diff --git a/GPT_SoVITS/TTS_infer_pack/TTS.py b/GPT_SoVITS/TTS_infer_pack/TTS.py
index 5cd618e..1571ef7 100644
--- a/GPT_SoVITS/TTS_infer_pack/TTS.py
+++ b/GPT_SoVITS/TTS_infer_pack/TTS.py
@@ -462,8 +462,6 @@ class TTS:
                 n_speakers=self.configs.n_speakers,
                 **kwargs
             )
-            if hasattr(vits_model, "enc_q"):
-                del vits_model.enc_q
             self.configs.is_v3_synthesizer = False
         else:
             vits_model = SynthesizerTrnV3(
@@ -474,7 +472,8 @@ class TTS:
             )
             self.configs.is_v3_synthesizer = True
             self.init_bigvgan()
-            
+            if "pretrained" not in weights_path and hasattr(vits_model, "enc_q"):
+                del vits_model.enc_q
 
         if if_lora_v3==False:
             print(f"Loading VITS weights from {weights_path}. {vits_model.load_state_dict(dict_s2['weight'], strict=False)}")
diff --git a/GPT_SoVITS/inference_webui.py b/GPT_SoVITS/inference_webui.py
index afae2cf..dd9086f 100644
--- a/GPT_SoVITS/inference_webui.py
+++ b/GPT_SoVITS/inference_webui.py
@@ -238,7 +238,7 @@ def change_sovits_weights(sovits_path,prompt_language=None,text_language=None):
         else:
             visible_sample_steps=False
             visible_inp_refs=True
-        yield  {'__type__':'update', 'choices':list(dict_language.keys())}, {'__type__':'update', 'choices':list(dict_language.keys())}, prompt_text_update, prompt_language_update, text_update, text_language_update,{"__type__": "update", "visible": visible_sample_steps},{"__type__": "update", "visible": visible_inp_refs},{"__type__": "update", "value": False,"interactive":True if model_version!="v3"else False},{"__type__": "update", "visible":True if model_version=="v3"else False}
+        yield  {'__type__':'update', 'choices':list(dict_language.keys())}, {'__type__':'update', 'choices':list(dict_language.keys())}, prompt_text_update, prompt_language_update, text_update, text_language_update,{"__type__": "update", "visible": visible_sample_steps,"value":32},{"__type__": "update", "visible": visible_inp_refs},{"__type__": "update", "value": False,"interactive":True if model_version!="v3"else False},{"__type__": "update", "visible":True if model_version=="v3"else False},{"__type__": "update", "value":i18n("模型加载中，请等待"),"interactive":False}
 
     dict_s2 = load_sovits_new(sovits_path)
     hps = dict_s2["config"]
@@ -294,6 +294,7 @@ def change_sovits_weights(sovits_path,prompt_language=None,text_language=None):
         # torch.save(vq_model.state_dict(),"merge_win.pth")
         vq_model.eval()
 
+    yield {'__type__':'update', 'choices':list(dict_language.keys())}, {'__type__':'update', 'choices':list(dict_language.keys())}, prompt_text_update, prompt_language_update, text_update, text_language_update,{"__type__": "update", "visible": visible_sample_steps,"value":32},{"__type__": "update", "visible": visible_inp_refs},{"__type__": "update", "value": False,"interactive":True if model_version!="v3"else False},{"__type__": "update", "visible":True if model_version=="v3"else False},{"__type__": "update", "value":i18n("合成语音"),"interactive":True}
     with open("./weight.json")as f:
         data=f.read()
         data=json.loads(data)
@@ -877,7 +878,7 @@ with gr.Blocks(title="GPT-SoVITS WebUI") as app:
         with gr.Row():
             inp_ref = gr.Audio(label=i18n("请上传3~10秒内参考音频，超过会报错！"), type="filepath", scale=13)
             with gr.Column(scale=13):
-                ref_text_free = gr.Checkbox(label=i18n("开启无参考文本模式。不填参考文本亦相当于开启。")+i18n("v3暂不支持该模式，使用了会报错。"), value=False, interactive=True, show_label=True,scale=1)
+                ref_text_free = gr.Checkbox(label=i18n("开启无参考文本模式。不填参考文本亦相当于开启。")+i18n("v3暂不支持该模式，使用了会报错。"), value=False, interactive=True if model_version!="v3"else False, show_label=True,scale=1)
                 gr.Markdown(html_left(i18n("使用无参考文本模式时建议使用微调的GPT")+"<br>"+i18n("听不清参考音频说的啥(不晓得写啥)可以开。开启后无视填写的参考文本。")))
                 prompt_text = gr.Textbox(label=i18n("参考音频的文本"), value="", lines=5, max_lines=5,scale=1)
             with gr.Column(scale=14):
@@ -915,7 +916,7 @@ with gr.Blocks(title="GPT-SoVITS WebUI") as app:
             #     phoneme=gr.Textbox(label=i18n("音素框"), value="")
             #     get_phoneme_button = gr.Button(i18n("目标文本转音素"), variant="primary")
         with gr.Row():
-            inference_button = gr.Button(i18n("合成语音"), variant="primary", size='lg', scale=25)
+            inference_button = gr.Button(value=i18n("合成语音"), variant="primary", size='lg', scale=25)
             output = gr.Audio(label=i18n("输出的语音"), scale=14)
 
