Spaces:
Paused
Paused
Nattapong Tapachoom
commited on
Commit
·
084df26
1
Parent(s):
b410a7f
Refactor app.py to improve model loading and PDF processing; update dataset generation logic and enhance UI components
Browse files
app.py
CHANGED
|
@@ -6,28 +6,30 @@ from datetime import datetime
|
|
| 6 |
from typing import List, Dict, Any, Tuple
|
| 7 |
|
| 8 |
import gradio as gr
|
|
|
|
| 9 |
|
| 10 |
-
|
| 11 |
-
OAUTH_AVAILABLE = bool(os.getenv("OAUTH_CLIENT_ID"))
|
| 12 |
|
| 13 |
-
#
|
| 14 |
-
|
| 15 |
-
REQUIRE_LOGIN = os.getenv("REQUIRE_LOGIN", _default_require).strip().lower() in ("1", "true", "yes", "y")
|
| 16 |
|
| 17 |
-
|
| 18 |
-
|
| 19 |
-
|
| 20 |
-
|
| 21 |
|
| 22 |
-
|
| 23 |
-
|
| 24 |
-
|
| 25 |
-
|
| 26 |
-
|
| 27 |
-
|
| 28 |
-
|
| 29 |
-
|
| 30 |
-
|
|
|
|
|
|
|
|
|
|
| 31 |
|
| 32 |
|
| 33 |
def ensure_output_dir() -> str:
|
|
@@ -37,11 +39,6 @@ def ensure_output_dir() -> str:
|
|
| 37 |
|
| 38 |
|
| 39 |
def read_pdfs(files: List[gr.File]) -> Tuple[str, List[Dict[str, Any]]]:
|
| 40 |
-
if not files:
|
| 41 |
-
return "", []
|
| 42 |
-
if PdfReader is None:
|
| 43 |
-
raise RuntimeError("pypdf is not installed. Please add it to requirements.txt or pip install pypdf.")
|
| 44 |
-
|
| 45 |
docs = []
|
| 46 |
combined_text_parts: List[str] = []
|
| 47 |
for f in files:
|
|
@@ -49,11 +46,7 @@ def read_pdfs(files: List[gr.File]) -> Tuple[str, List[Dict[str, Any]]]:
|
|
| 49 |
reader = PdfReader(path)
|
| 50 |
pages_text = []
|
| 51 |
for i, page in enumerate(reader.pages):
|
| 52 |
-
|
| 53 |
-
text = page.extract_text() or ""
|
| 54 |
-
except Exception:
|
| 55 |
-
text = ""
|
| 56 |
-
# Normalize whitespace
|
| 57 |
text = re.sub(r"\s+", " ", text).strip()
|
| 58 |
if text:
|
| 59 |
pages_text.append({"page": i + 1, "text": text})
|
|
@@ -63,665 +56,142 @@ def read_pdfs(files: List[gr.File]) -> Tuple[str, List[Dict[str, Any]]]:
|
|
| 63 |
return combined_text, docs
|
| 64 |
|
| 65 |
|
| 66 |
-
def chunk_text(text: str, chunk_size: int = 1500, overlap: int = 200, max_chunks: int = 5) -> List[
|
| 67 |
text = text.strip()
|
| 68 |
if not text:
|
| 69 |
return []
|
| 70 |
-
chunks: List[
|
| 71 |
start = 0
|
| 72 |
n = len(text)
|
| 73 |
while start < n and len(chunks) < max_chunks:
|
| 74 |
end = min(start + chunk_size, n)
|
| 75 |
chunk = text[start:end]
|
| 76 |
-
|
| 77 |
-
if end < n:
|
| 78 |
-
m = re.search(r"[\.!?]\s", text[end - 200:end] if end - 200 > start else text[start:end])
|
| 79 |
-
if m:
|
| 80 |
-
end = start + (m.end())
|
| 81 |
-
chunk = text[start:end]
|
| 82 |
-
chunks.append({"index": len(chunks), "start": start, "end": end, "text": chunk})
|
| 83 |
if end >= n:
|
| 84 |
break
|
| 85 |
start = max(end - overlap, 0)
|
| 86 |
-
if start == end: # safety
|
| 87 |
-
start += 1
|
| 88 |
return chunks
|
| 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 |
-
'สร้างคำถามแบบเลือกตอบจากเนื้อหา จำนวน {min_pairs} ถึง {max_pairs} ข้อ แต่ละข้อมี {num_options} ตัวเลือก '
|
| 121 |
-
'ส่งคืนเฉพาะ JSON array ของ objects ที่มี {{"question": str, "options": [str], "answer_index": int}} เท่านั้น ไม่ต้องมีข้อความเพิ่มเติม\n\n'
|
| 122 |
-
'เนื้อหา:\n{content}\n'
|
| 123 |
-
),
|
| 124 |
-
"True/False": (
|
| 125 |
-
'สร้างข้อความจริง/เท็จที่อิงจากเนื้อหาเท่านั้น จำนวน {min_pairs} ถึง {max_pairs} ข้อความ '
|
| 126 |
-
'ส่งคืนเฉพาะ JSON array ของ objects ที่มี {{"statement": str, "answer": bool, "explanation": str}} เท่านั้น ไม่ต้องมีข้อความเพิ่มเติม\n\n'
|
| 127 |
-
'เนื้อหา:\n{content}\n'
|
| 128 |
-
),
|
| 129 |
-
"Translation": (
|
| 130 |
-
'แปลเนื้อหาเป็น{target_language} สร้างคู่ประโยคแบบคู่ขนาน จำนวน {min_pairs} ถึง {max_pairs} คู่ '
|
| 131 |
-
'ส่งคืนเฉพาะ JSON array ของ objects ที่มี {{"source": str, "target": str}} เท่านั้น ไม่ต้องมีข้อความเพิ่มเติม\n\n'
|
| 132 |
-
'เนื้อหา:\n{content}\n'
|
| 133 |
-
),
|
| 134 |
-
"RLHF": (
|
| 135 |
-
'สร้างข้อมูลสำหรับ Reinforcement Learning from Human Feedback (RLHF) จากเนื้อหานี้ '
