add model_file

bart_demo_gradio.py

@@ -3,19 +3,21 @@ import gradio as gr
 import torch
 import transformers
 
+
+
 # saved_model
-def load_model(model_path, config):
+def load_model(model_path):
     saved_data = torch.load(
         model_path,
-        map_location="cpu" if config.gpu_id < 0 else "cuda:%d" % config.gpu_id
+        map_location="cpu"
     )
 
     bart_best = saved_data["model"]
     train_config = saved_data["config"]
-    tokenizer = transformers.PreTrainedTokenizerFast.from_pretrained(config.pretrained_model_name)
+    tokenizer = transformers.PreTrainedTokenizerFast.from_pretrained('gogamza/kobart-base-v1')
 
     ## Load weights.
-    model = transformers.BartForConditionalGeneration.from_pretrained(config.pretrained_model_name)
+    model = transformers.BartForConditionalGeneration.from_pretrained('gogamza/kobart-base-v1')
     model.load_state_dict(bart_best)
 
     return model, tokenizer
@@ -23,13 +25,10 @@ def load_model(model_path, config):
 
 # main
 def inference(prompt):
-
-    config = define_argparser()
-    model_path = config.model_fpath
+    model_path = "./kobart-model-logical.pth"
 
     model, tokenizer = load_model(
-        model_path=model_path, 
-        config=config
+        model_path=model_path
         )
 
     input_ids = tokenizer.encode(prompt)
@@ -40,6 +39,7 @@ def inference(prompt):
 
     return output
 
+
 demo = gr.Interface(
     fn=inference, 
     inputs="text", 

kobart-model-logical.pth

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:f108a09b663ed85f66e384b51922867ed9fab15b3b7de3a20c1f7389e4ffdeb4
+size 496665407