Spaces:
Sleeping
Sleeping
Create app.py
Browse files
app.py
ADDED
@@ -0,0 +1,169 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
import os
|
2 |
+
from threading import Thread
|
3 |
+
from typing import Iterator, List, Tuple
|
4 |
+
|
5 |
+
import gradio as gr
|
6 |
+
import spaces
|
7 |
+
import torch
|
8 |
+
from transformers import AutoModelForCausalLM, AutoTokenizer, TextIteratorStreamer
|
9 |
+
|
10 |
+
DESCRIPTION = """\
|
11 |
+
# Zero GPU Model Comparison Arena
|
12 |
+
Compare two language models using Hugging Face's Zero GPU initiative.
|
13 |
+
Select two different models from the dropdowns and see how they perform on the same input.
|
14 |
+
"""
|
15 |
+
|
16 |
+
MAX_MAX_NEW_TOKENS = 1024
|
17 |
+
DEFAULT_MAX_NEW_TOKENS = 256
|
18 |
+
MAX_INPUT_TOKEN_LENGTH = int(os.getenv("MAX_INPUT_TOKEN_LENGTH", "4096"))
|
19 |
+
|
20 |
+
device = torch.device("cuda:0" if torch.cuda.is_available() else "cpu")
|
21 |
+
|
22 |
+
MODEL_OPTIONS = [
|
23 |
+
"google/gemma-2b-it",
|
24 |
+
"mistralai/Mistral-7B-Instruct-v0.2",
|
25 |
+
"meta-llama/Llama-2-7b-chat-hf",
|
26 |
+
"tiiuae/falcon-7b-instruct"
|
27 |
+
]
|
28 |
+
|
29 |
+
models = {}
|
30 |
+
tokenizers = {}
|
31 |
+
|
32 |
+
for model_id in MODEL_OPTIONS:
|
33 |
+
tokenizers[model_id] = AutoTokenizer.from_pretrained(model_id)
|
34 |
+
models[model_id] = AutoModelForCausalLM.from_pretrained(
|
35 |
+
model_id,
|
36 |
+
device_map="auto",
|
37 |
+
load_in_8bit=True,
|
38 |
+
)
|
39 |
+
models[model_id].eval()
|
40 |
+
|
41 |
+
@spaces.GPU(duration=90)
|
42 |
+
def generate(
|
43 |
+
model_id: str,
|
44 |
+
message: str,
|
45 |
+
chat_history: List[Tuple[str, str]],
|
46 |
+
max_new_tokens: int = DEFAULT_MAX_NEW_TOKENS,
|
47 |
+
temperature: float = 0.7,
|
48 |
+
top_p: float = 0.95,
|
49 |
+
) -> Iterator[str]:
|
50 |
+
model = models[model_id]
|
51 |
+
tokenizer = tokenizers[model_id]
|
52 |
+
|
53 |
+
conversation = []
|
54 |
+
for user, assistant in chat_history:
|
55 |
+
conversation.extend([
|
56 |
+
{"role": "user", "content": user},
|
57 |
+
{"role": "assistant", "content": assistant},
|
58 |
+
])
|
59 |
+
conversation.append({"role": "user", "content": message})
|
60 |
+
|
61 |
+
input_ids = tokenizer.apply_chat_template(conversation, return_tensors="pt")
|
62 |
+
if input_ids.shape[1] > MAX_INPUT_TOKEN_LENGTH:
|
63 |
+
input_ids = input_ids[:, -MAX_INPUT_TOKEN_LENGTH:]
|
64 |
+
gr.Warning(f"Trimmed input from conversation as it was longer than {MAX_INPUT_TOKEN_LENGTH} tokens.")