|
| 136 |
-
'สร้างคำถามและการตอบสนองหลายแบบ พร้อมคะแนนความต้องการของมนุษย์ จำนวน {min_pairs} ถึง {max_pairs} ชุด '
|
| 137 |
-
'ส่งคืนเฉพาะ JSON array ของ objects ที่มี {{"prompt": str, "responses": [str], "scores": [float], "preferred_response": str}} เท่านั้น\n\n'
|
| 138 |
-
'เนื้อหา:\n{content}\n'
|
| 139 |
-
),
|
| 140 |
-
"DPO": (
|
| 141 |
-
'สร้างข้อมูลสำหรับ Direct Preference Optimization (DPO) จากเนื้อหานี้ '
|
| 142 |
-
'สร้างคำถามพร้อมการตอบสนองที่ดีและไม่ดี จำนวน {min_pairs} ถึง {max_pairs} คู่ '
|
| 143 |
-
'ส่งคืนเฉพาะ JSON array ของ objects ที่มี {{"prompt": str, "chosen": str, "rejected": str, "reason": str}} เท่านั้น\n\n'
|
| 144 |
-
'เนื้อหา:\n{content}\n'
|
| 145 |
-
),
|
| 146 |
-
"Instruction_Following": (
|
| 147 |
-
'สร้างคำสั่งและการตอบสนองสำหรับการฝึกการทำตามคำสั่ง จำนวน {min_pairs} ถึง {max_pairs} คู่ '
|
| 148 |
-
'ส่งคืนเฉพาะ JSON array ของ objects ที่มี {{"instruction": str, "input": str, "output": str, "difficulty": str}} เท่านั้น\n\n'
|
| 149 |
-
'เนื้อหา:\n{content}\n'
|
| 150 |
-
),
|
| 151 |
-
"Constitutional_AI": (
|
| 152 |
-
'สร้างข้อมูลสำหรับ Constitutional AI โดยสร้างคำถามที่อาจมีปัญหาทางจริยธรรมและคำตอบที่เหมาะสม '
|
| 153 |
-
'จำนวน {min_pairs} ถึง {max_pairs} คู่ '
|
| 154 |
-
'ส่งคืนเฉพาะ JSON array ของ objects ที่มี {{"problematic_prompt": str, "constitutional_response": str, "principle": str}} เท่านั้น\n\n'
|
| 155 |
-
'เนื้อหา:\n{content}\n'
|
| 156 |
-
),
|
| 157 |
-
"Chain_of_Thought": (
|
| 158 |
-
'สร้างตัวอย่างการคิดแบบขั้นตอน (Chain of Thought) จากเนื้อหา จำนวน {min_pairs} ถึง {max_pairs} ตัวอย่าง '
|
| 159 |
-
'ส่งคืนเฉพาะ JSON array ของ objects ที่มี {{"problem": str, "thinking_steps": [str], "final_answer": str}} เท่านั้น\n\n'
|
| 160 |
-
'เนื้อหา:\n{content}\n'
|
| 161 |
-
),
|
| 162 |
-
"Dialogue": (
|
| 163 |
-
'สร้างบทสนทนาระหว่างผู้ใช้และผู้ช่วย AI จากเนื้อหา จำนวน {min_pairs} ถึง {max_pairs} บทสนทนา '
|
| 164 |
-
'ส่งคืนเฉพาะ JSON array ของ objects ที่มี {{"dialogue": [{{"role": str, "content": str}}], "context": str}} เท่านั้น\n\n'
|
| 165 |
-
'เนื้อหา:\n{content}\n'
|
| 166 |
-
),
|
| 167 |
-
"Thai_Culture": (
|
| 168 |
-
'สร้างคำถาม-คำตอบเกี่ยวกับวัฒนธรรมไทยจากเนื้อหา เน้นความเข้าใจภาษาไทยและบริบททางวัฒนธรรม '
|
| 169 |
-
'จำนวน {min_pairs} ถึง {max_pairs} คู่ '
|
| 170 |
-
'ส่งคืนเฉพาะ JSON array ของ objects ที่มี {{"question_th": str, "answer_th": str, "cultural_context": str}} เท่านั้น\n\n'
|
| 171 |
-
'เนื้อหา:\n{content}\n'
|
| 172 |
-
),
|
| 173 |
-
}
|
| 174 |
-
|
| 175 |
-
|
| 176 |
-
def extract_json_array(text: str) -> List[Dict[str, Any]]:
|
| 177 |
-
if not text:
|
| 178 |
-
return []
|
| 179 |
-
# Remove code fences
|
| 180 |
-
text = re.sub(r"```[a-zA-Z]*", "```", text)
|
| 181 |
-
text = text.replace("```", "")
|
| 182 |
-
# Find first [ ... ] block
|
| 183 |
-
start = text.find("[")
|
| 184 |
-
end = text.rfind("]")
|
| 185 |
-
if start != -1 and end != -1 and end > start:
|
| 186 |
-
candidate = text[start : end + 1]
|
| 187 |
-
else:
|
| 188 |
-
candidate = text
|
| 189 |
-
try:
|
| 190 |
-
data = json.loads(candidate)
|
| 191 |
-
if isinstance(data, list):
|
| 192 |
-
# normalize
|
| 193 |
-
norm = []
|
| 194 |
-
for item in data:
|
| 195 |
-
if not isinstance(item, dict):
|
| 196 |
-
continue
|
| 197 |
-
q = str(item.get("question", "").strip())
|
| 198 |
-
a = str(item.get("answer", "").strip())
|
| 199 |
-
if q and a:
|
| 200 |
-
norm.append({"question": q, "answer": a})
|
| 201 |
-
return norm
|
| 202 |
-
except Exception:
|
| 203 |
-
pass
|
| 204 |
-
return []
|
| 205 |
-
|
| 206 |
-
|
| 207 |
-
def build_langchain(model_id: str, hf_token: str | None, max_new_tokens: int, temperature: float, template: str):
|
| 208 |
-
if any(x is None for x in [PromptTemplate, JsonOutputParser, HuggingFaceEndpoint]):
|
| 209 |
-
raise RuntimeError("langchain, langchain-community, and langchain-huggingface are required. Please add to requirements.txt.")
