Nattapong Tapachoom
commited on
Commit
·
18f1382
1
Parent(s):
861a5b2
Add
Browse files- app.py +747 -3
- requirements.txt +7 -0
- sample_data.csv +6 -0
app.py
CHANGED
@@ -1,7 +1,751 @@
|
|
1 |
import gradio as gr
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2 |
|
3 |
-
|
4 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
5 |
|
6 |
-
demo = gr.Interface(fn=greet, inputs="text", outputs="text")
|
7 |
demo.launch()
|
|
|
1 |
import gradio as gr
|
2 |
+
import os
|
3 |
+
import json
|
4 |
+
import uuid
|
5 |
+
import re
|
6 |
+
from typing import List, Optional, Dict, Any
|
7 |
+
from pydantic import BaseModel, ValidationError
|
8 |
+
from datasets import load_dataset, Dataset, DatasetDict
|
9 |
+
import pandas as pd
|
10 |
+
import requests
|
11 |
+
from datetime import datetime
|
12 |
+
import hashlib
|
13 |
|
14 |
+
# 1. Dataset Schema
|
15 |
+
class DataSample(BaseModel):
|
16 |
+
id: str
|
17 |
+
context: str
|
18 |
+
question: str
|
19 |
+
options: Optional[List[str]] = None
|
20 |
+
answer: str
|
21 |
+
rationale: str
|
22 |
+
category: str
|
23 |
+
difficulty: str
|
24 |
+
source: str
|
25 |
+
language: str
|
26 |
+
|
27 |
+
# 2. Load dataset (local file หรือ Hugging Face)
|
28 |
+
def load_data(source_type, path_or_name):
|
29 |
+
try:
|
30 |
+
if source_type == "local":
|
31 |
+
if not os.path.exists(path_or_name):
|
32 |
+
raise FileNotFoundError(f"ไฟล์ {path_or_name} ไม่พบ")
|
33 |
+
|
34 |
+
ext = os.path.splitext(path_or_name)[-1].lower()
|
35 |
+
if ext == ".jsonl":
|
36 |
+
data = []
|
37 |
+
with open(path_or_name, 'r', encoding="utf-8") as f:
|
38 |
+
for line_num, line in enumerate(f, 1):
|
39 |
+
try:
|
40 |
+
data.append(json.loads(line.strip()))
|
41 |
+
except json.JSONDecodeError as e:
|
42 |
+
print(f"Warning: บรรทัด {line_num} มีข้อผิดพลาด JSON: {e}")
|
43 |
+
continue
|
44 |
+
elif ext == ".csv":
|
45 |
+
df = pd.read_csv(path_or_name, encoding="utf-8")
|
46 |
+
data = df.to_dict(orient="records")
|
47 |
+
elif ext == ".json":
|
48 |
+
with open(path_or_name, 'r', encoding="utf-8") as f:
|
49 |
+
raw_data = json.load(f)
|
50 |
+
data = raw_data if isinstance(raw_data, list) else [raw_data]
|
51 |
+
else:
|
52 |
+
raise ValueError(f"ไม่รองรับไฟล์ประเภท {ext}")
|
53 |
+
|
54 |
+
# แปลงเป็น DataSample objects
|
55 |
+
samples = []
|
56 |
+
for i, item in enumerate(data):
|
57 |
+
try:
|
58 |
+
# เติมค่า default ถ้าไม่มี
|
59 |
+
if 'id' not in item:
|
60 |
+
item['id'] = str(uuid.uuid4())
|
61 |
+
if 'source' not in item:
|
62 |
+
item['source'] = f"local_{os.path.basename(path_or_name)}"
|
63 |
+
if 'difficulty' not in item:
|
64 |
+
item['difficulty'] = "medium"
|
65 |
+
if 'language' not in item:
|
66 |
+
item['language'] = "th"
|
67 |
+
|
68 |
+
samples.append(DataSample(**item))
|
69 |
+
except ValidationError as e:
|
70 |
+
print(f"Warning: รายการที่ {i+1} ข้อมูลไม่ถูกต้อง: {e}")
|
71 |
+
continue
|
72 |
+
|
73 |
+
return samples
|
74 |
+
|
75 |
+
elif source_type == "hf":
|
76 |
+
try:
|
77 |
+
ds = load_dataset(path_or_name)
|
78 |
+
# หา split ที่มีข้อมูล
|
79 |
+
available_splits = list(ds.keys())
|
80 |
+
if not available_splits:
|
81 |
+
raise ValueError("ไม่พบข้อมูลใน dataset")
|
82 |
+
|
83 |
+
# ใช้ split แรกที่มีข้อมูล
|
84 |
+
split_name = available_splits[0]
|
85 |
