reer to 90sec model

app.py

@@ -40,9 +40,12 @@ def clear_text():
     return gr.update(value="")  # Clears the text field
 
 #@spaces.GPU
-def infer_music(lrc, ref_audio_path, steps, file_type, cfg_strength, odeint_method, duration, prompt=None):
+#def infer_music(lrc, ref_audio_path, steps, file_type, cfg_strength, odeint_method, duration, prompt=None):
+def infer_music(lrc, ref_audio_path, steps, file_type, cfg_strength, odeint_method, prompt=None):
 
-    max_frames = math.floor(duration * 21.56)
+    duration = 95
+    max_frames = 2048
+    #max_frames = math.floor(duration * 21.56)
     sway_sampling_coef = -1 if steps < 32 else None
     vocal_flag = False
     lrc_prompt, start_time = get_lrc_token(max_frames, lrc, tokenizer, device)
@@ -227,7 +230,7 @@ with gr.Blocks(css=css) as demo:
                     
                 with gr.Column():
                   
-                    duration = gr.Slider(95, 285, value=285, label="Music Duration")
+                    #duration = gr.Slider(95, 285, value=285, label="Music Duration")
                     lyrics_btn = gr.Button("Submit", variant="primary")
                     audio_output = gr.Audio(label="Audio Result", type="filepath", elem_id="audio_output")
                     with gr.Accordion("Advanced Settings", open=False):
@@ -379,7 +382,8 @@ with gr.Blocks(css=css) as demo:
     
     lyrics_btn.click(
         fn=infer_music,
-        inputs=[lrc, audio_prompt, steps, file_type, cfg_strength, odeint_method, duration, text_prompt, ],
+        #inputs=[lrc, audio_prompt, steps, file_type, cfg_strength, odeint_method, duration, text_prompt, ],
+        inputs=[lrc, audio_prompt, steps, file_type, cfg_strength, odeint_method, text_prompt, ],
         outputs=audio_output
     )
 

diffrhythm/infer/infer_utils.py

@@ -12,8 +12,8 @@ from diffrhythm.model import DiT, CFM
 
 def prepare_model(device):
     # prepare cfm model
-    #dit_ckpt_path = hf_hub_download(repo_id="ASLP-lab/DiffRhythm-base", filename="cfm_model.pt")
-    dit_ckpt_path = hf_hub_download(repo_id="ASLP-lab/DiffRhythm-full", filename="cfm_model.pt")
+    dit_ckpt_path = hf_hub_download(repo_id="ASLP-lab/DiffRhythm-base", filename="cfm_model.pt")
+    #dit_ckpt_path = hf_hub_download(repo_id="ASLP-lab/DiffRhythm-full", filename="cfm_model.pt")
     dit_config_path = "./diffrhythm/config/diffrhythm-1b.json"
     with open(dit_config_path, encoding="utf-8") as f:
         model_config = json.load(f)