Update app.py

app.py

@@ -1,74 +1,57 @@
-# 导入必要的库
-import gradio as gr
-from datasets import load_dataset
-import numpy as np
-from sentence_transformers import SentenceTransformer, util
+# 导入库
+from transformers import VisionEncoderDecoderModel, ViTFeatureExtractor, AutoTokenizer
+from PIL import Image
+import torch
 from transformers import pipeline
-
-# 安装依赖（在Hugging Face Spaces中可省略，若空间环境未预装相关库可保留）
-#!pip install datasets sentence-transformers transformers torch
-
-# 加载数据集
-dataset = load_dataset("Pradeep016/career-guidance-qa-dataset", split="train")
-# 过滤无效数据（确保question和answer非空）
-dataset = dataset.filter(lambda x: x["question"] and x["answer"])
-
-# 构建职位知识库（职位名称 + 问题-答案对）
-def build_knowledge_base(dataset):
-    knowledge_base = []
-    for item in dataset:
-        role = item["role"]
-        question = item["question"]
-        answer = item["answer"]
-        # 合并职位名称与问题，增强语义关联
-        entry = f"{role} | {question}: {answer}"
-        knowledge_base.append(entry)
-    return knowledge_base
-
-knowledge_base = build_knowledge_base(dataset)
-
-# 初始化语义搜索模型
-embedder = SentenceTransformer("sentence-transformers/all-MiniLM-L6-v2")
-# 预计算知识库嵌入向量
-knowledge_embeddings = embedder.encode(knowledge_base, convert_to_tensor=True)
-
-# 智能问答函数
-def career_qa(user_input):
-    # 1. 语义搜索匹配相关职位
-    input_embedding = embedder.encode(user_input, convert_to_tensor=True)
-    # 计算余弦相似度
-    cos_scores = util.cos_sim(input_embedding, knowledge_embeddings)[0]
-    # 取前3个最相关条目
-    top_indices = np.argsort(cos_scores)[-3:][::-1]
-    top_matches = [knowledge_base[idx] for idx in top_indices]
+import requests
+from io import BytesIO
+
+# 1. 图像标题生成（使用指定模型）
+def generate_caption(image_url):
+    model_name = "bipin/image-caption-generator"
+    model = VisionEncoderDecoderModel.from_pretrained(model_name)
+    feature_extractor = ViTFeatureExtractor.from_pretrained(model_name)
+    tokenizer = AutoTokenizer.from_pretrained("gpt2")
+    device = torch.device("cuda" if torch.cuda.is_available() else "cpu")
+    model.to(device)
     
-    # 2. 从匹配条目中提取答案
-    qa_pipeline = pipeline("question-answering", model="distilbert-base-uncased-finetuned-squad2")
-    results = []
-    for match in top_matches:
-        role = match.split(" | ")[0]
-        context = match.split(" | ")[1]
-        # 固定问题为“请介绍这个职位”
-        result = qa_pipeline(question="请介绍这个职位", context=context)
-        results.append({
-            "职位名称": role,
-            "简介": result["answer"],
-            "置信度": result["score"]
-        })
-    return results
-
-# Gradio界面定义
-def demo(user_input):
-    results = career_qa(user_input)
-    output = "\n".join([f"📌 {res['职位名称']}\n{res['简介']}\n" for res in results])
-    return output
-
-iface = gr.Interface(
-    fn=demo,
-    inputs=gr.Textbox(label="输入职业关键词（如：零售经理）"),
-    outputs=gr.Textbox(label="职位介绍"),
-    title="职业咨询智能问答",
-)
-
-if __name__ == "__main__":
-    iface.launch()
+    # 下载并预处理图像
+    response = requests.get(image_url)
+    img = Image.open(BytesIO(response.content)).convert("RGB")
+    pixel_values = feature_extractor(images=[img], return_tensors="pt").pixel_values.to(device)
+    
+    # 生成标题（限制50字内）
+    output_ids = model.generate(pixel_values, num_beams=4, max_length=50)
+    caption = tokenizer.decode(output_ids[0], skip_special_tokens=True)
+    return caption
+
+# 2. 标题扩写为宣传文案（使用文本生成模型）
+def expand_to_copy(caption):
+    generator = pipeline("text-generation", model="gpt2", max_length=200)
+    prompt = f"根据以下图片标题生成宣传文案：{caption}\n要求：生动形象，突出产品优势，适合社交媒体传播。"
+    copy = generator(prompt, num_return_sequences=1)[0]['generated_text']
+    return copy.strip()
+
+# 3. 文本转语音（使用TTS模型）
+def text_to_speech(text, output_file="output.mp3"):
+    tts = pipeline("text-to-speech", model="facebook/t5-small")
+    speech = tts(text)
+    with open(output_file, "wb") as f:
+        f.write(speech["audio"])
+    return output_file
+
+# 主函数
+def marketing_pipeline(image_url):
+    # 生成标题
+    caption = generate_caption(image_url)
+    print(f"生成标题：{caption}")
+    
+    # 扩写文案
+    copy = expand_to_copy(caption)
+    print(f"宣传文案：\n{copy}")
+    
+    # 生成语音
+    audio_file = text_to_speech(copy)
+    print(f"语音文件已保存：{audio_file}")
+    
+    return caption, copy, audio_file