+
data = ds[split_name]
|
86 |
+
|
87 |
+
samples = []
|
88 |
+
for i, item in enumerate(data):
|
89 |
+
try:
|
90 |
+
# แปลง HF format เป็น DataSample
|
91 |
+
sample_dict = dict(item)
|
92 |
+
|
93 |
+
# เติมค่า default
|
94 |
+
if 'id' not in sample_dict:
|
95 |
+
sample_dict['id'] = f"hf_{i}"
|
96 |
+
if 'source' not in sample_dict:
|
97 |
+
sample_dict['source'] = f"hf_{path_or_name}"
|
98 |
+
if 'difficulty' not in sample_dict:
|
99 |
+
sample_dict['difficulty'] = "medium"
|
100 |
+
if 'language' not in sample_dict:
|
101 |
+
sample_dict['language'] = "en"
|
102 |
+
|
103 |
+
samples.append(DataSample(**sample_dict))
|
104 |
+
except ValidationError as e:
|
105 |
+
print(f"Warning: รายการที่ {i+1} จาก HF ข้อมูลไม่ถูกต้อง: {e}")
|
106 |
+
continue
|
107 |
+
|
108 |
+
return samples
|
109 |
+
|
110 |
+
except Exception as e:
|
111 |
+
raise ValueError(f"ไม่สามารถโหลด HF dataset '{path_or_name}': {e}")
|
112 |
+
else:
|
113 |
+
raise ValueError("source_type ต้องเป็น 'local' หรือ 'hf'")
|
114 |
+
|
115 |
+
except Exception as e:
|
116 |
+
raise Exception(f"ข้อผิดพลาดในการโหลดข้อมูล: {e}")
|
117 |
+
|
118 |
+
# 3. LLM API Integration (รองรับหลาย provider)
|
119 |
+
class LLMProvider:
|
120 |
+
def __init__(self, provider="ollama", api_key=None, base_url="http://localhost:11434"):
|
121 |
+
self.provider = provider
|
122 |
+
self.api_key = api_key
|
123 |
+
self.base_url = base_url
|
124 |
+
|
125 |
+
def generate(self, prompt, model="llama3.2", temperature=0.7, max_tokens=1000):
|
126 |
+
try:
|
127 |
+
if self.provider == "ollama":
|
128 |
+
return self._generate_ollama(prompt, model, temperature, max_tokens)
|
129 |
+
elif self.provider == "openai":
|
130 |
+
return self._generate_openai(prompt, model, temperature, max_tokens)
|
131 |
+
elif self.provider == "huggingface":
|
132 |
+
return self._generate_huggingface(prompt, model, temperature, max_tokens)
|
133 |
+
else:
|
134 |
+
raise ValueError(f"ไม่รองรับ provider: {self.provider}")
|
135 |
+
except Exception as e:
|
136 |
+
return f"Error generating response: {e}"
|
137 |
+
|
138 |
+
def _generate_ollama(self, prompt, model, temperature, max_tokens):
|
139 |
+
response = requests.post(
|
140 |
+
f"{self.base_url}/api/generate",
|
141 |
+
json={
|
142 |
+
"model": model,
|
143 |
+
"prompt": prompt,
|
144 |
+
"stream": False,
|
145 |
+
"options": {
|
146 |
+
"temperature": temperature,
|
147 |
+
"num_predict": max_tokens
|
148 |
+
}
|
149 |
+
}
|
150 |
+
)
|
151 |
+
response.raise_for_status()
|
152 |
+
return response.json()["response"]
|
153 |
+
|
154 |
+
def _generate_openai(self, prompt, model, temperature, max_tokens):
|
155 |
+
import openai
|
156 |
+
if self.api_key:
|
157 |
+
openai.api_key = self.api_key
|
158 |
+
|
159 |
+
response = openai.ChatCompletion.create(
|
160 |
+
model=model,
|
161 |
+
messages=[{"role": "user", "content": prompt}],
|
162 |
+
temperature=temperature,
|
163 |
+
max_tokens=max_tokens
|
164 |
+
)
|
165 |
+
return response.choices[0].message.content
|
166 |
+
|
167 |
+
def _generate_huggingface(self, prompt, model, temperature, max_tokens):
|
168 |
+
headers = {"Authorization": f"Bearer {self.api_key}"}
|
169 |
+
response = requests.post(
|
170 |
+
f"https://api-inference.huggingface.co/models/{model}",
|
171 |
+
headers=headers,