         inference_button.click(
@@ -923,7 +924,7 @@ with gr.Blocks(title="GPT-SoVITS WebUI") as app:
             [inp_ref, prompt_text, prompt_language, text, text_language, how_to_cut, top_k, top_p, temperature, ref_text_free,speed,if_freeze,inp_refs,sample_steps,if_sr_Checkbox,pause_second_slider],
             [output],
         )
-        SoVITS_dropdown.change(change_sovits_weights, [SoVITS_dropdown,prompt_language,text_language], [prompt_language,text_language,prompt_text,prompt_language,text,text_language,sample_steps,inp_refs,ref_text_free,if_sr_Checkbox])
+        SoVITS_dropdown.change(change_sovits_weights, [SoVITS_dropdown,prompt_language,text_language], [prompt_language,text_language,prompt_text,prompt_language,text,text_language,sample_steps,inp_refs,ref_text_free,if_sr_Checkbox,inference_button])
         GPT_dropdown.change(change_gpt_weights, [GPT_dropdown], [])
 
         # gr.Markdown(value=i18n("文本切分工具。太长的文本合成出来效果不一定好，所以太长建议先切。合成会根据文本的换行分开合成再拼起来。"))
diff --git a/GPT_SoVITS/inference_webui_fast.py b/GPT_SoVITS/inference_webui_fast.py
index 9017aa4..40cdae9 100644
--- a/GPT_SoVITS/inference_webui_fast.py
+++ b/GPT_SoVITS/inference_webui_fast.py
@@ -41,12 +41,13 @@ gpt_path = os.environ.get("gpt_path", None)
 sovits_path = os.environ.get("sovits_path", None)
 cnhubert_base_path = os.environ.get("cnhubert_base_path", None)
 bert_path = os.environ.get("bert_path", None)
-version=os.environ.get("version","v2")
+version=model_version=os.environ.get("version","v2")
 
 import gradio as gr
 from TTS_infer_pack.TTS import TTS, TTS_Config, NO_PROMPT_ERROR
 from TTS_infer_pack.text_segmentation_method import get_method
 from tools.i18n.i18n import I18nAuto, scan_language_list
+from inference_webui import DictToAttrRecursive
 
 language=os.environ.get("language","Auto")
 language=sys.argv[-1] if sys.argv[-1] in scan_language_list() else language
@@ -221,19 +222,16 @@ def get_weights_names(GPT_weight_root, SoVITS_weight_root):
 SoVITS_names, GPT_names = get_weights_names(GPT_weight_root, SoVITS_weight_root)
 
