delete local args for global models

app.py

@@ -360,11 +360,11 @@ def construction_all():
 
 @torch.no_grad()   
 @spaces.GPU(duration=120) 
-def evaluate_v1(inputs, model, quantizer, tokenizer, width, height, do_sample=False, temperature=1.0, top_p=1.0, top_k=50):
+def evaluate_v1(inputs, width, height, do_sample=False, temperature=1.0, top_p=1.0, top_k=50):
     print(f"evaluate_v1")
     print(f"evaluate_v1 {model.device} {model.lm.device} {pipeline.device}")
-    model = model.to("cuda")
-    print(f"after evaluate_v1 {model.device} {model.lm.device} {pipeline.device}")
+    # model = model.to("cuda")
+    # print(f"after evaluate_v1 {model.device} {model.lm.device} {pipeline.device}")
     
     json_example = inputs
     input_intension = '{"wholecaption":"' + json_example["wholecaption"] + '","layout":[{"layer":'
@@ -407,7 +407,7 @@ def evaluate_v1(inputs, model, quantizer, tokenizer, width, height, do_sample=Fa
         pred_json_example = None
     return pred_json_example
 
-def inference(generate_method, intention, model, quantizer, tokenizer, width, height, do_sample=True, temperature=1.0, top_p=1.0, top_k=50):
+def inference(generate_method, intention, width, height, do_sample=True, temperature=1.0, top_p=1.0, top_k=50):
     print(f"start inference")
     rawdata = {}
     rawdata["wholecaption"] = intention
@@ -418,7 +418,7 @@ def inference(generate_method, intention, model, quantizer, tokenizer, width, he
         preddata = None
         while preddata is None and max_try_time > 0:
             print(f"preddata = evaluate_v1")
-            preddata = evaluate_v1(rawdata, model, quantizer, tokenizer, width, height, do_sample=do_sample, temperature=temperature, top_p=top_p, top_k=top_k)
+            preddata = evaluate_v1(rawdata, width, height, do_sample=do_sample, temperature=temperature, top_p=top_p, top_k=top_k)
             max_try_time -= 1
     else:
         print("Please input correct generate method")
@@ -434,11 +434,11 @@ def process_preddate(intention, temperature, top_p, generate_method='v1'):
         # print("looking for greedy decoding strategies, set `do_sample=False`.")
         # preddata = inference_partial(generate_method, intention, do_sample=False)
         print(f"preddata = inference temperatrue = 0.0")
-        preddata = inference(generate_method, intention, model=model, quantizer=quantizer, tokenizer=tokenizer, width=512, height=512, do_sample=False)
+        preddata = inference(generate_method, intention, width=512, height=512, do_sample=False)
     else:
         # preddata = inference_partial(generate_method, intention, temperature=temperature, top_p=top_p)
         print(f"preddata = inference temperatrue != 0.0")
-        preddata = inference(generate_method, intention, model=model, quantizer=quantizer, tokenizer=tokenizer, width=512, height=512, temperature=temperature, top_p=top_p)
+        preddata = inference(generate_method, intention, width=512, height=512, temperature=temperature, top_p=top_p)
         
     layouts = preddata["layout"]
     list_box = []
@@ -471,8 +471,9 @@ def process_preddate(intention, temperature, top_p, generate_method='v1'):
     return str(filtered_boxes), intention, str(filtered_boxes)
 
 @spaces.GPU(duration=120)
-def test_one_sample(validation_box, validation_prompt, true_gs, inference_steps, pipeline, generator, transp_vae):
+def test_one_sample(validation_box, validation_prompt, true_gs, inference_steps, seed):
     print(f"test_one_sample: {validation_box}")
+    generator = torch.Generator(device=torch.device("cuda", index=0)).manual_seed(seed)
     output, rgba_output, _, _ = pipeline(
         prompt=validation_prompt,
         validation_box=validation_box,
@@ -500,11 +501,9 @@ def test_one_sample(validation_box, validation_prompt, true_gs, inference_steps,
     
     return output_gradio
 
-def gradio_test_one_sample(validation_prompt, validation_box_str, seed, true_gs, inference_steps, pipeline, transp_vae):
+def gradio_test_one_sample(validation_prompt, validation_box_str, seed, true_gs, inference_steps):
     print(f"svg_test_one_sample")
-    # print(f"svg_test_one_sample {model.device} {model.lm.device} {pipeline.device}")
-    # generator = torch.Generator().manual_seed(seed)
-    generator = torch.Generator(device=torch.device("cuda", index=0)).manual_seed(seed)
+    
     try:
         if isinstance(validation_box_str, (list, tuple)):
             validation_box = validation_box_str
@@ -518,7 +517,7 @@ def gradio_test_one_sample(validation_prompt, validation_box_str, seed, true_gs,
     validation_box = adjust_validation_box(validation_box)
     
     print("result_images = test_one_sample")
-    result_images = test_one_sample(validation_box, validation_prompt, true_gs, inference_steps, pipeline, generator, transp_vae)
+    result_images = test_one_sample(validation_box, validation_prompt, true_gs, inference_steps, seed)
     print("after result_images = test_one_sample")
     svg_img = pngs_to_svg(result_images[1:])
 
@@ -543,7 +542,7 @@ def process_svg(text_input, tuple_input, seed, true_gs, inference_steps):
     print(f"precess_svg")
     # print(f"precess_svg {model.device} {model.lm.device} {pipeline.device}")
     result_images = []
-    result_images, svg_file_path = gradio_test_one_sample(text_input, tuple_input, seed, true_gs, inference_steps, pipeline=pipeline, transp_vae=transp_vae)
+    result_images, svg_file_path = gradio_test_one_sample(text_input, tuple_input, seed, true_gs, inference_steps)
     # result_images, svg_file_path = gradio_test_one_sample_partial(text_input, tuple_input, seed, true_gs, inference_steps)
 
     url, unique_filename = upload_to_github(file_path=svg_file_path)