|
172 |
+
json={
|
173 |
+
"inputs": prompt,
|
174 |
+
"parameters": {
|
175 |
+
"temperature": temperature,
|
176 |
+
"max_new_tokens": max_tokens
|
177 |
+
}
|
178 |
+
}
|
179 |
+
)
|
180 |
+
response.raise_for_status()
|
181 |
+
result = response.json()
|
182 |
+
if isinstance(result, list) and len(result) > 0:
|
183 |
+
return result[0].get("generated_text", "").replace(prompt, "").strip()
|
184 |
+
return str(result)
|
185 |
+
|
186 |
+
# 4. Dataset Generation & Augmentation
|
187 |
+
def generate_new_samples(samples: List[DataSample], llm_provider: LLMProvider,
|
188 |
+
generation_type="augment", n_generate=1, custom_prompt=""):
|
189 |
+
"""
|
190 |
+
generation_type: 'augment', 'roleplay', 'topic_conditioning', 'self_critique'
|
191 |
+
"""
|
192 |
+
generated_samples = []
|
193 |
+
|
194 |
+
for sample in samples[:5]: # จำกัดแค่ 5 samples แรกเพื่อทดสอบ
|
195 |
+
for _ in range(n_generate):
|
196 |
+
try:
|
197 |
+
if generation_type == "augment":
|
198 |
+
prompt = f"""
|
199 |
+
Based on this context and question, create a similar but different scenario:
|
200 |
+
|
201 |
+
Context: {sample.context}
|
202 |
+
Question: {sample.question}
|
203 |
+
Answer: {sample.answer}
|
204 |
+
Rationale: {sample.rationale}
|
205 |
+
|
206 |
+
Generate a new scenario in the same category ({sample.category}) with:
|
207 |
+
- Different context but similar moral/logical challenge
|
208 |
+
- Appropriate question
|
209 |
+
- Clear answer
|
210 |
+
- Detailed rationale
|
211 |
+
|
212 |
+
Format as JSON:
|
213 |
+
{{
|
214 |
+
"context": "new context here",
|
215 |
+
"question": "new question here",
|
216 |
+
"answer": "new answer here",
|
217 |
+
"rationale": "detailed reasoning here"
|
218 |
+
}}"""
|
219 |
+
|
220 |
+
elif generation_type == "roleplay":
|
221 |
+
roles = ["ครูใหญ่", "หมอ", "นักบวช", "นักจิตวิทยา", "ผู้ปกครอง"]
|
222 |
+
role = roles[len(generated_samples) % len(roles)]
|
223 |
+
prompt = f"""
|
224 |
+
คุณคือ{role} กำลังให้คำแนะนำเกี่ยวกับสถานการณ์นี้:
|
225 |
+
|
226 |
+
Context: {sample.context}
|
227 |
+
Question: {sample.question}
|
228 |
+
|
229 |
+
ในฐานะ{role} จงสร้างคำตอบและเหตุผลที่เหมาะสมจากมุมมองของบทบาทนี้
|
230 |
+
|
231 |
+
Format as JSON:
|
232 |
+
{{
|
233 |
+
"context": "{sample.context}",
|
234 |
+
"question": "{sample.question}",
|
235 |
+
"answer": "คำตอบในฐานะ{role}",
|
236 |
+
"rationale": "เหตุผลจากมุมมอง{role}"
|
237 |
+
}}"""
|
238 |
+
|
239 |
+
elif generation_type == "topic_conditioning":
|
240 |
+
topics = ["ปัญหาวัยรุ่น", "ความยากจน", "เทคโนโลยี", "สิ่งแวดล้อม", "คร���บครัว"]
|
241 |
+
topic = topics[len(generated_samples) % len(topics)]
|
242 |
+
prompt = f"""
|
243 |
+
สร้างสถานการณ์ใหม่ในหัวข้อ "{topic}" ที่มีความซับซ้อนทางจริยธรรมคล้ายกับ:
|
244 |
+
|
245 |
+
Original context: {sample.context}
|
246 |
+
Category: {sample.category}
|
247 |
+
|
248 |
+
สร้างสถานการณ์ใหม่ที่เกี่ยวข้องกับ{topic}:
|
249 |
+
|
250 |
+
Format as JSON:
|
251 |
+
{{
|
252 |
+
"context": "สถานการณ์เกี่ยวกับ{topic}",
|
253 |
+
"question": "คำถามที่เหมาะสม",
|
254 |
+
"answer": "คำตอบที่ดีที่สุด",
|
255 |
+
"rationale": "เหตุผลโดยละเอียด"
|
256 |
+
}}"""
|
257 |
+
|
258 |
+
elif generation_type == "self_critique":
|
259 |
+
prompt = f"""
|
260 |
+
Analyze and improve this moral reasoning scenario:
