Spaces:
Runtime error
Runtime error
Update app.py
Browse files
app.py
CHANGED
@@ -6,9 +6,14 @@ import gradio as gr
|
|
6 |
import spaces
|
7 |
import torch
|
8 |
from transformers import AutoModelForCausalLM, AutoTokenizer, TextIteratorStreamer
|
9 |
-
|
10 |
DESCRIPTION = """\
|
11 |
-
#
|
|
|
|
|
|
|
|
|
|
|
|
|
12 |
"""
|
13 |
|
14 |
MAX_MAX_NEW_TOKENS = 2048
|
@@ -26,31 +31,19 @@ model = AutoModelForCausalLM.from_pretrained(
|
|
26 |
)
|
27 |
model.eval()
|
28 |
|
29 |
-
|
30 |
@spaces.GPU(duration=90)
|
31 |
def generate(
|
32 |
message: str,
|
33 |
-
chat_history: list[tuple[str, str]],
|
34 |
max_new_tokens: int = 1024,
|
35 |
temperature: float = 0.6,
|
36 |
top_p: float = 0.9,
|
37 |
top_k: int = 50,
|
38 |
repetition_penalty: float = 1.2,
|
39 |
) -> Iterator[str]:
|
40 |
-
|
41 |
-
for user, assistant in chat_history:
|
42 |
-
conversation.extend(
|
43 |
-
[
|
44 |
-
{"role": "user", "content": user},
|
45 |
-
{"role": "assistant", "content": assistant},
|
46 |
-
]
|
47 |
-
)
|
48 |
-
conversation.append({"role": "user", "content": message})
|
49 |
-
|
50 |
-
input_ids = tokenizer.apply_chat_template(conversation, add_generation_prompt=True, return_tensors="pt")
|
51 |
if input_ids.shape[1] > MAX_INPUT_TOKEN_LENGTH:
|
52 |
input_ids = input_ids[:, -MAX_INPUT_TOKEN_LENGTH:]
|
53 |
-
gr.Warning(f"Trimmed input
|
54 |
input_ids = input_ids.to(model.device)
|
55 |
|
56 |
streamer = TextIteratorStreamer(tokenizer, timeout=20.0, skip_prompt=True, skip_special_tokens=True)
|
@@ -71,63 +64,86 @@ def generate(
|
|
71 |
outputs = []
|
72 |
for text in streamer:
|
73 |
outputs.append(text)
|
74 |
-
yield "".join(outputs)
|
75 |
-
|
76 |
-
|
77 |
-
chat_interface = gr.ChatInterface(
|
78 |
-
fn=generate,
|
79 |
-
additional_inputs=[
|
80 |
-
gr.Slider(
|
81 |
-
label="Max new tokens",
|
82 |
-
minimum=1,
|
83 |
-
maximum=MAX_MAX_NEW_TOKENS,
|
84 |
-
step=1,
|
85 |
-
value=DEFAULT_MAX_NEW_TOKENS,
|
86 |
-
),
|
87 |
-
gr.Slider(
|
88 |
-
label="Temperature",
|
89 |
-
minimum=0.1,
|
90 |
-
maximum=4.0,
|
91 |
-
step=0.1,
|
92 |
-
value=0.6,
|
93 |
-
),
|
94 |
-
gr.Slider(
|
95 |
-
label="Top-p (nucleus sampling)",
|
96 |
-
minimum=0.05,
|
97 |
-
maximum=1.0,
|
98 |
-
step=0.05,
|
99 |
-
value=0.9,
|
100 |
-
),
|
101 |
-
gr.Slider(
|
102 |
-
label="Top-k",
|
103 |
-
minimum=1,
|
104 |
-
maximum=1000,
|
105 |
-
step=1,
|
106 |
-
value=50,
|
107 |
-
),
|
108 |
-
gr.Slider(
|
109 |
-
label="Repetition penalty",
|
110 |
-
minimum=1.0,
|
111 |
-
maximum=2.0,
|
112 |
-
step=0.05,
|
113 |
-
value=1.2,
|
114 |
-
),
|
115 |
-
],
|
116 |
-
stop_btn=None,
|
117 |
-
examples=[
|
118 |
-
["Hello there! How are you doing?"],
|
119 |
-
["Can you explain briefly to me what is the Python programming language?"],
|
120 |
-
["Explain the plot of Cinderella in a sentence."],
|
121 |
-
["How many hours does it take a man to eat a Helicopter?"],
|
122 |
-
["Write a 100-word article on 'Benefits of Open-Source in AI research'"],
|
123 |
-
],
|
124 |
-
cache_examples=False,
|
125 |
-
)
|
126 |
|
127 |
with gr.Blocks(css="style.css", fill_height=True) as demo:
|
128 |
gr.Markdown(DESCRIPTION)
|
129 |
gr.DuplicateButton(value="Duplicate Space for private use", elem_id="duplicate-button")
|
130 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
131 |
|
132 |
if __name__ == "__main__":
|
133 |
demo.queue(max_size=20).launch()
|
|
|
6 |
import spaces
|
7 |
import torch
|
8 |
from transformers import AutoModelForCausalLM, AutoTokenizer, TextIteratorStreamer
|
|
|
9 |
DESCRIPTION = """\
|
10 |
+
# Qwen 0.5B Text Completion
|
11 |
+
|
12 |
+
This is a demo of [`Qwen/Qwen2-0.5B-Instruct`](https://huggingface.co/Qwen/Qwen2-0.5B-Instruct), a lightweight language model fine-tuned for instruction following.