 
-from process_ckpt import get_sovits_version_from_path_fast
+from process_ckpt import get_sovits_version_from_path_fast,load_sovits_new
 def change_sovits_weights(sovits_path,prompt_language=None,text_language=None):
-    global version, dict_language
+    global version, model_version, dict_language,if_lora_v3
     version, model_version, if_lora_v3=get_sovits_version_from_path_fast(sovits_path)
-
+    # print(sovits_path,version, model_version, if_lora_v3)
     if if_lora_v3 and not os.path.exists(path_sovits_v3):
         info= path_sovits_v3 + i18n("SoVITS V3 底模缺失，无法加载相应 LoRA 权重")
         gr.Warning(info)
         raise FileExistsError(info)
-
-    tts_pipeline.init_vits_weights(sovits_path)
-
-    dict_language = dict_language_v1 if tts_pipeline.configs.version =='v1' else dict_language_v2
+    dict_language = dict_language_v1 if version =='v1' else dict_language_v2
     if prompt_language is not None and text_language is not None:
         if prompt_language in list(dict_language.keys()):
             prompt_text_update, prompt_language_update = {'__type__':'update'}, {'__type__':'update', 'value':prompt_language}
@@ -251,8 +249,11 @@ def change_sovits_weights(sovits_path,prompt_language=None,text_language=None):
         else:
             visible_sample_steps=False
             visible_inp_refs=True
-        yield  {'__type__':'update', 'choices':list(dict_language.keys())}, {'__type__':'update', 'choices':list(dict_language.keys())}, prompt_text_update, prompt_language_update, text_update, text_language_update,{"__type__": "update", "visible": visible_sample_steps},{"__type__": "update", "visible": visible_inp_refs},{"__type__": "update", "value": False,"interactive":True if model_version!="v3"else False},{"__type__": "update", "visible":True if model_version=="v3"else False}
+        #prompt_language,text_language,prompt_text,prompt_language,text,text_language,inp_refs,ref_text_free,
+        yield  {'__type__':'update', 'choices':list(dict_language.keys())}, {'__type__':'update', 'choices':list(dict_language.keys())}, prompt_text_update, prompt_language_update, text_update, text_language_update,{"__type__": "update", "interactive": visible_sample_steps,"value":32},{"__type__": "update", "visible": visible_inp_refs},{"__type__": "update", "interactive": True if model_version!="v3"else False},{"__type__": "update", "value":i18n("模型加载中，请等待"),"interactive":False}
 
+    tts_pipeline.init_vits_weights(sovits_path)
+    yield {'__type__':'update', 'choices':list(dict_language.keys())}, {'__type__':'update', 'choices':list(dict_language.keys())}, prompt_text_update, prompt_language_update, text_update, text_language_update,{"__type__": "update", "interactive": visible_sample_steps,"value":32},{"__type__": "update", "visible": visible_inp_refs},{"__type__": "update", "interactive": True if model_version!="v3"else False},{"__type__": "update", "value":i18n("合成语音"),"interactive":True}
     with open("./weight.json")as f:
         data=f.read()
         data=json.loads(data)
@@ -279,14 +280,14 @@ with gr.Blocks(title="GPT-SoVITS WebUI") as app:
             gr.Markdown(value=i18n("*请上传并填写参考信息"))
             with gr.Row():
                 inp_ref = gr.Audio(label=i18n("主参考音频(请上传3~10秒内参考音频，超过会报错！)"), type="filepath")
-                inp_refs = gr.File(label=i18n("辅参考音频(可选多个，或不选)"),file_count="multiple")
+                inp_refs = gr.File(label=i18n("辅参考音频(可选多个，或不选)"),file_count="multiple", visible=True if model_version!="v3"else False)
             prompt_text = gr.Textbox(label=i18n("主参考音频的文本"), value="", lines=2)
             with gr.Row():
                 prompt_language = gr.Dropdown(
                     label=i18n("主参考音频的语种"), choices=list(dict_language.keys()), value=i18n("中文")
                 )
                 with gr.Column():
-                    ref_text_free = gr.Checkbox(label=i18n("开启无参考文本模式。不填参考文本亦相当于开启。"), value=False, interactive=True, show_label=True)
+                    ref_text_free = gr.Checkbox(label=i18n("开启无参考文本模式。不填参考文本亦相当于开启。"), value=False, interactive=True if model_version!="v3"else False, show_label=True)
                     gr.Markdown(i18n("使用无参考文本模式时建议使用微调的GPT")+"<br>"+i18n("听不清参考音频说的啥(不晓得写啥)可以开。开启后无视填写的参考文本。"))
 
         with gr.Column():
@@ -355,7 +356,7 @@ with gr.Blocks(title="GPT-SoVITS WebUI") as app:
             [output, seed],
         )
         stop_infer.click(tts_pipeline.stop, [], [])
-        SoVITS_dropdown.change(change_sovits_weights, [SoVITS_dropdown,prompt_language,text_language], [prompt_language,text_language,prompt_text,prompt_language,text,text_language])
+        SoVITS_dropdown.change(change_sovits_weights, [SoVITS_dropdown,prompt_language,text_language], [prompt_language,text_language,prompt_text,prompt_language,text,text_language,sample_steps,inp_refs,ref_text_free,inference_button])#
         GPT_dropdown.change(tts_pipeline.init_t2s_weights, [GPT_dropdown], [])
 
