File size: 9,879 Bytes
53101ae
 
71c5946
53101ae
 
13264cd
 
 
 
 
 
 
 
 
 
 
 
 
 
71c5946
53101ae
71c5946
585d66a
f06be4b
53101ae
f06be4b
53101ae
 
f06be4b
53101ae
71c5946
 
13264cd
71c5946
 
 
 
 
53101ae
71c5946
13264cd
 
 
 
 
b6bb557
71c5946
13264cd
71c5946
 
 
 
 
13264cd
 
 
 
 
b6bb557
71c5946
 
13264cd
 
 
 
 
b6bb557
13264cd
71c5946
 
 
 
 
 
13264cd
 
 
 
 
 
 
 
 
71c5946
 
 
13264cd
 
 
 
 
71c5946
 
13264cd
 
 
 
 
71c5946
13264cd
71c5946
13264cd
 
 
 
 
 
 
 
 
 
 
 
4c8bad5
13264cd
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1910bd7
13264cd
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
71c5946
 
 
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
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
# Qwen2.5-0.5B + QLoRA (private adapter OK) — Gradio UI + config sanitize
import os, json, threading, torch, gradio as gr
from transformers import AutoTokenizer, AutoModelForCausalLM, TextIteratorStreamer
from huggingface_hub import hf_hub_download
from peft import PeftModel, LoraConfig, get_peft_model
import math
from typing import List, Optional
from fastapi import UploadFile, File, Form
from pydantic import BaseModel
from io import BytesIO

try:
    from pypdf import PdfReader
except:
    PdfReader = None
try:
    import docx
except:
    docx = None

# ---------- Env ----------
BASE_MODEL = os.environ.get("BASE_MODEL", "Qwen/Qwen2.5-0.5B-Instruct")
ADAPTER_REPO_ID = os.environ.get("ADAPTER_REPO_ID")
if not ADAPTER_REPO_ID:
    raise ValueError("ADAPTER_REPO_ID belum di-set di Space secrets.")
HF_TOKEN = os.environ.get("HF_TOKEN") or os.environ.get("HUGGINGFACEHUB_API_TOKEN")
AUTH_TFM = {"token": HF_TOKEN} if HF_TOKEN else {}
AUTH_PEFT = {"use_auth_token": HF_TOKEN} if HF_TOKEN else {}
ADAPTER_REVISION = os.environ.get("ADAPTER_REVISION", "main")
SYSTEM_PROMPT = os.environ.get("SYSTEM_PROMPT", "You are a concise, helpful cybersecurity assistant.")
MERGE_LORA = os.environ.get("MERGE_LORA", "0") == "1"

# ---------- Cache local ----------
HF_HOME = os.path.join(os.getcwd(), ".hfhome")
os.environ.setdefault("HF_HOME", HF_HOME)
os.environ.setdefault("TRANSFORMERS_CACHE", os.path.join(HF_HOME, "transformers"))
os.environ.setdefault("HF_HUB_DISABLE_SYMLINKS_WARNING", "1")

# ---------- Load base ----------
tokenizer = AutoTokenizer.from_pretrained(
    BASE_MODEL, 
    use_fast=True, 
    trust_remote_code=True, 
    padding_side="left", 
    cache_dir=os.environ["TRANSFORMERS_CACHE"],
    **AUTH_TFM
)

if tokenizer.pad_token is None:
    tokenizer.pad_token = tokenizer.eos_token

dtype = torch.bfloat16 if torch.cuda.is_available() else torch.float32
base = AutoModelForCausalLM.from_pretrained(
    BASE_MODEL,
    torch_dtype=dtype,
    device_map="auto",
    trust_remote_code=True,
    cache_dir=os.environ["TRANSFORMERS_CACHE"],
    **AUTH_TFM
)

model = PeftModel.from_pretrained(
    base,
    ADAPTER_REPO_ID,
    torch_dtype=dtype,
    cache_dir=os.environ["TRANSFORMERS_CACHE"],
    **AUTH_PEFT
)

if MERGE_LORA:
    model = model.merge_and_unload()

model.eval()

# ---------- Chat Logic ----------
def chat_generate_to_str(message, history=None, max_new_tokens=512):
    history = history or []
    last = ""
    for partial in chat_generate(message, history, max_new_tokens=max_new_tokens):
        last = partial
    return last.strip()
    
def chat_generate(message, history, max_new_tokens):
    messages = []
    if SYSTEM_PROMPT:
        messages.append({"role": "system", "content": SYSTEM_PROMPT})
    for user_msg, bot_msg in history:
        if user_msg:
            messages.append({"role": "user", "content": user_msg})
        if bot_msg:
            messages.append({"role": "assistant", "content": bot_msg})
    messages.append({"role": "user", "content": message})

    prompt_text = tokenizer.apply_chat_template(
        messages,
        tokenize=False,
        add_generation_prompt=True,
    )

    inputs = tokenizer(prompt_text, return_tensors="pt").to(model.device)

    streamer = TextIteratorStreamer(
        tokenizer, skip_prompt=True, skip_special_tokens=True
    )
    gen_kwargs = dict(
        **inputs,
        streamer=streamer,
        max_new_tokens= max_new_tokens,
        temperature=0.7,
        top_p=0.9,
        do_sample=True,
        repetition_penalty=1.1,
    )

    thread = threading.Thread(target=model.generate, kwargs=gen_kwargs)
    thread.start()

    partial_text = ""
    for new_text in streamer:
        partial_text += new_text
        yield partial_text

