File size: 5,419 Bytes
ee55b6f
a7ada72
 
 
 
 
ee55b6f
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
0fbec2e
a7ada72
 
0fbec2e
f5e3afe
ee55b6f
0fbec2e
ee55b6f
 
 
 
 
 
 
 
4ee6037
 
ee55b6f
a7ada72
ee55b6f
 
 
 
a7ada72
 
ee55b6f
9e7fc2a
ee55b6f
 
 
 
 
 
0fbec2e
ee55b6f
 
 
 
 
 
 
 
 
 
 
 
0fbec2e
ee55b6f
fedc021
0fbec2e
 
fedc021
 
0fbec2e
ee55b6f
0fbec2e
ee55b6f
 
0fbec2e
 
 
 
 
b96edb4
9cd225b
a527d95
9cd225b
 
 
 
 
 
 
 
 
 
 
 
ff5c7cf
0fbec2e
 
 
 
 
 
 
 
 
 
 
 
 
 
4ee6037
 
ee55b6f
0fbec2e
 
ee55b6f
4ee6037
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
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
import os, zipfile, shutil, glob
import gradio as gr
from langchain_community.vectorstores import FAISS
from langchain_community.embeddings import HuggingFaceEmbeddings
from transformers import AutoTokenizer, AutoModelForCausalLM, pipeline
import torch
import langchain

ZIP_NAME = "solo_leveling_faiss_ko.zip"
TARGET_DIR = "solo_leveling_faiss_ko"

def ensure_faiss_dir() -> str:
    if os.path.exists(os.path.join(TARGET_DIR, "index.faiss")) and \
       os.path.exists(os.path.join(TARGET_DIR, "index.pkl")):
        return TARGET_DIR
    if os.path.exists("index.faiss") and os.path.exists("index.pkl"):
        os.makedirs(TARGET_DIR, exist_ok=True)
        if not os.path.exists(os.path.join(TARGET_DIR, "index.faiss")):
            shutil.move("index.faiss", os.path.join(TARGET_DIR, "index.faiss"))
        if not os.path.exists(os.path.join(TARGET_DIR, "index.pkl")):
            shutil.move("index.pkl", os.path.join(TARGET_DIR, "index.pkl"))
        return TARGET_DIR
    if os.path.exists(ZIP_NAME):
        with zipfile.ZipFile(ZIP_NAME, 'r') as z:
            z.extractall(".")
        if os.path.exists(os.path.join(TARGET_DIR, "index.faiss")) and \
           os.path.exists(os.path.join(TARGET_DIR, "index.pkl")):
            return TARGET_DIR
        faiss_cand = glob.glob("**/index.faiss", recursive=True)
        pkl_cand   = glob.glob("**/index.pkl",   recursive=True)
        if faiss_cand and pkl_cand:
            os.makedirs(TARGET_DIR, exist_ok=True)
            shutil.copy2(faiss_cand[0], os.path.join(TARGET_DIR, "index.faiss"))
            shutil.copy2(pkl_cand[0],   os.path.join(TARGET_DIR, "index.pkl"))
            return TARGET_DIR
    raise FileNotFoundError("FAISS index files not found (index.faiss / index.pkl).")

# 0) FAISS ์ธ๋ฑ์Šค ์œ„์น˜ ํ™•๋ณด
base_dir = ensure_faiss_dir()

# 1) ๋ฒกํ„ฐ DB
embeddings = HuggingFaceEmbeddings(model_name="jhgan/ko-sroberta-multitask")
vectorstore = FAISS.load_local(base_dir, embeddings, allow_dangerous_deserialization=True)

# 2) ๋ชจ๋ธ ๋กœ๋”ฉ (CPU)
model_name = "kakaocorp/kanana-nano-2.1b-instruct"
tokenizer = AutoTokenizer.from_pretrained(model_name)
model = AutoModelForCausalLM.from_pretrained(model_name, torch_dtype=torch.float32, device_map=None)

# 3) ํ…์ŠคํŠธ ์ƒ์„ฑ ํŒŒ์ดํ”„๋ผ์ธ
lm = pipeline(
    "text-generation",
    model=model,
    tokenizer=tokenizer,
    max_new_tokens=100,
    temperature=0.6,
    do_sample=True,
    top_p=0.9,
    return_full_text=False
)

# ์„ ํƒ์ง€
choices = [
    "1: ํ™ฉ๋™์„ ๋ฌด๋ฆฌ๋ฅผ ๋ชจ๋‘ ์ฒ˜์น˜ํ•œ๋‹ค.",
    "2: ์ง„ํ˜ธ๋ฅผ ํฌํ•จํ•œ ํ™ฉ๋™์„ ๋ฌด๋ฆฌ๋ฅผ ๋ชจ๋‘ ์ฒ˜์น˜ํ•œ๋‹ค.",
    "3: ์ „๋ถ€ ๊ธฐ์ ˆ ์‹œํ‚ค๊ณ  ์‚ด๋ ค๋‘”๋‹ค.",
    "4: ์‹œ์Šคํ…œ์„ ๊ฑฐ๋ถ€ํ•˜๊ณ  ๊ทธ๋ƒฅ ๋„๋ง์นœ๋‹ค."
]

