Update app.py
Browse files
app.py
CHANGED
@@ -1,138 +1,63 @@
|
|
1 |
-
import os
|
2 |
-
from threading import Thread
|
3 |
-
from typing import Iterator
|
4 |
-
|
5 |
import gradio as gr
|
6 |
-
import
|
7 |
-
import torch
|
8 |
-
from transformers import AutoModelForCausalLM, GemmaTokenizerFast, TextIteratorStreamer
|
9 |
|
10 |
-
DESCRIPTION = """\
|
11 |
-
# Gemma 2 2B IT
|
12 |
-
Gemma 2 is Google's latest iteration of open LLMs.
|
13 |
-
This is a demo of [`google/gemma-2-2b-it`](https://huggingface.co/google/gemma-2-2b-it), fine-tuned for instruction following.
|
14 |
-
For more details, please check [our post](https://huggingface.co/blog/gemma2).
|
15 |
-
👉 Looking for a larger and more powerful version? Try the 27B version in [HuggingChat](https://huggingface.co/chat/models/google/gemma-2-27b-it) and the 9B version in [this Space](https://huggingface.co/spaces/huggingface-projects/gemma-2-9b-it).
|
16 |
"""
|
|
|
|
|
|
|
17 |
|
18 |
-
MAX_MAX_NEW_TOKENS = 2048
|
19 |
-
DEFAULT_MAX_NEW_TOKENS = 1024
|
20 |
-
MAX_INPUT_TOKEN_LENGTH = int(os.getenv("MAX_INPUT_TOKEN_LENGTH", "4096"))
|
21 |
-
|
22 |
-
device = torch.device("cuda:0" if torch.cuda.is_available() else "cpu")
|
23 |
|
24 |
-
|
25 |
-
|
26 |
-
|
27 |
-
|
28 |
-
|
29 |
-
|
30 |
-
|
31 |
-
|
32 |
-
|
33 |
|
|
|
|
|
|
|
|
|
|
|
34 |
|
35 |
-
|
36 |
-
def generate(
|
37 |
-
message: str,
|
38 |
-
chat_history: list[tuple[str, str]],
|
39 |
-
max_new_tokens: int = 1024,
|
40 |
-
temperature: float = 0.6,
|
41 |
-
top_p: float = 0.9,
|
42 |
-
top_k: int = 50,
|
43 |
-
repetition_penalty: float = 1.2,
|
44 |
-
) -> Iterator[str]:
|
45 |
-
conversation = []
|
46 |
-
for user, assistant in chat_history:
|
47 |
-
conversation.extend(
|
48 |
-
[
|
49 |
-
{"role": "user", "content": user},
|
50 |
-
{"role": "assistant", "content": assistant},
|
51 |
-
]
|
52 |
-
)
|
53 |
-
conversation.append({"role": "user", "content": message})
|
54 |
|
55 |
-
|
56 |
-
if input_ids.shape[1] > MAX_INPUT_TOKEN_LENGTH:
|
57 |
-
input_ids = input_ids[:, -MAX_INPUT_TOKEN_LENGTH:]
|
58 |
-
gr.Warning(f"Trimmed input from conversation as it was longer than {MAX_INPUT_TOKEN_LENGTH} tokens.")
|
59 |
-
input_ids = input_ids.to(model.device)
|
60 |
|
61 |
-
|
62 |
-
|
63 |
-
|
64 |
-
|
65 |
-
max_new_tokens=max_new_tokens,
|
66 |
-
do_sample=True,
|
67 |
-
top_p=top_p,
|
68 |
-
top_k=top_k,
|
69 |
temperature=temperature,
|
70 |
-
|
71 |
-
|
72 |
-
|
73 |
-
t = Thread(target=model.generate, kwargs=generate_kwargs)
|
74 |
-
t.start()
|
75 |
-
|
76 |
-
outputs = []
|
77 |
-
for text in streamer:
|
78 |
-
outputs.append(text)
|
79 |
-
yield "".join(outputs)
|
80 |
|
|
|
|
|
81 |
|
82 |
-
|
83 |
-
|
|
|
|
|
|
|
84 |
additional_inputs=[
|
|
|
|
|
|
|
85 |
gr.Slider(
|
86 |
-
label="Max new tokens",
|
87 |
-
minimum=1,
|
88 |
-
maximum=MAX_MAX_NEW_TOKENS,
|
89 |
-
step=1,
|
90 |
-
value=DEFAULT_MAX_NEW_TOKENS,
|
91 |
-
),
|
92 |
-
gr.Slider(
|
93 |
-
label="Temperature",
|
94 |
minimum=0.1,
|
95 |
-
maximum=4.0,
|
96 |
-
step=0.1,
|
97 |
-
value=0.6,
|
98 |
-
),
|
99 |
-
gr.Slider(
|
100 |
-
label="Top-p (nucleus sampling)",
|
101 |
-
minimum=0.05,
|
102 |
maximum=1.0,
|
|
|
103 |
step=0.05,
|
104 |
-
|
105 |
-
),
|
106 |
-
gr.Slider(
|
107 |
-
label="Top-k",
|
108 |
-
minimum=1,
|
109 |
-
maximum=1000,
|
110 |
-
step=1,
|
111 |