# ---------- Summarization logic ----------
def model_ctx_len() -> int:
    m = getattr(tokenizer, "model_max_length", 4096)
    try:
        return 4096 if m is None or m > 10_000_000 else int(m)
    except:
        return 4096

def smart_chunks(text: str, target_tokens: int = 1024) -> List[str]:
    approx_chars = target_tokens * 4
    parts = []
    start = 0
    n = len(text)
    while start < n:
        end = min(n, start + approx_chars)
        cut = text.rfind("\n", start, end)
        if cut == -1: cut = text.rfind(". ", start, end)
        if cut == -1 or cut <= start: cut = end
        parts.append(text[start:cut].strip())
        start = cut
    return [p for p in parts if p]

def summarize_text(text: str, bullets: bool = True, max_new_tokens: int = 256) -> str:
    ctx = model_ctx_len()
    chunk_tokens = max(256, min(1024, ctx // 3))
    chunks = smart_chunks(text, target_tokens=chunk_tokens)

    partial = []
    for i, c in enumerate(chunks, 1):
        instr = (
            "Summary this document. Focust on key points, entities, numbers, and conclusions. "
            "Give Output 5-8 Point \n\n"
            f"=== Section {i}/{len(chunks)} ===\n{c}"
        )
        summary = chat_generate_to_str(instr, history=[], max_new_tokens=max_new_tokens)
        partial.append(summary)

    join = "\n\n".join(f"- {s}" for s in partial)
    final_instr = (
        "Assemble a comprehensive, structured, and concise summary from the partial summaries below.\n "
        "Give, what is the most important information from the document.\n"
        + ("\nOutput at points" if bullets else "")
        + "\n\nPartial Summary:\n" + join
    )
    final_summary = chat_generate_to_str(final_instr, history=[], max_new_tokens=max_new_tokens)
    return final_summary

def read_pdf(file: UploadFile) -> str:
    if PdfReader is None:
        raise RuntimeError("pypdf not installed. pip install pypdf")
    with open(file, "rb") as f:
        pdf = PdfReader(f)
        texts = [(page.extract_text() or "") for page in pdf.pages]
    return "\n".join(texts)

def read_docx(file: UploadFile) -> str:
    if docx is None:
        raise RuntimeError("python-docx not installed. pip install python-docx")
    d = docx.Document(file)
    return "\n".join(p.text for p in d.paragraphs)

def summarize_file(filepath, bullets=True, max_new_tokens=256):
    if not filepath:
        return "Please upload a file."
    low = filepath.lower()
    if low.endswith(".pdf"):
        content = read_pdf(filepath)
    elif low.endswith(".docx"):
        content = read_docx(filepath)
    elif low.endswith(".txt"):
        with open(filepath, "r", encoding="utf-8", errors="ignore") as f:
            content = f.read()
    else:
        return "Unsupported file type. Use .pdf, .docx, or .txt"

    if not content or content.startswith("[Error]"):
        return content or "No readable content."
    return summarize_text(content, bullets=bullets,  max_new_tokens=max_new_tokens)

# ---------- UI ----------
with gr.Blocks(title="LLM Toolkit: Chat & Summaries", fill_height=True) as demo:
    gr.Markdown("## 🔧 LLM Toolkit — Chat, Summary Text, Summary File")

    with gr.Row():
        with gr.Column(scale=1):
            gr.Markdown("### ⚙️ Settings")
            sl_max_new = gr.Slider(32, 2048, value=512, step=32, label="Max New Tokens (Chat)")

            gr.Markdown("#### Summarization")
            sl_max_new_sum = gr.Slider(32, 1024, value=256, step=32, label="Max New Tokens (Summary)")
            ck_bullets = gr.Checkbox(value=True, label="Bullet points output")

        with gr.Column(scale=3):
            with gr.Tabs():
                # ---- Chat Tab ----
                with gr.Tab("💬 Chat"):
                    chat = gr.Chatbot(height=420, show_copy_button=True)
                    chat_state = gr.State([])
                    chat_input = gr.Textbox(placeholder="Type your prompt…", label="Your message", lines=2)
                    with gr.Row():
                        btn_send = gr.Button("Send", variant="primary")
                        btn_clear = gr.Button("Clear")

                    def _on_send(msg, history, m):
                        stream = chat_generate(msg, history or [],  max_new_tokens=m)
                        partial = ""
                        for chunk in stream:
                            partial = chunk
                            yield history + [(msg, partial)]
                        # finalize
                        yield history + [(msg, partial)]

                    btn_send.click(
                        _on_send,
                        inputs=[chat_input, chat_state, sl_max_new],
                        outputs=chat
                    ).then(lambda h: h, chat, chat_state).then(lambda: "", None, chat_input)

                    btn_clear.click(lambda: ([], []), None, [chat, chat_state])

                # ---- Summary Text Tab ----
                with gr.Tab("📝 Summary Text"):
                    txt_input = gr.Textbox(lines=14, label="Paste text here")
                    btn_sum_text = gr.Button("Summarize", variant="primary")
                    txt_out = gr.Markdown()

                    btn_sum_text.click(
                        summarize_text,
                        inputs=[txt_input, ck_bullets,  sl_max_new_sum],
                        outputs=txt_out
                    )

                # ---- Summary File Tab ----
                with gr.Tab("📄 Summary File"):
                    file_input = gr.File(
                        label="Upload .pdf / .docx / .txt",
                        file_types=[".pdf", ".docx", ".txt"],
                        type="filepath"
                    )
                    btn_sum_file = gr.Button("Summarize File", variant="primary")
                    file_out = gr.Markdown()

                    btn_sum_file.click(
                        summarize_file,
                        inputs=[file_input, ck_bullets, sl_max_new_sum],
                        outputs=file_out
                    )

if __name__ == "__main__":
    demo.launch()