|
| 210 |
-
# Prompt
|
| 211 |
-
prompt = PromptTemplate.from_template(template)
|
| 212 |
-
# Model wrapper (Hugging Face Inference API)
|
| 213 |
-
llm = HuggingFaceEndpoint(
|
| 214 |
-
model=model_id,
|
| 215 |
-
token=hf_token,
|
| 216 |
-
task="text-generation",
|
| 217 |
-
max_new_tokens=max_new_tokens,
|
| 218 |
-
temperature=temperature,
|
| 219 |
-
do_sample=temperature > 0.0,
|
| 220 |
-
)
|
| 221 |
-
parser = JsonOutputParser()
|
| 222 |
-
chain = prompt | llm | parser
|
| 223 |
-
return chain
|
| 224 |
-
|
| 225 |
-
|
| 226 |
-
def get_task_template(task: str, custom_instruction: str | None) -> str:
|
| 227 |
-
base = TASK_TEMPLATES.get(task, DEFAULT_QA_PROMPT_TMPL)
|
| 228 |
-
if custom_instruction and custom_instruction.strip():
|
| 229 |
-
# Allow user to override fully, but ensure {content} is present
|
| 230 |
-
if "{content}" not in custom_instruction:
|
| 231 |
-
custom_instruction = custom_instruction.strip() + "\n\nContent:\n{content}\n"
|
| 232 |
-
return custom_instruction
|
| 233 |
-
return base
|
| 234 |
-
|
| 235 |
-
|
| 236 |
-
def normalize_items(task: str, data: Any) -> List[Dict[str, Any]]:
|
| 237 |
-
# Convert model output to list[dict] per task
|
| 238 |
-
items: List[Dict[str, Any]] = []
|
| 239 |
-
if data is None:
|
| 240 |
-
return items
|
| 241 |
-
if isinstance(data, str):
|
| 242 |
-
data = extract_json_array(data)
|
| 243 |
-
if isinstance(data, dict):
|
| 244 |
-
# handle wrappers like {"items": [...]}
|
| 245 |
-
if "items" in data and isinstance(data["items"], list):
|
| 246 |
-
data = data["items"]
|
| 247 |
-
else:
|
| 248 |
-
data = [data]
|
| 249 |
-
if isinstance(data, list):
|
| 250 |
-
# keywords may be list[str]
|
| 251 |
-
if task == "Keywords" and data and all(isinstance(x, str) for x in data):
|
| 252 |
-
return [{"keyword": x} for x in data if x]
|
| 253 |
-
for el in data:
|
| 254 |
-
if isinstance(el, dict):
|
| 255 |
-
items.append(el)
|
| 256 |
-
# Validate per-task required fields and normalize variants
|
| 257 |
-
norm: List[Dict[str, Any]] = []
|
| 258 |
-
for it in items:
|
| 259 |
-
if task == "QA":
|
| 260 |
-
q = str(it.get("question", "")).strip()
|
| 261 |
-
a = str(it.get("answer", "")).strip()
|
| 262 |
-
if q and a:
|
| 263 |
-
norm.append({"question": q, "answer": a})
|
| 264 |
-
elif task == "Summarization":
|
| 265 |
-
s = str(it.get("summary", "")).strip()
|
| 266 |
-
if s:
|
| 267 |
-
norm.append({"summary": s})
|
| 268 |
-
elif task == "Keywords":
|
| 269 |
-
k = it.get("keyword")
|
| 270 |
-
if isinstance(k, str) and k.strip():
|
| 271 |
-
norm.append({"keyword": k.strip()})
|
| 272 |
-
elif isinstance(it.get("keywords"), list):
|
| 273 |
-
for kw in it["keywords"]:
|
| 274 |
-
if isinstance(kw, str) and kw.strip():
|
| 275 |
-
norm.append({"keyword": kw.strip()})
|
| 276 |
-
elif task == "NER":
|
| 277 |
-
txt = it.get("text")
|
| 278 |
-
label = it.get("label")
|
| 279 |
-
start = it.get("start")
|
| 280 |
-
end = it.get("end")
|
| 281 |
-
if isinstance(txt, str) and isinstance(label, str) and isinstance(start, int) and isinstance(end, int):
|
| 282 |
-
norm.append({"text": txt, "label": label, "start": start, "end": end})
|
| 283 |
-
elif isinstance(it.get("entities"), list):
|
| 284 |
-
for ent in it["entities"]:
|
| 285 |
-
if all(k in ent for k in ("text", "label", "start", "end")):
|
| 286 |
-
norm.append({
|
| 287 |
-
"text": str(ent.get("text", "")),
|
| 288 |
-
"label": str(ent.get("label", "")),
|
| 289 |
-
"start": int(ent.get("start", 0)),
|
| 290 |
-
"end": int(ent.get("end", 0)),
|
| 291 |
-
})
|
| 292 |
-
elif task == "Classification":
|
| 293 |
-
labels = it.get("labels")
|
| 294 |
-
if isinstance(labels, str):
|
| 295 |
-
labels = [labels]
|
| 296 |
-
if isinstance(labels, list):
|
| 297 |
-
labels = [str(x).strip() for x in labels if str(x).strip()]
|
| 298 |
-
rationale = str(it.get("rationale", "")).strip()
|
| 299 |
-
if labels:
|
| 300 |
-
norm.append({"labels": labels, "rationale": rationale})
|
| 301 |
-
elif task == "MCQ":
|
| 302 |
-
q = it.get("question")
|
| 303 |
-
options = it.get("options")
|
| 304 |
-
answer_index = it.get("answer_index")
|
| 305 |
-
answer = it.get("answer")
|
| 306 |
-
if isinstance(options, list) and all(isinstance(o, str) for o in options) and isinstance(q, str):
|
| 307 |
-
if isinstance(answer_index, int):
|
| 308 |
-
idx = answer_index
|
| 309 |
-
elif isinstance(answer, str) and answer in options:
|
| 310 |
-
idx = options.index(answer)
|
| 311 |
-
else:
|
| 312 |
-
continue
|
| 313 |
-
norm.append({"question": q, "options": options, "answer_index": idx})
|
| 314 |
-
elif task == "True/False":
|
| 315 |
-
st = it.get("statement")
|
| 316 |
-
ans = it.get("answer")
|
| 317 |
-
expl = it.get("explanation", "")
|
| 318 |
-
if isinstance(st, str):
|
| 319 |
-
if isinstance(ans, bool):
|
| 320 |
-
val = ans
|
| 321 |
-
elif isinstance(ans, str):
|
| 322 |
-
val = ans.strip().lower() in ("true", "t", "yes", "1")
|
| 323 |
-
else:
|
| 324 |
-
continue
|
| 325 |
-
norm.append({"statement": st, "answer": val, "explanation": str(expl)})
|
| 326 |
-
elif task == "Translation":
|
| 327 |
-
src = it.get("source")
|
| 328 |
-
tgt = it.get("target")
|
| 329 |
-
if isinstance(src, str) and isinstance(tgt, str) and src.strip() and tgt.strip():
|
| 330 |
-
norm.append({"source": src, "target": tgt})
|
| 331 |
-
elif task == "RLHF":
|
| 332 |
-
prompt = it.get("prompt")
|
| 333 |
-
responses = it.get("responses")
|
| 334 |
-
scores = it.get("scores")
|
| 335 |
-
preferred = it.get("preferred_response")
|
| 336 |
-
if isinstance(prompt, str) and isinstance(responses, list) and isinstance(scores, list):
|
| 337 |
-
norm.append({
|
| 338 |
-
"prompt": prompt,
|
| 339 |
-
"responses": responses,
|
| 340 |
-
"scores": scores,
|
| 341 |
-
"preferred_response": str(preferred) if preferred else ""
|
| 342 |
-
})
|
| 343 |
-
elif task == "DPO":
|
| 344 |
-
prompt = it.get("prompt")
|
| 345 |
-
chosen = it.get("chosen")
|
| 346 |
-
rejected = it.get("rejected")
|
| 347 |
-
reason = it.get("reason", "")
|
| 348 |
-
if isinstance(prompt, str) and isinstance(chosen, str) and isinstance(rejected, str):
|
| 349 |
-
norm.append({
|
| 350 |
-
"prompt": prompt,
|
| 351 |
-
"chosen": chosen,
|
| 352 |
-
"rejected": rejected,
|
| 353 |
-
"reason": str(reason)
|
| 354 |
-
})
|
| 355 |
-
elif task == "Instruction_Following":
|
| 356 |
-
instruction = it.get("instruction")
|
| 357 |
-
input_text = it.get("input", "")
|
| 358 |
-
output = it.get("output")
|
| 359 |
-
difficulty = it.get("difficulty", "medium")
|
| 360 |
-
if isinstance(instruction, str) and isinstance(output, str):
|
| 361 |
-
norm.append({
|
| 362 |
-
"instruction": instruction,
|
| 363 |
-
"input": str(input_text),
|
| 364 |
-
"output": output,
|
| 365 |
-
"difficulty": str(difficulty)
|
| 366 |
-
})
|
| 367 |
-
elif task == "Constitutional_AI":
|
| 368 |
-
problematic = it.get("problematic_prompt")
|
| 369 |
-
constitutional = it.get("constitutional_response")
|
| 370 |
-
principle = it.get("principle", "")
|
| 371 |
-
if isinstance(problematic, str) and isinstance(constitutional, str):
|
| 372 |
-
norm.append({
|
| 373 |
-
"problematic_prompt": problematic,
|
| 374 |
-
"constitutional_response": constitutional,
|
| 375 |
-
"principle": str(principle)
|
| 376 |
-
})
|
| 377 |
-
elif task == "Chain_of_Thought":
|
| 378 |
-
problem = it.get("problem")
|
| 379 |
-
steps = it.get("thinking_steps")
|
| 380 |
-
answer = it.get("final_answer")
|
| 381 |
-
if isinstance(problem, str) and isinstance(steps, list) and isinstance(answer, str):
|
| 382 |
-
norm.append({
|
| 383 |
-
"problem": problem,
|
| 384 |
-
"thinking_steps": steps,
|
| 385 |
-
"final_answer": answer
|
| 386 |
-
})
|
| 387 |
-
elif task == "Dialogue":
|
| 388 |
-
dialogue = it.get("dialogue")
|
| 389 |
-
context = it.get("context", "")
|
| 390 |
-
if isinstance(dialogue, list):
|
| 391 |
-
norm.append({
|
| 392 |
-
"dialogue": dialogue,
|
| 393 |
-
"context": str(context)
|
| 394 |
-
})
|
| 395 |
-
elif task == "Thai_Culture":
|
| 396 |
-
question_th = it.get("question_th")
|
| 397 |
-
answer_th = it.get("answer_th")
|
| 398 |
-
cultural_context = it.get("cultural_context", "")
|
| 399 |
-
if isinstance(question_th, str) and isinstance(answer_th, str):
|
| 400 |
-
norm.append({
|
| 401 |
-
"question_th": question_th,
|
| 402 |
-
"answer_th": answer_th,
|
| 403 |
-
"cultural_context": str(cultural_context)
|
| 404 |
-
})
|
| 405 |
-
return norm
|
| 406 |
-
|
| 407 |
-
|
| 408 |
-
def generate_dataset(
|
| 409 |
-
user_profile: Any | None,
|
| 410 |
-
files: List[gr.File],
|
| 411 |
-
task: str,
|
| 412 |
-
preset_model: str,
|
| 413 |
-
custom_model_id: str,
|
| 414 |
-
hf_token: str,
|
| 415 |
-
chunk_size: int,
|
| 416 |
-
overlap: int,
|
| 417 |
-
max_chunks: int,
|
| 418 |
-
max_new_tokens: int,
|
| 419 |
-
temperature: float,
|
| 420 |
-
custom_instruction: str,
|
| 421 |
-
min_pairs: int,
|
| 422 |
-
max_pairs: int,
|
| 423 |
-
class_labels_text: str,
|
| 424 |
-
multi_label: bool,
|
| 425 |
-
target_language: str,
|
| 426 |
-
num_options: int,
|
| 427 |
-
ner_labels_text: str,
|
| 428 |
-
):
|
| 429 |
-
# Enforce login if required
|
| 430 |
-
if REQUIRE_LOGIN and not user_profile:
|
| 431 |
-
return "กรุณาเข้าสู่ระบบก่อนเพื่อสร้างชุดข้อมูล", None, None
|
| 432 |
-
|
| 433 |
-
# Read and chunk
|
| 434 |
-
full_text, _docs = read_pdfs(files)
|
| 435 |
-
chunks = chunk_text(full_text, chunk_size=chunk_size, overlap=overlap, max_chunks=max_chunks)
|
| 436 |
if not chunks:
|
| 437 |
-
return "
|
| 438 |
-
|
| 439 |
-
|
| 440 |
-
# Prepare template per task
|
| 441 |
-
base_template = get_task_template(task, custom_instruction)
|
| 442 |
-
# enrich template with conditional clauses
|
| 443 |
-
ner_clause = ""
|
| 444 |
-
if ner_labels_text.strip():
|
| 445 |
-
ner_clause = f" (limit to: {ner_labels_text.strip()})"
|
| 446 |
-
base_template = base_template.replace("{ner_labels_clause}", ner_clause)
|
| 447 |
-
if "{labels}" in base_template:
|
| 448 |
-
labels_text = class_labels_text.strip() or "[]"
|
| 449 |
-
base_template = base_template.replace("{labels}", labels_text)
|
| 450 |
-
if "{multi_label_clause}" in base_template:
|
| 451 |
-
base_template = base_template.replace("{multi_label_clause}", " Allow multiple labels." if multi_label else " Choose a single best label.")
|
| 452 |
-
if "{num_options}" in base_template:
|
| 453 |
-
base_template = base_template.replace("{num_options}", str(int(num_options)))
|
| 454 |
-
try:
|
| 455 |
-
chain = build_langchain(model_id, hf_token or None, max_new_tokens, temperature, base_template)
|
| 456 |
-
except Exception as e:
|
| 457 |
-
return f"ข้อผิดพลาดในการเตรียม LangChain: {e}", None, None
|
| 458 |
-
|
| 459 |
-
results: List[Dict[str, Any]] = []
|
| 460 |
for ch in chunks:
|
| 461 |
-
|
| 462 |
-
|
| 463 |
-
|
| 464 |
-
|
| 465 |
-
|
| 466 |
-
|
| 467 |
-
|
| 468 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 469 |
try:
|
| 470 |
-
|
| 471 |
-
|
|
|
|
| 472 |
except Exception:
|
| 473 |
-
|
| 474 |
-
|
| 475 |
-
for it in items:
|
| 476 |
-
# Enrich with context and task
|
| 477 |
-
it["context"] = (ch["text"][:500] + ("..." if len(ch["text"]) > 500 else ""))
|
| 478 |
-
it["task"] = task
|
| 479 |
-
results.append(it)
|
| 480 |
|
| 481 |
if not results:
|
| 482 |
-
return
|
| 483 |
-
|
| 484 |
-
#
|
| 485 |
-
unique: List[Dict[str, Any]] = []
|
| 486 |
-
seen = set()
|
| 487 |
-
def key_of(item: Dict[str, Any]) -> str:
|
| 488 |
-
if task == "QA":
|
| 489 |
-
return (item.get("question") or "").strip().lower()
|
| 490 |
-
if task == "Summarization":
|
| 491 |
-
return (item.get("summary") or "").strip().lower()
|
| 492 |
-
if task == "Keywords":
|
| 493 |
-
return (item.get("keyword") or "").strip().lower()
|
| 494 |
-
if task == "NER":
|
| 495 |
-
return f"{item.get('text')}|{item.get('label')}|{item.get('start')}|{item.get('end')}"
|
| 496 |
-
if task == "Classification":
|
| 497 |
-
return ",".join(sorted([str(x).lower() for x in item.get("labels", [])]))
|
| 498 |
-
if task == "MCQ":
|
| 499 |
-
return (item.get("question") or "").strip().lower()
|
| 500 |
-
if task == "True/False":
|
| 501 |
-
return (item.get("statement") or "").strip().lower()
|
| 502 |
-
if task == "Translation":
|
| 503 |
-
return f"{item.get('source')}|{item.get('target')}"
|
| 504 |
-
if task == "RLHF":
|
| 505 |
-
return (item.get("prompt") or "").strip().lower()
|
| 506 |
-
if task == "DPO":
|
| 507 |
-
return (item.get("prompt") or "").strip().lower()
|
| 508 |
-
if task == "Instruction_Following":
|
| 509 |
-
return (item.get("instruction") or "").strip().lower()
|
| 510 |
-
if task == "Constitutional_AI":
|
| 511 |
-
return (item.get("problematic_prompt") or "").strip().lower()
|
| 512 |
-
if task == "Chain_of_Thought":
|
| 513 |
-
return (item.get("problem") or "").strip().lower()
|
| 514 |
-
if task == "Dialogue":
|
| 515 |
-
dialogue = item.get("dialogue", [])
|
| 516 |
-
if dialogue and isinstance(dialogue, list):
|
| 517 |
-
return str(dialogue[0].get("content", "")).strip().lower()
|
| 518 |
-
return ""
|
| 519 |
-
if task == "Thai_Culture":
|
| 520 |
-
return (item.get("question_th") or "").strip().lower()
|
| 521 |
-
return json.dumps(item, ensure_ascii=False)
|
| 522 |
-
for r in results:
|
| 523 |
-
k = key_of(r)
|
| 524 |
-
if k and k not in seen:
|
| 525 |
-
unique.append(r)
|
| 526 |
-
seen.add(k)
|
| 527 |
-
|
| 528 |
-
# Save to outputs
|
| 529 |
outdir = ensure_output_dir()
|
| 530 |
ts = datetime.utcnow().strftime("%Y%m%d_%H%M%S")
|
| 531 |
-
|
| 532 |
-
|
| 533 |
-
|
| 534 |
with io.open(json_path, "w", encoding="utf-8") as f:
|
| 535 |
-
json.dump(
|
| 536 |
with io.open(jsonl_path, "w", encoding="utf-8") as f:
|
| 537 |
-
for item in
|
| 538 |
f.write(json.dumps(item, ensure_ascii=False) + "\n")
|
| 539 |
|
| 540 |
-
return f"สร้างข้อมูลสำเร็จ {len(
|
| 541 |
|
| 542 |
|
|
|
|
| 543 |
PRESET_MODELS = [
|
| 544 |
-
|
| 545 |
-
"openthaigpt/openthaigpt-1.0.0-alpha-7b-chat",
|
| 546 |
-
"scb10x/llama-3-typhoon-v1.5-8b-instruct",
|
| 547 |
-
"airesearch/wangchanberta-base-att-spm-uncased",
|
| 548 |
-
|
| 549 |
-
# Multilingual models good for Thai
|
| 550 |
-
"google/mt5-large",
|
| 551 |
-
"microsoft/mdeberta-v3-base",
|
| 552 |
-
"facebook/xglm-7.5B",
|
| 553 |
-
"microsoft/DialoGPT-medium",
|
| 554 |
-
|
| 555 |
-
# General powerful models
|
| 556 |
-
"HuggingFaceH4/zephyr-7b-beta",
|
| 557 |
"mistralai/Mistral-7B-Instruct-v0.2",
|
| 558 |
-
"google/flan-t5-large",
|
| 559 |
"meta-llama/Llama-2-7b-chat-hf",
|
| 560 |
-
"
|
| 561 |
]
|
| 562 |
|
|
|
|
|
|
|
| 563 |
|
| 564 |
-
with gr.Blocks(title="AutoGDataset Thai - PDF to Dataset Generator") as demo:
|
| 565 |
-
gr.Markdown("""
|
| 566 |
-
# AutoGDataset Thai 🇹🇭
|
| 567 |
-
สร้างชุดข้อมูล (Dataset) ภาษาไทยจากไฟล์ PDF โดยใช้ LangChain กับโมเดล Hugging Face
|
| 568 |
-
|
| 569 |
-
**คุณสมบัติ:**
|
| 570 |
-
- รองรับงานหลากหลายประเภท: QA, RLHF, DPO, Constitutional AI และอื่นๆ
|
| 571 |
-
- เน้นการสร้างข้อมูลภาษาไทยคุณภาพสูง
|
| 572 |
-
- รองรับโมเดลภาษาไทยและ multilingual models
|
| 573 |
-
- สามารถปรับแต่ง prompt เพื่อเพิ่มประสิทธิภาพ
|
| 574 |
-
|
| 575 |
-
เลือกโมเดลที่มีอยู่หรือระบุ repo id ที่กำหนดเอง ระบุ `HF_TOKEN` หากจำเป็นสำหรับโมเดล
|
| 576 |
-
""")
|
| 577 |
-
|
| 578 |
-
# Login requirement (Hugging Face OAuth via Gradio LoginButton when available)
|
| 579 |
-
user_state = gr.State(value=None)
|
| 580 |
-
effective_require_login = bool(REQUIRE_LOGIN and OAUTH_AVAILABLE)
|
| 581 |
with gr.Row():
|
| 582 |
-
|
| 583 |
-
value=(
|
| 584 |
-
"กรุณาเข้าสู่ระบบด้วยบัญชี Hugging Face เพื่อใช้งานแอป"
|
| 585 |
-
if effective_require_login
|
| 586 |
-
else (
|
| 587 |
-
"การเข้าสู่ระบบเป็นทางเลือก" if OAUTH_AVAILABLE else "ไม่ได้ตั้งค่าการเข้าสู่ระบบ OAuth ในการติดตั้งนี้"
|
| 588 |
-
)
|
| 589 |
-
),
|
| 590 |
-
elem_id="login-info",
|
| 591 |
-
)
|
| 592 |
-
if OAUTH_AVAILABLE:
|
| 593 |
-
with gr.Row():
|
| 594 |
-
login_btn = gr.LoginButton(value="เข้าสู่ระบบด้วย Hugging Face")
|
| 595 |
|
| 596 |
with gr.Row():
|
| 597 |
-
|
| 598 |
-
|
| 599 |
-
|
| 600 |
-
|
| 601 |
-
task = gr.Dropdown(
|
| 602 |
-
label="งานที่ต้องการ (Task Type)",
|
| 603 |
-
choices=[
|
| 604 |
-
"QA",
|
| 605 |
-
"Summarization",
|
| 606 |
-
"Keywords",
|
| 607 |
-
"NER",
|
| 608 |
-
"Classification",
|
| 609 |
-
"MCQ",
|
| 610 |
-
"True/False",
|
| 611 |
-
"Translation",
|
| 612 |
-
"RLHF",
|
| 613 |
-
"DPO",
|
| 614 |
-
"Instruction_Following",
|
| 615 |
-
"Constitutional_AI",
|
| 616 |
-
"Chain_of_Thought",
|
| 617 |
-
"Dialogue",
|
| 618 |
-
"Thai_Culture",
|
| 619 |
-
],
|
| 620 |
-
value="Thai_Culture",
|
| 621 |
-
)
|
| 622 |
-
with gr.Row():
|
| 623 |
-
preset_model = gr.Dropdown(label="โมเดลที่กำหนดไว้ (Preset Model)", choices=PRESET_MODELS, value=PRESET_MODELS[0])
|
| 624 |
-
custom_model_id = gr.Textbox(label="รหัสโมเดลกำหนดเอง (ไม่บังคับ)", placeholder="org/model-name")
|
| 625 |
-
with gr.Row():
|
| 626 |
-
hf_token = gr.Textbox(label="HF Token", type="password", value=os.environ.get("HF_TOKEN", ""), placeholder="hf_xxx (จำเป็นสำหรับหลายโมเดล)")
|
| 627 |
-
with gr.Row():
|
| 628 |
-
max_new_tokens = gr.Slider(64, 1024, value=512, step=16, label="จำนวน Token สูงสุด")
|
| 629 |
-
temperature = gr.Slider(0.0, 1.5, value=0.2, step=0.05, label="อุณหภูมิ (ความสร้างสรรค์)")
|
| 630 |
-
|
| 631 |
-
with gr.Accordion("การตั้งค่าขั้นสูง (Advanced Settings)", open=False):
|
| 632 |
-
with gr.Row():
|
| 633 |
-
chunk_size = gr.Slider(500, 4000, value=1500, step=50, label="ขนาดส่วนข้อความ (ตัวอักษร)")
|
| 634 |
-
overlap = gr.Slider(0, 1000, value=200, step=50, label="การทับซ้อน (ตัวอักษร)")
|
| 635 |
-
max_chunks = gr.Slider(1, 40, value=5, step=1, label="จำนวนส่วนสูงสุด")
|
| 636 |
-
with gr.Row():
|
| 637 |
-
min_pairs = gr.Slider(1, 10, value=3, step=1, label="คู่ข้อมูลต่ำสุด/ส่วน")
|
| 638 |
-
max_pairs = gr.Slider(1, 12, value=6, step=1, label="คู่ข้อมูลสูงสุด/ส่วน")
|
| 639 |
-
custom_instruction = gr.Textbox(
|
| 640 |
-
label="คำสั่งกำหนดเอง (ไม่บังคับ)",
|
| 641 |
-
lines=3,
|
| 642 |
-
placeholder="แทนที่คำสั่งเริ่มต้น ต้องส่งคืน JSON array บริสุทธิ์ตามโครงสร้างงาน",
|
| 643 |
-
value="สร้างข้อมูลภาษาไทยคุณภาพสูงที่เข้าใจบริบททางวัฒนธรรมไทย ใช้ภาษาไทยที่เป็นธรรมชาติและเหมาะสมกับเนื้อหา"
|
| 644 |
-
)
|
| 645 |
-
|
| 646 |
-
# Task-specific controls
|
| 647 |
-
classification_labels = gr.Textbox(label="ป้ายกำกับการจำแนก (คั่นด้วยคอมมา)", visible=False)
|
| 648 |
-
multi_label = gr.Checkbox(label="อนุญาตหลายป้ายกำกับ", value=False, visible=False)
|
| 649 |
-
target_language = gr.Textbox(label="ภาษาเป้าหมาย (การแปล)", value="ไทย", visible=False)
|
| 650 |
-
num_options = gr.Slider(3, 6, value=4, step=1, label="ตัวเลือก MCQ", visible=False)
|
| 651 |
-
ner_labels = gr.Textbox(label="ป้ายกำกับ NER (คั่นด้วยคอมมา, ไม่บังคับ)", visible=False)
|
| 652 |
|
| 653 |
-
|
|
|
|
|
|
|
| 654 |
|
| 655 |
with gr.Row():
|
| 656 |
-
|
|
|
|
|
|
|
|
|
|
| 657 |
with gr.Row():
|
| 658 |
-
|
| 659 |
-
|
| 660 |
-
|
| 661 |
-
# Toggle visibility for task-specific controls
|
| 662 |
-
def _switch_task(t: str):
|
| 663 |
-
is_cls = t == "Classification"
|
| 664 |
-
is_tr = t == "Translation"
|
| 665 |
-
is_mcq = t == "MCQ"
|
| 666 |
-
is_ner = t == "NER"
|
| 667 |
-
return (
|
| 668 |
-
gr.update(visible=is_cls), # classification_labels
|
| 669 |
-
gr.update(visible=is_cls), # multi_label
|
| 670 |
-
gr.update(visible=is_tr), # target_language
|
| 671 |
-
gr.update(visible=is_mcq), # num_options
|
| 672 |
-
gr.update(visible=is_ner), # ner_labels
|
| 673 |
-
)
|
| 674 |
|
| 675 |
-
|
|
|
|
|
|
|
|
|
|
| 676 |
|
| 677 |
generate_btn.click(
|
| 678 |
fn=generate_dataset,
|
| 679 |
-
inputs=[
|
| 680 |
-
|
| 681 |
-
|
| 682 |
-
|
| 683 |
-
preset_model,
|
| 684 |
-
custom_model_id,
|
| 685 |
-
hf_token,
|
| 686 |
-
chunk_size,
|
| 687 |
-
overlap,
|
| 688 |
-
max_chunks,
|
| 689 |
-
max_new_tokens,
|
| 690 |
-
temperature,
|
| 691 |
-
custom_instruction,
|
| 692 |
-
min_pairs,
|
| 693 |
-
max_pairs,
|
| 694 |
-
classification_labels,
|
| 695 |
-
multi_label,
|
| 696 |
-
target_language,
|
| 697 |
-
num_options,
|
| 698 |
-
ner_labels,
|
| 699 |
-
],
|
| 700 |
-
outputs=[status, out_json, out_jsonl],
|
| 701 |
-
show_progress=True,
|
| 702 |
-
api_name="generate",
|
| 703 |
)
|
| 704 |
|
| 705 |
-
if OAUTH_AVAILABLE:
|
| 706 |
-
def _on_login(user):
|
| 707 |
-
try:
|
| 708 |
-
username = None
|
| 709 |
-
if isinstance(user, dict):
|
| 710 |
-
username = user.get("username") or user.get("name")
|
| 711 |
-
if not username and hasattr(user, "username"):
|
| 712 |
-
username = getattr(user, "username")
|
| 713 |
-
msg = f"เข้าสู่ระบบแล้วในนาม @{username}" if username else "เข้าสู่ระบบแล้ว"
|
| 714 |
-
except Exception:
|
| 715 |
-
msg = "เข้าสู่ระบบแล้ว"
|
| 716 |
-
return user, gr.update(value=msg), gr.update(interactive=True)
|
| 717 |
-
|
| 718 |
-
# Enable Generate button after login and store user profile
|
| 719 |
-
if hasattr(login_btn, "login"):
|
| 720 |
-
login_btn.login(_on_login, inputs=None, outputs=[user_state, login_info, generate_btn])
|
| 721 |
-
else:
|
| 722 |
-
# In local/dev without OAuth routing, clicking will mock-login
|
| 723 |
-
login_btn.click(lambda: ("local_user", gr.update(value="เข้าสู่ระบบแล้ว (ภายในเครื่อง)"), gr.update(interactive=True)), inputs=None, outputs=[user_state, login_info, generate_btn])
|
| 724 |
-
|
| 725 |
if __name__ == "__main__":
|
| 726 |
-
|
| 727 |
-
demo.queue().launch()
|
|
|
|
| 6 |
from typing import List, Dict, Any, Tuple
|
| 7 |
|
| 8 |
import gradio as gr
|
| 9 |
+
from pypdf import PdfReader
|
| 10 |
|
| 11 |
+
from transformers import AutoTokenizer, AutoModelForCausalLM, pipeline
|
|
|
|
| 12 |
|
| 13 |
+
# โหลดโมเดลเริ่มต้น (default)
|
| 14 |
+
DEFAULT_MODEL = "HuggingFaceH4/zephyr-7b-beta"
|
|
|
|
| 15 |
|
| 16 |
+
# สร้าง pipeline global
|
| 17 |
+
gen_pipe = None
|
| 18 |
+
tokenizer = None
|
| 19 |
+
current_model_id = None
|
| 20 |
|
| 21 |
+
|
| 22 |
+
def load_model(model_id: str, hf_token: str = None):
|
| 23 |
+
global gen_pipe, tokenizer, current_model_id
|
| 24 |
+
if current_model_id == model_id and gen_pipe is not None:
|
| 25 |
+
return gen_pipe
|
| 26 |
+
|
| 27 |
+
print(f"🔄 Loading model: {model_id}")
|
| 28 |
+
tokenizer = AutoTokenizer.from_pretrained(model_id, token=hf_token)
|
| 29 |
+
model = AutoModelForCausalLM.from_pretrained(model_id, token=hf_token, device_map="auto")
|
| 30 |
+
gen_pipe = pipeline("text-generation", model=model, tokenizer=tokenizer, device_map="auto")
|
| 31 |
+
current_model_id = model_id
|
| 32 |
+
return gen_pipe
|
| 33 |
|
| 34 |
|
| 35 |
def ensure_output_dir() -> str:
|
|
|
|
| 39 |
|
| 40 |
|
| 41 |
def read_pdfs(files: List[gr.File]) -> Tuple[str, List[Dict[str, Any]]]:
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 42 |
docs = []
|
| 43 |
combined_text_parts: List[str] = []
|
| 44 |
for f in files:
|
|
|
|
| 46 |
reader = PdfReader(path)
|
| 47 |
pages_text = []
|
| 48 |
for i, page in enumerate(reader.pages):
|
| 49 |
+
text = page.extract_text() or ""
|
|
|
|
|
|
|
|
|
|
|
|
|
| 50 |
text = re.sub(r"\s+", " ", text).strip()
|
| 51 |
if text:
|
| 52 |
pages_text.append({"page": i + 1, "text": text})
|
|
|
|
| 56 |
return combined_text, docs
|
| 57 |
|
| 58 |
|
| 59 |
+
def chunk_text(text: str, chunk_size: int = 1500, overlap: int = 200, max_chunks: int = 5) -> List[str]:
|
| 60 |
text = text.strip()
|
| 61 |
if not text:
|
| 62 |
return []
|
| 63 |
+
chunks: List[str] = []
|
| 64 |
start = 0
|
| 65 |
n = len(text)
|
| 66 |
while start < n and len(chunks) < max_chunks:
|
| 67 |
end = min(start + chunk_size, n)
|
| 68 |
chunk = text[start:end]
|
| 69 |
+
chunks.append(chunk)
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 70 |
if end >= n:
|
| 71 |
break
|
| 72 |
start = max(end - overlap, 0)
|
|
|
|
|
|
|
| 73 |
return chunks
|
| 74 |
|
| 75 |
|
| 76 |
+
# เทมเพลต prompt พื้นฐาน
|
| 77 |
+
DEFAULT_QA_PROMPT = (
|
| 78 |
+
"คุณเป็นผู้ช่วยสร้างชุดข้อมูล อ่านเนื้อหานี้แล้วสร้างคำถาม-คำตอบ "
|
| 79 |
+
"จำนวน {min_pairs} ถึ�� {max_pairs} คู่ "
|
| 80 |
+
"ส่งคืน JSON array ที่มี objects รูปแบบ {{\"question\": str, \"answer\": str}} เท่านั้น\n\n"
|
| 81 |
+
"เนื้อหา:\n{content}\n"
|
| 82 |
)
|
| 83 |
|
| 84 |
+
|
| 85 |
+
def generate_dataset(files: List[gr.File],
|
| 86 |
+
task: str,
|
| 87 |
+
preset_model: str,
|
| 88 |
+
custom_model_id: str,
|
| 89 |
+
hf_token: str,
|
| 90 |
+
chunk_size: int,
|
| 91 |
+
overlap: int,
|
| 92 |
+
max_chunks: int,
|
| 93 |
+
max_new_tokens: int,
|
| 94 |
+
temperature: float,
|
| 95 |
+
min_pairs: int,
|
| 96 |
+
max_pairs: int):
|
| 97 |
+
if not files:
|
| 98 |
+
