Update app.py

app.py

@@ -4,18 +4,17 @@ import time
 import torch
 import shutil
 import argparse
+import warnings
 import gradio as gr
 from config import *
 from utils import Patchilizer, TunesFormer, DEVICE
-from convert import abc_to_midi, abc_to_musicxml, musicxml_to_mxl, mxl2jpg, midi2wav
+from convert import abc2xml, xml2, xml2img
 from modelscope import snapshot_download
 from transformers import GPT2Config
-import warnings
 
-warnings.filterwarnings("ignore")
 
 # 模型下载
-MODEL_DIR = snapshot_download("MuGeminorum/hoyoGPT")
+WEIGHTS_PATH = snapshot_download("MuGeminorum/hoyoGPT") + "/weights.pth"
 
 
 def get_args(parser: argparse.ArgumentParser):
@@ -61,7 +60,7 @@ def get_args(parser: argparse.ArgumentParser):
     return args
 
 
-def generate_abc(args, epochs: str, region: str):
+def generate_abc(args, region: str):
     patchilizer = Patchilizer()
 
     patch_config = GPT2Config(
@@ -79,14 +78,11 @@ def generate_abc(args, epochs: str, region: str):
     )
 
     model = TunesFormer(patch_config, char_config, share_weights=SHARE_WEIGHTS)
-    filename = f"{MODEL_DIR}/{epochs}/weights.pth"
-    checkpoint = torch.load(filename, map_location=torch.device("cpu"))
+    checkpoint = torch.load(WEIGHTS_PATH, map_location=torch.device("cpu"))
     model.load_state_dict(checkpoint["model"])
     model = model.to(DEVICE)
     model.eval()
-
     prompt = f"A:{region}\n"
-
     tunes = ""
     num_tunes = args.num_tunes
     max_patch = args.max_patch
@@ -96,14 +92,12 @@ def generate_abc(args, epochs: str, region: str):
     seed = args.seed
     show_control_code = args.show_control_code
 
-    print(" HYPERPARAMETERS ".center(60, "#"), "\n")
+    print(" Hyper parms ".center(60, "#"), "\n")
     arg_dict: dict = vars(args)
-
     for key in arg_dict.keys():
         print(f"{key}: {str(arg_dict[key])}")
 
-    print("\n", " OUTPUT TUNES ".center(60, "#"))
-
+    print("\n", " Output tunes ".center(60, "#"))
     start_time = time.time()
     for i in range(num_tunes):
         title_artist = f"T:{region} Fragment\nC:Generated by AI\n"
@@ -179,40 +173,37 @@ def generate_abc(args, epochs: str, region: str):
     os.makedirs(TEMP_DIR, exist_ok=True)
     timestamp = time.strftime("%a_%d_%b_%Y_%H_%M_%S", time.localtime())
     try:
-        out_midi = abc_to_midi(tunes, f"{TEMP_DIR}/[{region}]{timestamp}.mid")
-        out_xml = abc_to_musicxml(tunes, f"{TEMP_DIR}/[{region}]{timestamp}.musicxml")
-        out_mxl = musicxml_to_mxl(f"{TEMP_DIR}/[{region}]{timestamp}.musicxml")
-        pdf_file, jpg_file = mxl2jpg(out_mxl)
-        wav_file = midi2wav(out_midi)
+        xml = abc2xml(tunes, f"{TEMP_DIR}/[{region}]{timestamp}.musicxml")
+        midi = xml2(xml, "mid")
+        audio = xml2(xml, "wav")
+        pdf, jpg = xml2img(xml)
+        mxl = xml2(xml, "mxl")
 
-        return tunes, out_midi, pdf_file, out_xml, out_mxl, jpg_file, wav_file
+        return tunes, midi, pdf, xml, mxl, jpg, audio
 
     except Exception as e:
         print(f"Invalid abc generated: {e}, retrying...")
-        return generate_abc(args, epochs, region)
+        return generate_abc(args, region)
 
 
-def inference(epochs: str, region: str):
+def inference(region: str):
     if os.path.exists(TEMP_DIR):
         shutil.rmtree(TEMP_DIR)
 
     parser = argparse.ArgumentParser()
     args = get_args(parser)
-    return generate_abc(args, epochs, region)
+    return generate_abc(args, region)
 
 
 if __name__ == "__main__":
+    warnings.filterwarnings("ignore")
+
     with gr.Blocks() as demo:
         gr.Markdown(
-            """<center>Welcome to this Space, made by bilibili <a href="https://space.bilibili.com/30620472">@MuGeminorum</a> based on the Tunesformer open source project, completely free.</center>"""
+            """<center>Welcome to this space, made by bilibili <a href="https://space.bilibili.com/30620472">@MuGeminorum</a> based on Tunesformer open source project, totally free.</center>"""
         )
         with gr.Row():
             with gr.Column():
-                weight_opt = gr.Dropdown(
-                    choices=["5", "15", "20"],
-                    value="15",
-                    label="Model Selection(epochs)",
-                )
                 region_opt = gr.Dropdown(
                     choices=["Mondstadt", "Liyue", "Inazuma", "Sumeru", "Fontaine"],
                     value="Mondstadt",
@@ -222,9 +213,7 @@ if __name__ == "__main__":
                 gr.Markdown(
                     """
     <center>
-    Currently, the model is still under debugging, and since the training data is converted from MIDI, there are a lot of spectral irregularities, which leads to many generated results with spectral irregularities and other problems.<br>
-
-    Planned in the Genshin main line killed, all countries and regions after all the characters are open, the second creation of the concert will be complete and balanced samples, then re-fine-tune the model and add the reality of the style of screening to assist the game of the various countries output gatekeepers, in order to enhance the output of the differentiation and quality.
+    Currently, the model is still under debugging. Planned in the Genshin main line killed, all countries and regions after all the characters are open, the second creation of the concert will be complete and balanced samples, then re-fine-tune the model and add the reality of the style of screening to assist the game of the various countries output gatekeepers, in order to enhance the output of the differentiation and quality.
 
     Data source: <a href="https://musescore.org">MuseScore</a><br>
     Tag embedded data source: <a href="https://genshin-impact.fandom.com/wiki/Genshin_Impact_Wiki">Genshin Impact Wiki | Fandom</a><br>
@@ -236,7 +225,7 @@ if __name__ == "__main__":
             with gr.Column():
                 wav_output = gr.Audio(label="Audio", type="filepath")
                 dld_midi = gr.components.File(label="Download MIDI")
-                pdf_score = gr.components.File(label="Download PDF Score")
+                pdf_score = gr.components.File(label="Download PDF score")
                 dld_xml = gr.components.File(label="Download MusicXML")
                 dld_mxl = gr.components.File(label="Download MXL")
                 abc_output = gr.Textbox(label="abc notation", show_copy_button=True)
@@ -244,7 +233,7 @@ if __name__ == "__main__":
 
         gen_btn.click(
             inference,
-            inputs=[weight_opt, region_opt],
+            inputs=region_opt,
             outputs=[
                 abc_output,
                 dld_midi,