-
value=50,
|
112 |
-
),
|
113 |
-
gr.Slider(
|
114 |
-
label="Repetition penalty",
|
115 |
-
minimum=1.0,
|
116 |
-
maximum=2.0,
|
117 |
-
step=0.05,
|
118 |
-
value=1.2,
|
119 |
),
|
120 |
],
|
121 |
-
stop_btn=None,
|
122 |
-
examples=[
|
123 |
-
["Hello there! How are you doing?"],
|
124 |
-
["Can you explain briefly to me what is the Python programming language?"],
|
125 |
-
["Explain the plot of Cinderella in a sentence."],
|
126 |
-
["How many hours does it take a man to eat a Helicopter?"],
|
127 |
-
["Write a 100-word article on 'Benefits of Open-Source in AI research'"],
|
128 |
-
],
|
129 |
-
cache_examples=False,
|
130 |
)
|
131 |
|
132 |
-
with gr.Blocks(css="style.css", fill_height=True) as demo:
|
133 |
-
gr.Markdown(DESCRIPTION)
|
134 |
-
gr.DuplicateButton(value="Duplicate Space for private use", elem_id="duplicate-button")
|
135 |
-
chat_interface.render()
|
136 |
|
137 |
if __name__ == "__main__":
|
138 |
-
demo.
|
|
|
|
|
|
|
|
|
|
|
1 |
import gradio as gr
|
2 |
+
from huggingface_hub import InferenceClient
|
|
|
|
|
3 |
|
|
|
|
|
|
|
|
|
|
|
|
|
4 |
"""
|
5 |
+
For more information on `huggingface_hub` Inference API support, please check the docs: https://huggingface.co/docs/huggingface_hub/v0.22.2/en/guides/inference
|
6 |
+
"""
|
7 |
+
client = InferenceClient("google/gemma-2-2b-it")
|
8 |
|
|
|
|
|
|
|
|
|
|
|
9 |
|
10 |
+
def respond(
|
11 |
+
message,
|
12 |
+
history: list[tuple[str, str]],
|
13 |
+
system_message,
|
14 |
+
max_tokens,
|
15 |
+
temperature,
|
16 |
+
top_p,
|
17 |
+
):
|
18 |
+
messages = [{"role": "system", "content": system_message}]
|
19 |
|
20 |
+
for val in history:
|
21 |
+
if val[0]:
|
22 |
+
messages.append({"role": "user", "content": val[0]})
|
23 |
+
if val[1]:
|
24 |
+
messages.append({"role": "assistant", "content": val[1]})
|
25 |
|
26 |
+
messages.append({"role": "user", "content": message})
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
27 |
|
28 |
+
response = ""
|
|
|
|
|
|
|
|
|
29 |
|
30 |
+
for message in client.chat_completion(
|
31 |
+
messages,
|
32 |
+
max_tokens=max_tokens,
|
33 |
+
stream=True,
|
|
|
|
|
|
|
|
|
34 |
temperature=temperature,
|
35 |
+
top_p=top_p,
|
36 |
+
):
|
37 |
+
token = message.choices[0].delta.content
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
38 |
|
39 |
+
response += token
|
40 |
+
yield response
|
41 |
|
42 |
+
"""
|
43 |
+
For information on how to customize the ChatInterface, peruse the gradio docs: https://www.gradio.app/docs/chatinterface
|
44 |
+
"""
|
45 |
+
demo = gr.ChatInterface(
|
46 |
+
respond,
|
47 |
additional_inputs=[
|
48 |
+
gr.Textbox(value="You are a friendly Chatbot.", label="System message"),
|
49 |
+
gr.Slider(minimum=1, maximum=2048, value=512, step=1, label="Max new tokens"),
|
50 |
+
gr.Slider(minimum=0.1, maximum=4.0, value=0.7, step=0.1, label="Temperature"),
|
51 |
gr.Slider(
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
52 |
minimum=0.1,
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
53 |
maximum=1.0,
|
54 |
+
value=0.95,
|
55 |
step=0.05,
|
56 |
+
label="Top-p (nucleus sampling)",
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
57 |
),
|
58 |
],
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
59 |
)
|
60 |
|
|
|
|
|
|
|
|
|
61 |
|
62 |
if __name__ == "__main__":
|
63 |
+
demo.launch()
|