PMAlpha

Sleeping

Sergidev commited on Aug 7

Commit

2036440

•

1 Parent(s): f54c000

Update modules/pmbl.py

Files changed (1) hide show

modules/pmbl.py CHANGED Viewed

@@ -13,7 +13,8 @@ class PMBL:
             model_type="llama",
             gpu_layers=self.gpu_layers,
             context_length=32768,
-            config={'use_flash_attention': True}
         )
     def init_db(self):

             model_type="llama",
             gpu_layers=self.gpu_layers,
             context_length=32768,
+            attn_implementation="flash_attention_2",
+            torch_dtype="auto"
         )
     def init_db(self):