Spaces:
Runtime error
Runtime error
ffreemt
commited on
Commit
•
634ed9b
1
Parent(s):
d82d66e
Update layout
Browse files
app.py
CHANGED
@@ -15,7 +15,6 @@ from transformers import AutoModel, AutoTokenizer
|
|
15 |
# os.system("pip install torch transformers sentencepiece loguru")
|
16 |
|
17 |
|
18 |
-
|
19 |
# fix timezone in Linux
|
20 |
os.environ["TZ"] = "Asia/Shanghai"
|
21 |
try:
|
@@ -42,9 +41,9 @@ if has_cuda:
|
|
42 |
AutoModel.from_pretrained(model_name, trust_remote_code=True).cuda().half()
|
43 |
)
|
44 |
else:
|
45 |
-
model =
|
46 |
-
model_name, trust_remote_code=True
|
47 |
-
)
|
48 |
|
49 |
model = model.eval()
|
50 |
logger.debug("done load")
|
@@ -54,7 +53,11 @@ logger.debug("done load")
|
|
54 |
|
55 |
# locate model file cache
|
56 |
cache_loc = Path("~/.cache/huggingface/hub").expanduser()
|
57 |
-
model_cache_path = [
|
|
|
|
|
|
|
|
|
58 |
|
59 |
logger.debug(f"{model_cache_path=}")
|
60 |
|
@@ -64,25 +67,70 @@ if model_cache_path:
|
|
64 |
|
65 |
|
66 |
def respond(message, chat_history):
|
67 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
68 |
chat_history.append((message, response))
|
69 |
-
return
|
|
|
70 |
|
71 |
theme = gr.themes.Soft(text_size="sm")
|
72 |
with gr.Blocks(theme=theme) as block:
|
73 |
chatbot = gr.Chatbot()
|
74 |
|
75 |
-
with gr.
|
76 |
with gr.Column(scale=12):
|
77 |
msg = gr.Textbox()
|
78 |
with gr.Column(scale=1, min_width=16):
|
79 |
-
with gr.Row():
|
80 |
btn = gr.Button("Send")
|
|
|
81 |
clear = gr.ClearButton([msg, chatbot])
|
82 |
|
83 |
# do not clear prompt
|
84 |
-
msg.submit(lambda x, y:
|
85 |
|
86 |
btn.click(respond, [msg, chatbot], [msg, chatbot])
|
87 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
88 |
block.queue().launch()
|
|
|
15 |
# os.system("pip install torch transformers sentencepiece loguru")
|
16 |
|
17 |
|
|
|
18 |
# fix timezone in Linux
|
19 |
os.environ["TZ"] = "Asia/Shanghai"
|
20 |
try:
|
|
|
41 |
AutoModel.from_pretrained(model_name, trust_remote_code=True).cuda().half()
|
42 |
)
|
43 |
else:
|
44 |
+
model = (
|
45 |
+
AutoModel.from_pretrained(model_name, trust_remote_code=True).half().float()
|
46 |
+
) # .float() .half().float(): must use float for cpu
|
47 |
|
48 |
model = model.eval()
|
49 |
logger.debug("done load")
|
|
|
53 |
|
54 |
# locate model file cache
|
55 |
cache_loc = Path("~/.cache/huggingface/hub").expanduser()
|
56 |
+
model_cache_path = [
|
57 |
+
elm
|
58 |
+
for elm in Path(cache_loc).rglob("*")
|
59 |
+
if Path(model_name).name in elm.as_posix() and "pytorch_model.bin" in elm.as_posix()
|
60 |
+
]
|
61 |
|
62 |
logger.debug(f"{model_cache_path=}")
|
63 |
|
|
|
67 |
|
68 |
|
69 |
def respond(message, chat_history):
|
70 |
+
"""Gen a response."""
