Update custom_generate/generate.py

custom_generate/generate.py

@@ -43,6 +43,7 @@ from transformers.modeling_utils import  PreTrainedModel
 from .functions_2_patch import _validate_model_kwargs, llama_atten_forward
 from .monkey_patching_utils import monkey_patching
 from .sep_cache_utils import SepCache
+
 # except :
 #     from ..functions_2_patch import _validate_model_kwargs, llama_atten_forward
 #     from ..monkey_patching_utils import monkey_patching
@@ -120,9 +121,9 @@ def generate(model,
     print(f"__file__: {__file__}")
     print(f"os.path.abspath(__file__): {os.path.abspath(__file__)}")
 
-    from ..functions_2_patch import _validate_model_kwargs, llama_atten_forward
-    from ..monkey_patching_utils import monkey_patching
-    from ..sep_cache_utils import SepCache
+    # from ..functions_2_patch import _validate_model_kwargs, llama_atten_forward
+    # from ..monkey_patching_utils import monkey_patching
+    # from ..sep_cache_utils import SepCache
 
     
 
@@ -228,7 +229,7 @@ def generate(model,
     """
 
     # 0. Monkey Patching for the `update` function of `SepCache`    
-    model_layers = monkey_patching(model,  model_atten_forward=llama_atten_forward, verbose=monkey_patch_verbose)
+    # model_layers = monkey_patching(model,  model_atten_forward=llama_atten_forward, verbose=monkey_patch_verbose)
 
     # 1. General sanity checks
     # 1.a. A few arguments are not allowed, especially arguments that control caches.