app.py

import os
import gradio as gr
from transformers import AutoTokenizer, AutoModelForCausalLM, TextGenerationPipeline

# 使用你在 Spaces 中添加的 Secret
hf_token = os.environ.get("HUGGINGFACE_TOKEN")

# 加载 tokenizer 和模型（带token）
model_id = "google/gemma-3-4b-it"
tokenizer = AutoTokenizer.from_pretrained(model_id, token=hf_token)
model = AutoModelForCausalLM.from_pretrained(model_id, token=hf_token, device_map="auto")

# 构造推理管道
pipe = TextGenerationPipeline(model=model, tokenizer=tokenizer)

# 推理函数
def generate(prompt):
    output = pipe(prompt, max_new_tokens=100, do_sample=True, temperature=0.7)
    return output[0]["generated_text"]

# Gradio UI
gr.Interface(
    fn=generate,
    inputs=gr.Text(label="Enter your prompt"),
    outputs=gr.Textbox(label="Generated Text"),
    title="Gemma-3-27B Text Generation"
).launch()