Update app.py

app.py

@@ -13,76 +13,6 @@ from midi_synthesizer import synthesis
 
 in_space = os.getenv("SYSTEM") == "spaces"
 
-
-def find_midi():
-    if disable_channels is not None:
-        disable_channels = [tokenizer.parameter_ids["channel"][c] for c in disable_channels]
-    else:
-        disable_channels = []
-    max_token_seq = tokenizer.max_token_seq
-    if prompt is None:
-        input_tensor = np.full((1, max_token_seq), tokenizer.pad_id, dtype=np.int64)
-        input_tensor[0, 0] = tokenizer.bos_id  # bos
-    else:
-        prompt = prompt[:, :max_token_seq]
-        if prompt.shape[-1] < max_token_seq:
-            prompt = np.pad(prompt, ((0, 0), (0, max_token_seq - prompt.shape[-1])),
-                            mode="constant", constant_values=tokenizer.pad_id)
-        input_tensor = prompt
-    input_tensor = input_tensor[None, :, :]
-    cur_len = input_tensor.shape[1]
-    bar = tqdm.tqdm(desc="generating", total=max_len - cur_len, disable=in_space)
-    with bar:
-        while cur_len < max_len:
-            end = False
-            hidden = model[0].run(None, {'x': input_tensor})[0][:, -1]
-            next_token_seq = np.empty((1, 0), dtype=np.int64)
-            event_name = ""
-            for i in range(max_token_seq):
-                mask = np.zeros(tokenizer.vocab_size, dtype=np.int64)
-                if i == 0:
-                    mask_ids = list(tokenizer.event_ids.values()) + [tokenizer.eos_id]
-                    if disable_patch_change:
-                        mask_ids.remove(tokenizer.event_ids["patch_change"])
-                    if disable_control_change:
-                        mask_ids.remove(tokenizer.event_ids["control_change"])
-                    mask[mask_ids] = 1
-                else:
-                    param_name = tokenizer.events[event_name][i - 1]
-                    mask_ids = tokenizer.parameter_ids[param_name]
-                    if param_name == "channel":
-                        mask_ids = [i for i in mask_ids if i not in disable_channels]
-                    mask[mask_ids] = 1
-                logits = model[1].run(None, {'x': next_token_seq, "hidden": hidden})[0][:, -1:]
-                scores = softmax(logits / temp, -1) * mask
-                sample = sample_top_p_k(scores, top_p, top_k)
-                if i == 0:
-                    next_token_seq = sample
-                    eid = sample.item()
-                    if eid == tokenizer.eos_id:
-                        end = True
-                        break
-                    event_name = tokenizer.id_events[eid]
-                else:
-                    next_token_seq = np.concatenate([next_token_seq, sample], axis=1)
-                    if len(tokenizer.events[event_name]) == i:
-                        break
-            if next_token_seq.shape[1] < max_token_seq:
-                next_token_seq = np.pad(next_token_seq, ((0, 0), (0, max_token_seq - next_token_seq.shape[-1])),
-                                        mode="constant", constant_values=tokenizer.pad_id)
-            next_token_seq = next_token_seq[None, :, :]
-            input_tensor = np.concatenate([input_tensor, next_token_seq], axis=1)
-            cur_len += 1
-            bar.update(1)
-            yield next_token_seq.reshape(-1)
-            if end:
-                break
-
-
-def create_msg(name, data):
-    return {"name": name, "data": data}
-
-
 def run(search_prompt, mid=None):
     mid_seq = []
 
