File size: 2,408 Bytes
dc13b6d
 
 
 
 
 
da40aec
dc13b6d
 
 
 
 
 
 
 
 
 
 
 
 
8722708
dc13b6d
 
 
 
 
da40aec
 
 
 
 
 
 
 
 
 
 
8722708
da40aec
dc13b6d
 
 
 
 
 
da40aec
 
 
 
dc13b6d
da40aec
 
 
dc13b6d
 
da40aec
 
 
 
 
 
 
 
 
dc13b6d
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
# app.py
import spaces
import gradio as gr
from functools import lru_cache
from transformers import AutoModelForCausalLM, AutoTokenizer, pipeline

# 可選模型列表
MODEL_LIST = [
    "ckiplab/gpt2-tiny-chinese",
    "ckiplab/gpt2-base-chinese",
    "liswei/Taiwan-ELM-270M-Instruct",
    "liswei/Taiwan-ELM-1_1B",
    "google/gemma-3-1b-pt",
    "benchang1110/Qwen2.5-Taiwan-1.5B-Instruct",
    "benchang1110/Taiwan-tinyllama-v1.0-base",
]

@lru_cache(maxsize=None)
def get_pipeline(model_name):
    tok = AutoTokenizer.from_pretrained(model_name)
    mdl = AutoModelForCausalLM.from_pretrained(model_name, weights_only=False)
    mdl.to("cuda")
    return pipeline("text-generation", model=mdl, tokenizer=tok, device=0)

@spaces.GPU
def suggest_next(text, model_name, k, m):
    """
    使用 Beam Search 產生 M 條最可能的下段建議(每條最多 K 個新詞元)。
    """
    gen_pipe = get_pipeline(model_name)
    outs = gen_pipe(
        text,
        max_new_tokens=k,
        num_beams=m,
        num_return_sequences=m,
        do_sample=False,
        early_stopping=True
    )
    # 只取掉 prompt 的部份
    return [out["generated_text"][len(text):] for out in outs]

def append_suggestion(current, choice):
    return current + choice

with gr.Blocks() as demo:
    gr.Markdown(
        "## 🇹🇼 台灣中文下段預測\n"
        "結合小型語言模型與 ZeroGPU,提供 Beam Search 風格的多條下段建議。"
    )

    input_text = gr.TextArea(
        label="輸入文字", lines=4, placeholder="請在此輸入起始片段…"
    )

    with gr.Row():
        model_selector = gr.Dropdown(
            MODEL_LIST, value=MODEL_LIST[0], label="選擇模型"
        )
        k_slider = gr.Slider(
            minimum=1, maximum=50, step=1, value=5, label="K(最大新生成詞元)"
        )
        m_slider = gr.Slider(
            minimum=1, maximum=10, step=1, value=5, label="M(建議數量 / Beam 數)"
        )

    suggestions = gr.Dropdown([], label="建議清單", interactive=True)
    gpu_button = gr.Button("使用 GPU 生成建議")

    gpu_button.click(
        fn=suggest_next,
        inputs=[input_text, model_selector, k_slider, m_slider],
        outputs=suggestions,
    )
    suggestions.change(
        fn=append_suggestion,
        inputs=[input_text, suggestions],
        outputs=input_text,
    )

    demo.launch()