def rag_answer(message, history):
    try:
        user_idx = int(message.strip()) - 1
        user_choice = choices[user_idx]
    except:
        return "โ—์˜ฌ๋ฐ”๋ฅธ ๋ฒˆํ˜ธ๋ฅผ ์ž…๋ ฅํ•ด์ฃผ์„ธ์š”. (์˜ˆ: 1 ~ 4)"
    docs = vectorstore.similarity_search(user_choice, k=3)
    context = "\n".join([doc.page_content for doc in docs])
    prompt = f"""๋‹น์‹ ์€ ์›นํˆฐ '๋‚˜ ํ˜ผ์ž๋งŒ ๋ ˆ๋ฒจ์—…'์˜ ์„ฑ์ง„์šฐ์ž…๋‹ˆ๋‹ค.
ํ˜„์žฌ ์ƒํ™ฉ:
{context}
์‚ฌ์šฉ์ž ์„ ํƒ: {user_choice}
์„ฑ์ง„์šฐ์˜ ๋งํˆฌ๋กœ ๊ฐ„๊ฒฐํ•˜๊ณ  ์ž์—ฐ์Šค๋Ÿฌ์šด ๋Œ€์‚ฌ๋ฅผ 1~2๋ฌธ์žฅ ์ƒ์„ฑํ•˜์„ธ์š”.
์ค‘๋ณต๋œ ๋‚ด์šฉ์ด๋‚˜ ๋น„์Šทํ•œ ๋ฌธ์žฅ์€ ๋งŒ๋“ค์ง€ ๋งˆ์„ธ์š”.
"""
    response = lm(prompt)[0]["generated_text"]
    only_dialogue = response.strip().split("\n")[-1]
    if not only_dialogue.startswith("๋Œ€์‚ฌ:"):
        only_dialogue = "๋Œ€์‚ฌ: " + only_dialogue
    return only_dialogue

# ===== UI (๋ณ€๊ฒฝ ์ง€์ ) =====
css_code = """
.quest-title {
  display:flex; align-items:center; gap:10px;
  font-weight:700; font-size:22px; margin-bottom:6px;
}
.quest-title img {
  width:72px; height:auto; opacity:.95;
}
.quest-desc { line-height:1.5; margin-bottom:14px; }
"""

header_html = """
<div class="quest-title">
  [๊ธด๊ธ‰ ํ€˜์ŠคํŠธ: ์ ์„ ์ฒ˜์น˜ํ•˜๋ผ!]
</div>
<div class="quest-desc">
  
<div style="display: flex; align-items: flex-start; gap: 20px;">
   
        <div class="quest-desc">
            'ํ”Œ๋ ˆ์ด์–ด'์—๊ฒŒ ์‚ด์˜๋ฅผ ๊ฐ€์ง„ ์ด๋“ค์ด ์ฃผ๋ณ€์— ์žˆ์Šต๋‹ˆ๋‹ค...
            <!-- ๋‚˜๋จธ์ง€ ํ…์ŠคํŠธ -->
        </div>
    </div>
    <div>
        <img src="https://huggingface.co/spaces/min24ss/r-story-selection/resolve/main/system.png"
             alt="quest"
             style="width: 250px; height: auto;">
    </div>
</div>
     

  'ํ”Œ๋ ˆ์ด์–ด'์—๊ฒŒ ์‚ด์˜๋ฅผ ๊ฐ€์ง„ ์ด๋“ค์ด ์ฃผ์œ„์— ์žˆ์Šต๋‹ˆ๋‹ค. ์ด๋“ค์„ ๋ชจ๋‘ ์ฒ˜์น˜ํ•˜์—ฌ ์•ˆ์ „์„ ํ™•๋ณดํ•˜์‹ญ์‹œ์˜ค.<br>
  ์ง€์‹œ์— ๋”ฐ๋ฅด์ง€ ์•Š์œผ๋ฉด ๋‹น์‹ ์˜ ์‹ฌ์žฅ์€ ์ •์ง€(!)ํ•˜๊ฒŒ ๋ฉ๋‹ˆ๋‹ค.<br>
  ์ฒ˜์น˜ํ•ด์•ผ ํ•  ์ ์˜ ์ˆซ์ž: 8๋ช… / ์ฒ˜์น˜ํ•œ ์ ์˜ ์ˆซ์ž: 0๋ช…<br><br>
  ๐Ÿ’ฌ ์„ ํƒ์ง€๋ฅผ ์ž…๋ ฅํ•˜์„ธ์š”:<br>
  1: ํ™ฉ๋™์„ ๋ฌด๋ฆฌ๋ฅผ ๋ชจ๋‘ ์ฒ˜์น˜ํ•œ๋‹ค.<br>
  2: ํ™ฉ๋™์„ ๋ฌด๋ฆฌ์™€ ์ง„ํ˜ธ๋ฅผ ํฌํ•จํ•˜์—ฌ ๋ชจ๋‘ ์ฒ˜์น˜ํ•œ๋‹ค.<br>
  3: ์ „๋ถ€ ๊ธฐ์ ˆ ์‹œํ‚ค๊ณ  ์‚ด๋ ค๋‘”๋‹ค.<br>
  4: ์‹œ์Šคํ…œ์„ ๊ฑฐ๋ถ€ํ•˜๊ณ  ๊ทธ๋ƒฅ ๋„๋ง์นœ๋‹ค.
</div>
"""

with gr.Blocks(css=css_code) as demo:
    gr.HTML(header_html)  # โ† ์—ฌ๊ธฐ์„œ HTML ๊ทธ๋Œ€๋กœ ๋ Œ๋”๋ง (์ด๋ฏธ์ง€ ๋ณด์žฅ)
    gr.ChatInterface(fn=rag_answer)  # title/description์€ ์“ฐ์ง€ ์•Š์Œ

if __name__ == "__main__":
    print("Torch:", torch.__version__)
    import transformers as _t
    print("Transformers:", _t.__version__)
    print("LangChain:", langchain.__version__)
    demo.launch()