     with gr.Group():
diff --git a/GPT_SoVITS/text/g2pw/onnx_api.py b/GPT_SoVITS/text/g2pw/onnx_api.py
index dcb4604..78a4c93 100644
--- a/GPT_SoVITS/text/g2pw/onnx_api.py
+++ b/GPT_SoVITS/text/g2pw/onnx_api.py
@@ -58,7 +58,7 @@ def download_and_decompress(model_dir: str='G2PWModel/'):
         extract_dir = os.path.join(parent_directory,"G2PWModel_1.1")
         extract_dir_new = os.path.join(parent_directory,"G2PWModel")
         print("Downloading g2pw model...")
-        modelscope_url = "https://paddlespeech.cdn.bcebos.com/Parakeet/released_models/g2p/G2PWModel_1.1.zip"
+        modelscope_url = "https://www.modelscope.cn/models/kamiorinn/g2pw/resolve/master/G2PWModel_1.1.zip"#"https://paddlespeech.cdn.bcebos.com/Parakeet/released_models/g2p/G2PWModel_1.1.zip"
         with requests.get(modelscope_url, stream=True) as r:
             r.raise_for_status()
             with open(zip_dir, 'wb') as f:
diff --git a/docs/cn/Changelog_CN.md b/docs/cn/Changelog_CN.md
index 4666f6e..58a67d3 100644
--- a/docs/cn/Changelog_CN.md
+++ b/docs/cn/Changelog_CN.md
@@ -286,3 +286,17 @@ https://github.com/RVC-Boss/GPT-SoVITS/pull/2112 https://github.com/RVC-Boss/GPT
 修复短文本语种选择出错 https://github.com/RVC-Boss/GPT-SoVITS/pull/2122
 
 修复v3sovits未传参以支持调节语速
+
+### 202503
+
+修复一批由依赖的库版本不对导致的问题https://github.com/RVC-Boss/GPT-SoVITS/commit/6c468583c5566e5fbb4fb805e4cc89c403e997b8
+
+修复模型加载异步逻辑https://github.com/RVC-Boss/GPT-SoVITS/commit/03b662a769946b7a6a8569a354860e8eeeb743aa
+
+修复其他若干bug
+
+重点更新：
+
+1-v3支持并行推理 https://github.com/RVC-Boss/GPT-SoVITS/commit/03b662a769946b7a6a8569a354860e8eeeb743aa
+
+2-整合包修复onnxruntime GPU推理的支持，影响：（1）g2pw有个onnx模型原先是CPU推理现在用GPU，显著降低推理的CPU瓶颈 （2）foxjoy去混响模型现在可使用GPU推理
diff --git a/webui.py b/webui.py
index b73ed89..41955a5 100644
--- a/webui.py
+++ b/webui.py
@@ -298,9 +298,9 @@ def change_tts_inference(bert_path,cnhubert_base_path,gpu_number,gpt_path,sovits
         cmd = '"%s" GPT_SoVITS/inference_webui_fast.py "%s"'%(python_exec, language)
     else:
         cmd = '"%s" GPT_SoVITS/inference_webui.py "%s"'%(python_exec, language)
-    #####v3暂不支持加速推理
-    if version=="v3":
-        cmd = '"%s" GPT_SoVITS/inference_webui.py "%s"'%(python_exec, language)
+    # #####v3暂不支持加速推理
+    # if version=="v3":
+    #     cmd = '"%s" GPT_SoVITS/inference_webui.py "%s"'%(python_exec, language)
     if p_tts_inference is None:
         os.environ["gpt_path"]=gpt_path if "/" in gpt_path else "%s/%s"%(GPT_weight_root,gpt_path)
         os.environ["sovits_path"]=sovits_path if "/"in sovits_path else "%s/%s"%(SoVITS_weight_root,sovits_path)
@@ -849,8 +849,8 @@ def switch_version(version_):
         {'__type__': 'update', "value": default_sovits_save_every_epoch,"maximum": max_sovits_save_every_epoch}, \
         {'__type__': 'update', "visible": True if version!="v3"else False}, \
         {'__type__': 'update', "value": False if not if_force_ckpt else True, "interactive": True if not if_force_ckpt else False}, \
-        {'__type__': 'update', "interactive": False if version == "v3" else True, "value": False}, \
-        {'__type__': 'update', "visible": True if version== "v3" else False}
+        {'__type__': 'update', "interactive": True, "value": False}, \
+        {'__type__': 'update', "visible": True if version== "v3" else False}        # {'__type__': 'update', "interactive": False if version == "v3" else True, "value": False}, \ ####batch infer
 
 if os.path.exists('GPT_SoVITS/text/G2PWModel'):...
 else: