Update app.py

只保留7B模型，否则会超过内存限制

app.py

@@ -3,34 +3,25 @@ import torch
 from transformers import AutoTokenizer, AutoModelForCausalLM
 
 # 设置设备，如果有GPU则使用GPU
+# 在免费CPU服务器上，这里会是 'cpu'
 device = "cuda" if torch.cuda.is_available() else "cpu"
+print(f"Using device: {device}")
 
-# 加载基础模型和分词器
-tokenizer_base = AutoTokenizer.from_pretrained("tencent/Hunyuan-MT-7B", trust_remote_code=True)
-model_base = AutoModelForCausalLM.from_pretrained("tencent/Hunyuan-MT-7B", trust_remote_code=True).to(device)
+# --- 修改部分开始 ---
+# 只加载Chimera集成模型和分词器
+print("Loading Hunyuan-MT-Chimera-7B model and tokenizer...")
+tokenizer = AutoTokenizer.from_pretrained("tencent/Hunyuan-MT-Chimera-7B", trust_remote_code=True)
+model = AutoModelForCausalLM.from_pretrained("tencent/Hunyuan-MT-Chimera-7B", trust_remote_code=True).to(device)
+print("Model loaded successfully.")
+# --- 修改部分结束 ---
 
-# 加载Chimera集成模型和分词器
-tokenizer_chimera = AutoTokenizer.from_pretrained("tencent/Hunyuan-MT-Chimera-7B", trust_remote_code=True)
-model_chimera = AutoModelForCausalLM.from_pretrained("tencent/Hunyuan-MT-Chimera-7B", trust_remote_code=True).to(device)
 
-
-def translate(model_choice, text_to_translate, source_lang, target_lang):
+def translate(text_to_translate, source_lang, target_lang):
     """
-    根据选择的模型进行翻译
+    使用已加载的Chimera模型进行翻译
     """
-    if model_choice == "Hunyuan-MT-7B (基础版)":
-        tokenizer = tokenizer_base
-        model = model_base
-        # 基础版Prompt模板
-        prompt = f"Translate the following text from {source_lang} to {target_lang}:\n{text_to_translate}"
-    else: # Chimera-7B
-        tokenizer = tokenizer_chimera
-        model = model_chimera
-        # Chimera版需要一个特殊的、包含候选翻译的模板，这里我们简化一下，
-        # 实际应用中会先用基础模型生成多个候选。为简化Demo，我们直接套用基础模板。
-        # 官方的Chimera用法更复杂，需要输入多个候选翻译进行精炼。
-        prompt = f"Translate the following text from {source_lang} to {target_lang}:\n{text_to_translate}"
-
+    prompt = f"Translate the following text from {source_lang} to {target_lang}:\n{text_to_translate}"
+    
     inputs = tokenizer(prompt, return_tensors="pt").to(device)
     
     # 生成翻译结果
@@ -47,14 +38,7 @@ def translate(model_choice, text_to_translate, source_lang, target_lang):
 # --- 创建Gradio界面 ---
 with gr.Blocks() as demo:
     gr.Markdown("# 腾讯混元翻译模型体验Demo")
-    gr.Markdown("选择一个模型，输入源语言、目标语言和待翻译的文本。")
-    
-    with gr.Row():
-        model_selector = gr.Radio(
-            ["Hunyuan-MT-7B (基础版)", "Hunyuan-MT-Chimera-7B (集成优化版)"], 
-            label="选择模型", 
-            value="Hunyuan-MT-7B (基础版)"
-        )
+    gr.Markdown("模型: Hunyuan-MT-Chimera-7B (集成优化版)")
     
     with gr.Row():
         source_language = gr.Textbox(label="源语言", value="Chinese")
@@ -67,7 +51,7 @@ with gr.Blocks() as demo:
     
     translate_button.click(
         fn=translate,
-        inputs=[model_selector, input_text, source_language, target_language],
+        inputs=[input_text, source_language, target_language],
         outputs=output_text
     )