no function

app.py

@@ -5,23 +5,48 @@ from threading import Thread
 import gradio as gr
 import torch
 
-config = PeftConfig.from_pretrained("Junity/Genshin-World-Model", trust_remote_code=True)
-model = AutoModelForCausalLM.from_pretrained("baichuan-inc/Baichuan-13B-Base", torch_dtype=torch.float32, trust_remote_code=True)
-model = PeftModel.from_pretrained(model, r"Junity/Genshin-World-Model", torch_dtype=torch.float32, trust_remote_code=True)
-tokenizer = AutoTokenizer.from_pretrained("baichuan-inc/Baichuan-13B-Base", trust_remote_code=True)
-history = []
-device = torch.device("cuda" if torch.cuda.is_available() else "cpu")
-if device == "cuda":
-    model.cuda()
-    model = model.half()
 
-def respond(role_name, msg, textbox):
+# lora_folder = ''
+# model_folder = ''
+#
+# config = PeftConfig.from_pretrained(("Junity/Genshin-World-Model" if lora_folder == ''
+#                                      else lora_folder),
+#                                     trust_remote_code=True)
+# model = AutoModelForCausalLM.from_pretrained(("baichuan-inc/Baichuan-13B-Base" if model_folder == ''
+#                                               else model_folder), torch_dtype=torch.float32, trust_remote_code=True)
+# model = PeftModel.from_pretrained(model,
+#                                   ("Junity/Genshin-World-Model" if lora_folder == ''
+#                                    else lora_folder)
+#                                   , torch_dtype=torch.float32, trust_remote_code=True)
+# tokenizer = AutoTokenizer.from_pretrained(("baichuan-inc/Baichuan-13B-Base" if model_folder == ''
+#                                            else model_folder),
+#                                           trust_remote_code=True)
+# history = []
+# device = torch.device("cuda" if torch.cuda.is_available() else "cpu")
+# if device == "cuda":
+#     model.cuda()
+#     model = model.half()
+
+
+def respond(role_name, character_name, msg, textbox, temp, rep, max_len, top_p, top_k):
     if textbox != '':
-        textbox = textbox + "\n" + role_name + "：" + msg + ('。' if msg[-1] not in ['。', '！', '？'] else '') + '\n'
+        textbox = (textbox
+                   + "\n"
+                   + role_name
+                   + ("：" if role_name != '' else '')
+                   + msg
+                   + ('。\n' if msg[-1] not in ['。', '！', '？'] else ''))
         yield ["", textbox]
     else:
-        textbox = textbox + role_name + "：" + msg + ('。' if msg[-1] not in ['。', '！', '？'] else '') + '\n'
+        textbox = (textbox
+                   + role_name
+                   + ("：" if role_name != '' else '')
+                   + msg
+                   + ('。' if msg[-1] not in ['。', '！', '？', ')', '}', '：', ':', '('] else '')
+                   + ('\n' if msg[-1] in ['。', '！', '？', ')', '}'] else ''))
         yield ["", textbox]
+    if character_name != '':
+        textbox += ('\n' if textbox[-1] != '\n' else '') + character_name + '：'
     input_ids = tokenizer.encode(textbox)[-4096:]
     input_ids = torch.LongTensor([input_ids]).to(device)
     generation_config = model.generation_config
@@ -48,21 +73,32 @@ def respond(role_name, msg, textbox):
         textbox += new_text
         yield ["", textbox]
 
+
 with gr.Blocks() as demo:
     gr.Markdown(
         """
         ## Genshin-World-Model
         - 模型地址 [https://huggingface.co/Junity/Genshin-World-Model](https://huggingface.co/Junity/Genshin-World-Model)
         - 此模型不支持要求对方回答什么，只支持续写。
+        - 目前运行不了，因为没有钱租卡。
         """
     )
-    with gr.Tab("聊天") as chat:
-        role_name = gr.Textbox(label="你将扮演的角色")
-        msg = gr.Textbox(label="输入")
+    with gr.Tab("创作") as chat:
+        role_name = gr.Textbox(label="你将扮演的角色(可留空)")
+        character_name = gr.Textbox(label="对方的角色(可留空)")
+        msg = gr.Textbox(label="你说的话")
+    with gr.Row():
+        clear = gr.ClearButton()
+        sub = gr.Button("Submit", variant="primary")
     with gr.Row():
-        clear = gr.Button("Clear")
-        sub = gr.Button("Submit")
-    textbox = gr.Textbox(interactive=False)
-    sub.click(fn=respond, inputs=[role_name, msg, textbox], outputs=[msg, textbox])
-    clear.click(lambda: None, None, textbox, queue=False)
-    demo.queue().launch(server_port=6006)
+        temp = gr.Slider(minimum=0, maximum=2.0, step=0.1, value=1.5, label="温度(调大则更随机)", interactive=True)
+        rep = gr.Slider(minimum=0, maximum=2.0, step=0.1, value=1.0, label="对重复生成的惩罚", interactive=True)
+        max_len = gr.Slider(minimum=4, maximum=512, step=4, value=256, label="对方回答的最大长度", interactive=True)
+        top_p = gr.Slider(minimum=0, maximum=1.0, step=0.1, value=0.7, label="Top-p(调大则更随机)", interactive=True)
+        top_k = gr.Slider(minimum=0, maximum=100, step=1, value=50, label="Top-k(调大则更随机)", interactive=True)
+    textbox = gr.Textbox(interactive=True, label="全部文本(可修改)")
+    clear.add([msg, role_name, textbox])
+    sub.click(fn=respond,
+              inputs=[role_name, character_name, msg, textbox, temp, rep, max_len, top_p, top_k],
+              outputs=[msg, textbox])
+    demo.queue().launch(server_port=6006, share=True)