@@ -95,24 +25,11 @@ def run(search_prompt, mid=None):
        
     elif mid is not None:
         mid_seq = MIDI.midi2score(mid)
-
-    init_msgs = [create_msg("visualizer_clear", None)]
-    for event in mid_seq:
-        if event[0] == 'note':
-            init_msgs.append(create_msg("visualizer_append", event))
-    yield mid_seq, None, None, init_msgs
-
-    # j = 0
-    # for i in range(len(mid_seq)-1):
-    #    if mid_seq[i][0] == 'note':
-    #        j += 1
-    #        yield mid_seq, None, None, [create_msg("visualizer_append", mid_seq[i]), create_msg("progress", [j + 1, len(mid_seq)])]
-            
-
+ 
     with open(f"output.mid", 'wb') as f:
         f.write(MIDI.score2midi([mid_seq_ticks, mid_seq]))
     audio = synthesis(MIDI.score2opus([mid_seq_ticks, mid_seq]), soundfont_path)
-    yield mid_seq, "output.mid", (44100, audio), [create_msg("visualizer_end", None)]
+    yield mid_seq, "output.mid", (44100, audio)
 
 
 def cancel_run(mid_seq):
@@ -120,40 +37,9 @@ def cancel_run(mid_seq):
         return None, None
 
     with open(f"output.mid", 'wb') as f:
-        f.write(MIDI.score2midi(mid_seq))
-    audio = synthesis(MIDI.score2opus(mid_seq), soundfont_path)
-    return "output.mid", (44100, audio), [create_msg("visualizer_end", None)]
-
-
-def load_javascript(dir="javascript"):
-    scripts_list = glob.glob(f"{dir}/*.js")
-    javascript = ""
-    for path in scripts_list:
-        with open(path, "r", encoding="utf8") as jsfile:
-            javascript += f"\n<!-- {path} --><script>{jsfile.read()}</script>"
-    template_response_ori = gr.routes.templates.TemplateResponse
-
-    def template_response(*args, **kwargs):
-        res = template_response_ori(*args, **kwargs)
-        res.body = res.body.replace(
-            b'</head>', f'{javascript}</head>'.encode("utf8"))
-        res.init_headers()
-        return res
-
-    gr.routes.templates.TemplateResponse = template_response
-
-
-class JSMsgReceiver(gr.HTML):
-    def __init__(self, **kwargs):
-        super().__init__(elem_id="msg_receiver", visible=False, **kwargs)
-    
-    def postprocess(self, y):
-        if y:
-            y = f"<p>{json.dumps(y)}</p>"
-        return super().postprocess(y)
-    
-    def get_block_name(self) -> str:
-        return "html"
+        f.write(MIDI.score2midi([1000, mid_seq]))
+    audio = synthesis(MIDI.score2opus([1000, mid_seq]), soundfont_path)
+    return "output.mid", (44100, audio)
 
 if __name__ == "__main__":
     parser = argparse.ArgumentParser()
@@ -176,8 +62,6 @@ if __name__ == "__main__":
         meta_data = pickle.load(f)
     print('Done!')
     
-    
-    load_javascript()
     app = gr.Blocks()
     with app:
         gr.Markdown("<h1 style='text-align: center; margin-bottom: 1rem'>MIDI Search</h1>")
@@ -189,8 +73,6 @@ if __name__ == "__main__":
                     " for faster running and longer generation"
                     )
         
-        js_msg = JSMsgReceiver()
-        
         with gr.Tabs():
             with gr.TabItem("instrument prompt") as tab1:
                 
@@ -206,10 +88,9 @@ if __name__ == "__main__":
         search_btn = gr.Button("search", variant="primary")
         stop_btn = gr.Button("stop and output")
         output_midi_seq = gr.Textbox()
-        output_midi_visualizer = gr.HTML(elem_id="midi_visualizer_container")
         output_audio = gr.Audio(label="output audio", format="mp3", elem_id="midi_audio")
         output_midi = gr.File(label="output midi", file_types=[".mid"])
         run_event = search_btn.click(run, [search_prompt],
-                                  [output_midi_seq, output_midi, output_audio, js_msg])
-        stop_btn.click(cancel_run, output_midi_seq, [output_midi, output_audio, js_msg], cancels=run_event, queue=False)
+                                  [output_midi_seq, output_midi, output_audio])
+        stop_btn.click(cancel_run, output_midi_seq, [output_midi, output_audio], cancels=run_event, queue=False)
     app.queue(1).launch(server_port=opt.port, share=opt.share, inbrowser=True)