Spaces:
Running
Running
app.py
CHANGED
@@ -14,6 +14,8 @@ tokenizer = AutoTokenizer.from_pretrained(
|
|
14 |
trust_remote_code=True,
|
15 |
use_fast=True,
|
16 |
)
|
|
|
|
|
17 |
|
18 |
model = AutoModelForCausalLM.from_pretrained(
|
19 |
model_path,
|
@@ -22,10 +24,11 @@ model = AutoModelForCausalLM.from_pretrained(
|
|
22 |
device_map="auto",
|
23 |
offload_folder="offload",
|
24 |
offload_state_dict=True,
|
|
|
25 |
trust_remote_code=True,
|
26 |
)
|
27 |
|
28 |
-
|
29 |
|
30 |
# Создаём пайплайн
|
31 |
pipe = pipeline("text-generation", model=model, tokenizer=tokenizer)
|
|
|
14 |
trust_remote_code=True,
|
15 |
use_fast=True,
|
16 |
)
|
17 |
+
# Ensure offload directory exists before loading
|
18 |
+
os.makedirs("offload", exist_ok=True)
|
19 |
|
20 |
model = AutoModelForCausalLM.from_pretrained(
|
21 |
model_path,
|
|
|
24 |
device_map="auto",
|
25 |
offload_folder="offload",
|
26 |
offload_state_dict=True,
|
27 |
+
low_cpu_mem_usage=True,
|
28 |
trust_remote_code=True,
|
29 |
)
|
30 |
|
31 |
+
|
32 |
|
33 |
# Создаём пайплайн
|
34 |
pipe = pipeline("text-generation", model=model, tokenizer=tokenizer)
|