return "❌ กรุณาอัปโหลดไฟล์ PDF", None, None
|
| 99 |
+
|
| 100 |
+
# โหลดโมเดล
|
| 101 |
+
model_id = (custom_model_id or "").strip() or preset_model or DEFAULT_MODEL
|
| 102 |
+
pipe = load_model(model_id, hf_token or None)
|
| 103 |
+
|
| 104 |
+
# อ่าน PDF และตัดเป็น chunk
|
| 105 |
+
full_text, _ = read_pdfs(files)
|
| 106 |
+
chunks = chunk_text(full_text, chunk_size, overlap, max_chunks)
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 107 |
if not chunks:
|
| 108 |
+
return "❌ ไม่สามารถดึงข้อความจาก PDF", None, None
|
| 109 |
+
|
| 110 |
+
results = []
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 111 |
for ch in chunks:
|
| 112 |
+
prompt = DEFAULT_QA_PROMPT.format(
|
| 113 |
+
min_pairs=min_pairs,
|
| 114 |
+
max_pairs=max_pairs,
|
| 115 |
+
content=ch
|
| 116 |
+
)
|
| 117 |
+
output = pipe(prompt,
|
| 118 |
+
max_new_tokens=max_new_tokens,
|
| 119 |
+
temperature=temperature,
|
| 120 |
+
do_sample=temperature > 0.0)[0]["generated_text"]
|
| 121 |
+
|
| 122 |
+
# พยายาม extract JSON
|
| 123 |
+
start, end = output.find("["), output.rfind("]")
|
| 124 |
+
if start != -1 and end != -1:
|
| 125 |
try:
|
| 126 |
+
data = json.loads(output[start:end + 1])
|
| 127 |
+
if isinstance(data, list):
|
| 128 |
+
results.extend(data)
|
| 129 |
except Exception:
|
| 130 |
+
pass
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 131 |
|
| 132 |
if not results:
|
| 133 |
+
return "❌ ไม่สามารถสร้างข้อมูล JSON ได้", None, None
|
| 134 |
+
|
| 135 |
+
# Save output
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 136 |
outdir = ensure_output_dir()
|
| 137 |
ts = datetime.utcnow().strftime("%Y%m%d_%H%M%S")
|
| 138 |
+
json_path = os.path.join(outdir, f"dataset_{task}_{ts}.json")
|
| 139 |
+
jsonl_path = os.path.join(outdir, f"dataset_{task}_{ts}.jsonl")
|
| 140 |
+
|
| 141 |
with io.open(json_path, "w", encoding="utf-8") as f:
|
| 142 |
+
json.dump(results, f, ensure_ascii=False, indent=2)
|
| 143 |
with io.open(jsonl_path, "w", encoding="utf-8") as f:
|
| 144 |
+
for item in results:
|
| 145 |
f.write(json.dumps(item, ensure_ascii=False) + "\n")
|
| 146 |
|
| 147 |
+
return f"✅ สร้างข้อมูลสำเร็จ {len(results)} รายการ", json_path, jsonl_path
|
| 148 |
|
| 149 |
|
| 150 |
+
# ---------------- Gradio UI ----------------
|
| 151 |
PRESET_MODELS = [
|
| 152 |
+
DEFAULT_MODEL,
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 153 |
"mistralai/Mistral-7B-Instruct-v0.2",
|
|
|
|
| 154 |
"meta-llama/Llama-2-7b-chat-hf",
|
| 155 |
+
"google/flan-t5-large"
|
| 156 |
]
|
| 157 |
|
| 158 |
+
with gr.Blocks(title="Thai PDF → Dataset Generator") as demo:
|
| 159 |
+
gr.Markdown("# 📚 Thai Auto Dataset Generator")
|
| 160 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 161 |
with gr.Row():
|
| 162 |
+
pdf_files = gr.File(label="อัปโหลด PDF", file_count="multiple", file_types=[".pdf"])
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 163 |
|
| 164 |
with gr.Row():
|
| 165 |
+
task = gr.Textbox(label="Task", value="QA")
|
| 166 |
+
preset_model = gr.Dropdown(label="Preset Model", choices=PRESET_MODELS, value=DEFAULT_MODEL)
|
| 167 |
+
custom_model_id = gr.Textbox(label="Custom Model ID", placeholder="org/model-name")
|
| 168 |
+
hf_token = gr.Textbox(label="HF Token", type="password")
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 169 |
|
| 170 |
+
with gr.Row():
|
| 171 |
+
max_new_tokens = gr.Slider(64, 1024, value=512, step=16, label="Max New Tokens")
|
| 172 |
+
temperature = gr.Slider(0.0, 1.5, value=0.3, step=0.05, label="Temperature")
|
| 173 |
|
| 174 |
with gr.Row():
|
| 175 |
+
chunk_size = gr.Slider(500, 4000, value=1500, step=50, label="Chunk Size")
|
| 176 |
+
overlap = gr.Slider(0, 1000, value=200, step=50, label="Overlap")
|
| 177 |
+
max_chunks = gr.Slider(1, 20, value=5, step=1, label="Max Chunks")
|
| 178 |
+
|
| 179 |
with gr.Row():
|
| 180 |
+
min_pairs = gr.Slider(1, 10, value=3, step=1, label="Min Pairs")
|
| 181 |
+
max_pairs = gr.Slider(1, 12, value=6, step=1, label="Max Pairs")
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 182 |
|
| 183 |
+
generate_btn = gr.Button("🚀 Generate Dataset")
|
| 184 |
+
status = gr.Markdown()
|
| 185 |
+
out_json = gr.File(label="JSON")
|
| 186 |
+
out_jsonl = gr.File(label="JSONL")
|
| 187 |
|
| 188 |
generate_btn.click(
|
| 189 |
fn=generate_dataset,
|
| 190 |
+
inputs=[pdf_files, task, preset_model, custom_model_id, hf_token,
|
| 191 |
+
chunk_size, overlap, max_chunks, max_new_tokens, temperature,
|
| 192 |
+
min_pairs, max_pairs],
|
| 193 |
+
outputs=[status, out_json, out_jsonl]
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 194 |
)
|
| 195 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 196 |
if __name__ == "__main__":
|
| 197 |
+
demo.queue().launch()
|
|
|