|
13 |
+
|
14 |
+
This space allows you to input text and have the AI complete it. Simply type your text in the input box, click "Complete", and watch as the AI generates a continuation of your text.
|
15 |
+
|
16 |
+
You can adjust various parameters such as temperature and top-p sampling to control the generation process.
|
17 |
"""
|
18 |
|
19 |
MAX_MAX_NEW_TOKENS = 2048
|
|
|
31 |
)
|
32 |
model.eval()
|
33 |
|
|
|
34 |
@spaces.GPU(duration=90)
|
35 |
def generate(
|
36 |
message: str,
|
|
|
37 |
max_new_tokens: int = 1024,
|
38 |
temperature: float = 0.6,
|
39 |
top_p: float = 0.9,
|
40 |
top_k: int = 50,
|
41 |
repetition_penalty: float = 1.2,
|
42 |
) -> Iterator[str]:
|
43 |
+
input_ids = tokenizer.encode(message, return_tensors="pt")
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
44 |
if input_ids.shape[1] > MAX_INPUT_TOKEN_LENGTH:
|
45 |
input_ids = input_ids[:, -MAX_INPUT_TOKEN_LENGTH:]
|
46 |
+
gr.Warning(f"Trimmed input as it was longer than {MAX_INPUT_TOKEN_LENGTH} tokens.")
|
47 |
input_ids = input_ids.to(model.device)
|
48 |
|
49 |
streamer = TextIteratorStreamer(tokenizer, timeout=20.0, skip_prompt=True, skip_special_tokens=True)
|
|
|
64 |
outputs = []
|
65 |
for text in streamer:
|
66 |
outputs.append(text)
|
67 |
+
yield message + "".join(outputs)
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
68 |
|
69 |
with gr.Blocks(css="style.css", fill_height=True) as demo:
|
70 |
gr.Markdown(DESCRIPTION)
|
71 |
gr.DuplicateButton(value="Duplicate Space for private use", elem_id="duplicate-button")
|
72 |
+
|
73 |
+
with gr.Row():
|
74 |
+
with gr.Column(scale=4):
|
75 |
+
input_box = gr.Textbox(
|
76 |
+
label="Enter your text",
|
77 |
+
placeholder="Type your message here...",
|
78 |
+
lines=5
|
79 |
+
)
|
80 |
+
output_box = gr.Textbox(
|
81 |
+
label="Completed text",
|
82 |
+
lines=10,
|
83 |
+
interactive=False
|
84 |
+
)
|
85 |
+
with gr.Column(scale=1):
|
86 |
+
max_new_tokens = gr.Slider(
|
87 |
+
label="Max new tokens",
|
88 |
+
minimum=1,
|
89 |
+
maximum=MAX_MAX_NEW_TOKENS,
|
90 |
+
step=1,
|
91 |
+
value=DEFAULT_MAX_NEW_TOKENS,
|
92 |
+
)
|
93 |
+
temperature = gr.Slider(
|
94 |
+
label="Temperature",
|
95 |
+
minimum=0.1,
|
96 |
+
maximum=4.0,
|
97 |
+
step=0.1,
|
98 |
+
value=0.6,
|
99 |
+
)
|
100 |
+
top_p = gr.Slider(
|
101 |
+
label="Top-p (nucleus sampling)",
|
102 |
+
minimum=0.05,
|
103 |
+
maximum=1.0,
|
104 |
+
step=0.05,
|
105 |
+
value=0.9,
|
106 |
+
)
|
107 |
+
top_k = gr.Slider(
|
108 |
+
label="Top-k",
|
109 |
+
minimum=1,
|
110 |
+
maximum=1000,
|
111 |
+
step=1,
|
112 |
+
value=50,
|
113 |
+
)
|
114 |
+
repetition_penalty = gr.Slider(
|
115 |
+
label="Repetition penalty",
|
116 |
+
minimum=1.0,
|
117 |
+
maximum=2.0,
|
118 |
+
step=0.05,
|
119 |
+
value=1.2,
|
120 |
+
)
|
121 |
+
|
122 |
+
complete_btn = gr.Button("Complete")
|
123 |
+
|
124 |
+
complete_btn.click(
|
125 |
+
fn=generate,
|
126 |
+
inputs=[
|
127 |
+
input_box,
|
128 |
+
max_new_tokens,
|
129 |
+
temperature,
|
130 |
+
top_p,
|
131 |
+
top_k,
|
132 |
+
repetition_penalty
|
133 |
+
],
|
134 |
+
outputs=output_box
|
135 |
+
)
|
136 |
+
|
137 |
+
gr.Examples(
|
138 |
+
examples=[
|
139 |
+
"Hello there! How are you doing?",
|
140 |
+
"Can you explain briefly to me what is the Python programming language?",
|
141 |
+
"Explain the plot of Cinderella in a sentence.",
|
142 |
+
"How many hours does it take a man to eat a Helicopter?",
|
143 |
+
"Write a 100-word article on 'Benefits of Open-Source in AI research'",
|
144 |
+
],
|
145 |
+
inputs=input_box
|
146 |
+
)
|
147 |
|
148 |
if __name__ == "__main__":
|
149 |
demo.queue(max_size=20).launch()
|