a code snippet of a website's sticky header in CSS and JavaScript."],
163
+ ],
164
+ cache_examples=False,
165
+ )
166
+
167
+
168
+ if __name__ == "__main__":
169
+ demo.launch()
requirements.txt ADDED
@@ -0,0 +1,6 @@
 
 
 
 
 
 
 
1
+ accelerate
2
+ bitsandbytes
3
+ torch
4
+ transformers==4.43.1
5
+ einops
6
+ sentencepiece