Rename

- forward_old → forward_org
- output filename : llama → tinyllama
- LlamaDecoderLayerWithCache → LlamaDecoderLayerWithKVCache

Author: Sanggyu Lee

…odel/LlamaDecoderLayerWithCache/model.py …ache/LlamaDecoderLayerWithCache/model.pytest/modules/model/LlamaDecoderLayerWithCache/model.py renamed to test/modules/model/LlamaDecoderLayerWithKVCache/LlamaDecoderLayerWithCache/model.py test/modules/model/LlamaDecoderLayerWithCache/model.py renamed to test/modules/model/LlamaDecoderLayerWithKVCache/LlamaDecoderLayerWithCache/model.py

@@ -8,15 +8,15 @@
 
 from transformers.models.llama.modeling_llama import LlamaDecoderLayer
 
-forward_old = LlamaDecoderLayer.forward
+forward_org = LlamaDecoderLayer.forward
 
 
 def capture_and_forward(self, *args, **kwargs):
     global captured_input
 
     # Prepare args tuple for TICO.convert()
     # Get arg_names in positional args order using inspect
-    sig = inspect.signature(forward_old)
+    sig = inspect.signature(forward_org)
     args_names = [
         # signature includes `self`` and `kwargs``.
         # Just retrieve the ordinary positional inputs only
@@ -38,7 +38,7 @@ def populate_args(args_dict, filter):
         input_to_remove = ["use_cache"]
         captured_input = populate_args(args_dict, input_to_remove)
 
-    return forward_old(self, *args, **kwargs)
+    return forward_org(self, *args, **kwargs)
 
 
 # Tokenizer
@@ -82,4 +82,4 @@ def populate_args(args_dict, filter):
 model = AutoModelForCausalLM.from_pretrained(model_name)
 model.eval()
 circle_model = tico.convert(model.model.layers[0], captured_input)
-circle_model.save(f"llama.decoderlayer.circle")
+circle_model.save(f"tinyllama.decoderlayer.circle")