|
65 |
+
input_ids = input_ids.to(model.device)
|
66 |
+
|
67 |
+
streamer = TextIteratorStreamer(tokenizer, timeout=10.0, skip_prompt=True, skip_special_tokens=True)
|
68 |
+
generate_kwargs = dict(
|
69 |
+
input_ids=input_ids,
|
70 |
+
streamer=streamer,
|
71 |
+
max_new_tokens=max_new_tokens,
|
72 |
+
do_sample=True,
|
73 |
+
top_p=top_p,
|
74 |
+
temperature=temperature,
|
75 |
+
num_beams=1,
|
76 |
+
)
|
77 |
+
t = Thread(target=model.generate, kwargs=generate_kwargs)
|
78 |
+
t.start()
|
79 |
+
|
80 |
+
outputs = []
|
81 |
+
for text in streamer:
|
82 |
+
outputs.append(text)
|
83 |
+
yield "".join(outputs)
|
84 |
+
|
85 |
+
def compare_models(
|
86 |
+
model1_name: str,
|
87 |
+
model2_name: str,
|
88 |
+
message: str,
|
89 |
+
chat_history1: List[Tuple[str, str]],
|
90 |
+
chat_history2: List[Tuple[str, str]],
|
91 |
+
max_new_tokens: int,
|
92 |
+
temperature: float,
|
93 |
+
top_p: float,
|
94 |
+
) -> Tuple[str, str, List[Tuple[str, str]], List[Tuple[str, str]]]:
|
95 |
+
if model1_name == model2_name:
|
96 |
+
return "Error: Please select two different models.", "Error: Please select two different models.", chat_history1, chat_history2
|
97 |
+
|
98 |
+
output1 = "".join(list(generate(model1_name, message, chat_history1, max_new_tokens, temperature, top_p)))
|
99 |
+
output2 = "".join(list(generate(model2_name, message, chat_history2, max_new_tokens, temperature, top_p)))
|
100 |
+
|
101 |
+
chat_history1.append((message, output1))
|
102 |
+
chat_history2.append((message, output2))
|
103 |
+
|
104 |
+
log_results(model1_name, model2_name, message, output1, output2)
|
105 |
+
|
106 |
+
return output1, output2, chat_history1, chat_history2
|
107 |
+
|
108 |
+
def log_results(model1_name: str, model2_name: str, question: str, answer1: str, answer2: str, winner: str = None):
|
109 |
+
log_data = {
|
110 |
+
"question": question,
|
111 |
+
"model1": {"name": model1_name, "answer": answer1},
|
112 |
+
"model2": {"name": model2_name, "answer": answer2},
|
113 |
+
"winner": winner
|
114 |
+
}
|
115 |
+
|
116 |
+
# Here you would implement the actual logging logic, e.g., sending to a server or writing to a file
|
117 |
+
print("Logged:", log_data)
|
118 |
+
|
119 |
+
def vote_better(model1_name, model2_name, question, answer1, answer2, choice):
|
120 |
+
winner = model1_name if choice == "Model 1" else model2_name
|
121 |
+
log_results(model1_name, model2_name, question, answer1, answer2, winner)
|
122 |
+
return f"You voted that {winner} performs better. This has been logged."
|
123 |
+
|
124 |
+
with gr.Blocks(css="style.css") as demo:
|
125 |
+
gr.Markdown(DESCRIPTION)
|
126 |
+
|
127 |
+
with gr.Row():
|
128 |
+
with gr.Column():
|
129 |
+
model1_dropdown = gr.Dropdown(choices=MODEL_OPTIONS, label="Model 1", value=MODEL_OPTIONS[0])
|
130 |
+
chatbot1 = gr.Chatbot(label="Model 1 Output")
|
131 |
+
with gr.Column():
|
132 |
+
model2_dropdown = gr.Dropdown(choices=MODEL_OPTIONS, label="Model 2", value=MODEL_OPTIONS[1])
|
133 |
+
chatbot2 = gr.Chatbot(label="Model 2 Output")
|
134 |
+
|
135 |
+
text_input = gr.Textbox(label="Input Text", lines=3)
|
136 |
+
|
137 |
+
with gr.Row():
|
138 |
+
max_new_tokens = gr.Slider(label="Max new tokens", minimum=1, maximum=MAX_MAX_NEW_TOKENS, value=DEFAULT_MAX_NEW_TOKENS)
|
139 |
+
temperature = gr.Slider(label="Temperature", minimum=0.1, maximum=2.0, value=0.7)
|
140 |
+
top_p = gr.Slider(label="Top-p (nucleus sampling)", minimum=0.05, maximum=1.0, value=0.95)
|
141 |
+
|
142 |
+
compare_btn = gr.Button("Compare Models")
|
143 |
+
|
144 |
+
with gr.Row():
|
145 |
+
better1_btn = gr.Button("Model 1 is Better")
|
146 |
+
better2_btn = gr.Button("Model 2 is Better")
|
147 |
+
|
148 |
+
vote_output = gr.Textbox(label="Voting Result")
|
149 |
+
|
150 |
+
compare_btn.click(
|
151 |
+
compare_models,
|
152 |
+
inputs=[model1_dropdown, model2_dropdown, text_input, chatbot1, chatbot2, max_new_tokens, temperature, top_p],
|
153 |
+
outputs=[chatbot1, chatbot2, chatbot1, chatbot2]
|
154 |
+
)
|
155 |
+
|
156 |
+
better1_btn.click(
|
157 |
+
vote_better,
|
158 |
+
inputs=[model1_dropdown, model2_dropdown, text_input, chatbot1, chatbot2, gr.Textbox(value="Model 1", visible=False)],
|
159 |
+
outputs=[vote_output]
|
160 |
+
)
|
161 |
+
|
162 |
+
better2_btn.click(
|
163 |
+
vote_better,
|
164 |
+
inputs=[model1_dropdown, model2_dropdown, text_input, chatbot1, chatbot2, gr.Textbox(value="Model 2", visible=False)],
|
165 |
+
outputs=[vote_output]
|
166 |
+
)
|
167 |
+
|
168 |
+
if __name__ == "__main__":
|
169 |
+
demo.queue(max_size=10).launch()
|