|
71 |
+
message = message.strip()
|
72 |
+
response, chat_history = model.chat(
|
73 |
+
tokenizer,
|
74 |
+
message,
|
75 |
+
history=chat_history,
|
76 |
+
temperature=0.7,
|
77 |
+
repetition_penalty=1.2,
|
78 |
+
max_length=128,
|
79 |
+
)
|
80 |
chat_history.append((message, response))
|
81 |
+
return message, chat_history
|
82 |
+
|
83 |
|
84 |
theme = gr.themes.Soft(text_size="sm")
|
85 |
with gr.Blocks(theme=theme) as block:
|
86 |
chatbot = gr.Chatbot()
|
87 |
|
88 |
+
with gr.Row():
|
89 |
with gr.Column(scale=12):
|
90 |
msg = gr.Textbox()
|
91 |
with gr.Column(scale=1, min_width=16):
|
|
|
92 |
btn = gr.Button("Send")
|
93 |
+
with gr.Column(scale=1, min_width=8):
|
94 |
clear = gr.ClearButton([msg, chatbot])
|
95 |
|
96 |
# do not clear prompt
|
97 |
+
msg.submit(lambda x, y: (x,) + respond(x, y)[1:], [msg, chatbot], [msg, chatbot])
|
98 |
|
99 |
btn.click(respond, [msg, chatbot], [msg, chatbot])
|
100 |
|
101 |
+
with gr.Accordion("Example inputs", open=True):
|
102 |
+
etext = """In America, where cars are an important part of the national psyche, a decade ago people had suddenly started to drive less, which had not happened since the oil shocks of the 1970s. """
|
103 |
+
examples = gr.Examples(
|
104 |
+
examples=[
|
105 |
+
["Explain the plot of Cinderella in a sentence."],
|
106 |
+
[
|
107 |
+
"How long does it take to become proficient in French, and what are the best methods for retaining information?"
|
108 |
+
],
|
109 |
+
["What are some common mistakes to avoid when writing code?"],
|
110 |
+
["Build a prompt to generate a beautiful portrait of a horse"],
|
111 |
+
["Suggest four metaphors to describe the benefits of AI"],
|
112 |
+
["Write a pop song about leaving home for the sandy beaches."],
|
113 |
+
["Write a summary demonstrating my ability to tame lions"],
|
114 |
+
["鲁迅和周树人什么关系"],
|
115 |
+
["从前有一头牛,这头牛后面有什么?"],
|
116 |
+
["正无穷大加一大于正无穷大吗?"],
|
117 |
+
["正无穷大加正无穷大大于正无穷大吗?"],
|
118 |
+
["-2的平方根等于什么"],
|
119 |
+
["树上有5只鸟,猎人开枪打死了一只。树上还有几只鸟?"],
|
120 |
+
["树上有11只鸟,猎人开枪打死了一只。树上还有几只鸟?提示:需考虑鸟可能受惊吓飞走。"],
|
121 |
+
["鲁迅和周树人什么关系 用英文回答"],
|
122 |
+
["以红楼梦的行文风格写一张委婉的请假条。不少于320字。"],
|
123 |
+
[f"{etext} 翻成中文,列出3个版本"],
|
124 |
+
[f"{etext} \n 翻成中文,保留原意,但使用文学性的语言。不要写解释。列出3个版本"],
|
125 |
+
["js 判断一个数是不是质数"],
|
126 |
+
["js 实现python 的 range(10)"],
|
127 |
+
["js 实现python 的 [*(range(10)]"],
|
128 |
+
["假定 1 + 2 = 4, 试求 7 + 8"],
|
129 |
+
["Erkläre die Handlung von Cinderella in einem Satz."],
|
130 |
+
["Erkläre die Handlung von Cinderella in einem Satz. Auf Deutsch"],
|
131 |
+
],
|
132 |
+
inputs=[msg],
|
133 |
+
examples_per_page=60,
|
134 |
+
)
|
135 |
+
|
136 |
block.queue().launch()
|