Commit
•
f4d4880
1
Parent(s):
da3b30c
bug fix
Browse files
app.py
CHANGED
|
@@ -43,7 +43,7 @@ def load_lora_model(
|
|
| 43 |
model_path,
|
| 44 |
load_in_8bit=load_8bit,
|
| 45 |
device_map="auto" if device == "cuda" else {"": device},
|
| 46 |
-
max_memory=max_gpu_memory,
|
| 47 |
torch_dtype=torch.float16,
|
| 48 |
)
|
| 49 |
if lora_weight is not None:
|
|
|
|
| 43 |
model_path,
|
| 44 |
load_in_8bit=load_8bit,
|
| 45 |
device_map="auto" if device == "cuda" else {"": device},
|
| 46 |
+
max_memory={i: max_gpu_memory for i in range(num_gpus)},
|
| 47 |
torch_dtype=torch.float16,
|
| 48 |
)
|
| 49 |
if lora_weight is not None:
|