Update app.py
Browse files
app.py
CHANGED
|
@@ -63,7 +63,7 @@ model = AutoModelForCausalLM.from_pretrained(model_id, token= token,
|
|
| 63 |
load_in_8bit=True,
|
| 64 |
# # # torch_dtype=torch.fl,
|
| 65 |
attn_implementation="flash_attention_2",
|
| 66 |
-
|
| 67 |
# device_map='cuda',
|
| 68 |
# device_map=accelerator.device_map,
|
| 69 |
|
|
|
|
| 63 |
load_in_8bit=True,
|
| 64 |
# # # torch_dtype=torch.fl,
|
| 65 |
attn_implementation="flash_attention_2",
|
| 66 |
+
low_cpu_mem_usage=True,
|
| 67 |
# device_map='cuda',
|
| 68 |
# device_map=accelerator.device_map,
|
| 69 |
|