Spaces:
Runtime error
Runtime error
Update app.py
Browse files
app.py
CHANGED
@@ -1,18 +1,45 @@
|
|
1 |
import os
|
|
|
|
|
2 |
# os.system("pip install --upgrade torch transformers sentencepiece scipy cpm_kernels accelerate bitsandbytes loguru")
|
3 |
-
os.system("pip install transformers loguru")
|
4 |
|
5 |
import gradio as gr
|
6 |
from transformers import AutoTokenizer, AutoModel, AutoModelForCausalLM
|
7 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
8 |
tokenizer = AutoTokenizer.from_pretrained("THUDM/chatglm2-6b-int4", trust_remote_code=True)
|
|
|
|
|
|
|
|
|
9 |
logger.debug("load")
|
10 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
11 |
logger.debug("done load")
|
|
|
12 |
# tokenizer = AutoTokenizer.from_pretrained("openchat/openchat_v2_w")
|
13 |
# model = AutoModelForCausalLM.from_pretrained("openchat/openchat_v2_w", load_in_8bit_fp32_cpu_offload=True, load_in_8bit=True)
|
14 |
-
model.half()
|
15 |
-
model = model.eval()
|
16 |
|
17 |
model_path = model.config._dict['model_name_or_path']
|
18 |
logger.debug(f"{model_path=}")
|
|
|
1 |
import os
|
2 |
+
import time
|
3 |
+
|
4 |
# os.system("pip install --upgrade torch transformers sentencepiece scipy cpm_kernels accelerate bitsandbytes loguru")
|
5 |
+
os.system("pip install torch transformers sentencepiece loguru")
|
6 |
|
7 |
import gradio as gr
|
8 |
from transformers import AutoTokenizer, AutoModel, AutoModelForCausalLM
|
9 |
|
10 |
+
# fix timezone in Linux
|
11 |
+
os.environ["TZ"] = "Asia/Shanghai"
|
12 |
+
try:
|
13 |
+
time.tzset() # type: ignore # pylint: disable=no-member
|
14 |
+
except Exception:
|
15 |
+
# Windows
|
16 |
+
logger.warning("Windows, cant run time.tzset()")
|
17 |
+
|
18 |
+
model_name = "THUDM/chatglm2-6b-int4" # 3.9G
|
19 |
+
|
20 |
tokenizer = AutoTokenizer.from_pretrained("THUDM/chatglm2-6b-int4", trust_remote_code=True)
|
21 |
+
|
22 |
+
has_cuda = torch.cuda.is_available()
|
23 |
+
# has_cuda = False # force cpu
|
24 |
+
|
25 |
logger.debug("load")
|
26 |
+
if has_cuda:
|
27 |
+
if model_name.endswith("int4"):
|
28 |
+
model = AutoModel.from_pretrained(model_name, trust_remote_code=True).cuda()
|
29 |
+
else:
|
30 |
+
model = (
|
31 |
+
AutoModel.from_pretrained(model_name, trust_remote_code=True).cuda().half()
|
32 |
+
)
|
33 |
+
else:
|
34 |
+
model = AutoModel.from_pretrained(
|
35 |
+
model_name, trust_remote_code=True
|
36 |
+
).half() # .float() .half().float()
|
37 |
+
|
38 |
+
model = model.eval()
|
39 |
logger.debug("done load")
|
40 |
+
|
41 |
# tokenizer = AutoTokenizer.from_pretrained("openchat/openchat_v2_w")
|
42 |
# model = AutoModelForCausalLM.from_pretrained("openchat/openchat_v2_w", load_in_8bit_fp32_cpu_offload=True, load_in_8bit=True)
|
|
|
|
|
43 |
|
44 |
model_path = model.config._dict['model_name_or_path']
|
45 |
logger.debug(f"{model_path=}")
|