try onnx again

README.md

@@ -5,7 +5,7 @@ colorFrom: red
 colorTo: indigo
 sdk: gradio
 sdk_version: 4.43.0
-app_file: app.py
+app_file: app_onnx.py
 pinned: true
 license: apache-2.0
 ---

app_onnx.py

@@ -170,9 +170,10 @@ def run(model_name, tab, mid_seq, continuation_state, continuation_select, instr
         key_sig, mid, midi_events, reduce_cc_st, remap_track_channel, add_default_instr, remove_empty_channels,
         seed, seed_rand, gen_events, temp, top_p, top_k, allow_cc):
     model = models[model_name]
-    model[0].set_providers(['CUDAExecutionProvider', 'CPUExecutionProvider'])
-    model[1].set_providers(['CUDAExecutionProvider', 'CPUExecutionProvider'])
+    model_base = rt.InferenceSession(model[0], providers=providers)
+    model_token = rt.InferenceSession(model[1], providers=providers)
     tokenizer = model[2]
+    model = [model_base, model_token, tokenizer]
     bpm = int(bpm)
     if time_sig == "auto":
         time_sig = None
@@ -426,22 +427,18 @@ if __name__ == "__main__":
         ]
     }
     models = {}
-    providers = ['CPUExecutionProvider']
+    providers = ['CUDAExecutionProvider', 'CPUExecutionProvider']
 
     for name, (repo_id, path, config, loras) in models_info.items():
         model_base_path = hf_hub_download_retry(repo_id=repo_id, filename=f"{path}onnx/model_base.onnx")
         model_token_path = hf_hub_download_retry(repo_id=repo_id, filename=f"{path}onnx/model_token.onnx")
-        model_base = rt.InferenceSession(model_base_path, providers=providers)
-        model_token = rt.InferenceSession(model_token_path, providers=providers)
         tokenizer = get_tokenizer(config)
-        models[name] = [model_base, model_token, tokenizer]
+        models[name] = [model_base_path, model_token_path, tokenizer]
         for lora_name, lora_repo in loras.items():
             model_base_path = hf_hub_download_retry(repo_id=lora_repo, filename=f"onnx/model_base.onnx")
             model_token_path = hf_hub_download_retry(repo_id=lora_repo, filename=f"onnx/model_token.onnx")
-            model_base = rt.InferenceSession(model_base_path, providers=providers)
-            model_token = rt.InferenceSession(model_token_path, providers=providers)
             tokenizer = get_tokenizer(config)
-            models[f"{name} with {lora_name} lora"] = [model_base, model_token, tokenizer]
+            models[f"{name} with {lora_name} lora"] = [model_base_path, model_token_path, tokenizer]
 
     load_javascript()
     app = gr.Blocks()