|
261 |
+
|
262 |
+
Context: {sample.context}
|
263 |
+
Question: {sample.question}
|
264 |
+
Answer: {sample.answer}
|
265 |
+
Rationale: {sample.rationale}
|
266 |
+
|
267 |
+
1. First, critique the reasoning - what could be improved?
|
268 |
+
2. Then provide an enhanced version with better rationale
|
269 |
+
|
270 |
+
Format as JSON:
|
271 |
+
{{
|
272 |
+
"context": "{sample.context}",
|
273 |
+
"question": "{sample.question}",
|
274 |
+
"answer": "improved answer",
|
275 |
+
"rationale": "enhanced rationale with deeper analysis"
|
276 |
+
}}"""
|
277 |
+
|
278 |
+
else: # custom prompt
|
279 |
+
prompt = custom_prompt.format(**sample.dict())
|
280 |
+
|
281 |
+
# Generate ด้วย LLM
|
282 |
+
response = llm_provider.generate(prompt)
|
283 |
+
|
284 |
+
# Parse JSON response
|
285 |
+
try:
|
286 |
+
# ลองหา JSON ใน response
|
287 |
+
json_match = re.search(r'\{.*\}', response, re.DOTALL)
|
288 |
+
if json_match:
|
289 |
+
json_str = json_match.group()
|
290 |
+
parsed_data = json.loads(json_str)
|
291 |
+
|
292 |
+
# สร้าง DataSample ใหม่
|
293 |
+
new_sample = DataSample(
|
294 |
+
id=str(uuid.uuid4()),
|
295 |
+
context=parsed_data.get("context", sample.context),
|
296 |
+
question=parsed_data.get("question", sample.question),
|
297 |
+
answer=parsed_data.get("answer", sample.answer),
|
298 |
+
rationale=parsed_data.get("rationale", sample.rationale),
|
299 |
+
category=sample.category,
|
300 |
+
difficulty=sample.difficulty,
|
301 |
+
source=f"generated_{generation_type}",
|
302 |
+
language=sample.language,
|
303 |
+
options=sample.options
|
304 |
+
)
|
305 |
+
generated_samples.append(new_sample)
|
306 |
+
|
307 |
+
except (json.JSONDecodeError, KeyError) as e:
|
308 |
+
print(f"Warning: ไม่สามารถ parse JSON response: {e}")
|
309 |
+
continue
|
310 |
+
|
311 |
+
except Exception as e:
|
312 |
+
print(f"Warning: ไม่สามารถ generate sample: {e}")
|
313 |
+
continue
|
314 |
+
|
315 |
+
return generated_samples
|
316 |
+
|
317 |
+
# 5. Post-processing & Filtering
|
318 |
+
def remove_duplicates(samples: List[DataSample]) -> List[DataSample]:
|
319 |
+
"""Remove duplicate samples based on context and question"""
|
320 |
+
seen = set()
|
321 |
+
unique = []
|
322 |
+
for s in samples:
|
323 |
+
# สร้าง hash จาก context + question
|
324 |
+
content_hash = hashlib.md5(f"{s.context.lower().strip()}{s.question.lower().strip()}".encode()).hexdigest()
|
325 |
+
if content_hash not in seen:
|
326 |
+
unique.append(s)
|
327 |
+
seen.add(content_hash)
|
328 |
+
return unique
|
329 |
+
|
330 |
+
def syntax_check(samples: List[DataSample]) -> List[DataSample]:
|
331 |
+
"""Check for basic syntax issues and filter out problematic samples"""
|
332 |
+
valid_samples = []
|
333 |
+
for s in samples:
|
334 |
+
# Check ว่ามีเนื้อหาครบถ้วน
|
335 |
+
if (len(s.context.strip()) < 10 or
|
336 |
+
len(s.question.strip()) < 5 or
|
337 |
+
len(s.answer.strip()) < 3 or
|
338 |
+
len(s.rationale.strip()) < 10):
|
339 |
+
continue
|
340 |
+
|
341 |
+
# Check ว่าไม่มี placeholder text
|
342 |
+
placeholder_texts = ["[ใส่ข้อความ]", "TODO", "xxx", "example", "sample"]
|
343 |
+
has_placeholder = any(placeholder in s.context.lower() or
|
344 |
+
placeholder in s.question.lower() or
|
345 |
+
placeholder in s.answer.lower() or
|
346 |
+
placeholder in s.rationale.lower()
|
347 |
+
for placeholder in placeholder_texts)
|
348 |
+
if has_placeholder:
|
349 |
+
continue
|
350 |
+
|
351 |
+
valid_samples.append(s)
|
352 |
+
|
353 |
+
return valid_samples
|
354 |
+
|
355 |
+
def difficulty_assessment(samples: List[DataSample]) -> List[DataSample]:
|
356 |
+
"""Assess and update difficulty based on heuristics"""
|
357 |
+
for sample in samples:
|
358 |
+
# Heuristic based on token count and complexity
|
359 |
+
total_tokens = len(sample.context.split()) + len(sample.question.split()) + len(sample.rationale.split())
|
360 |
+
|
361 |
+
# Count complexity indicators
|
362 |
+
complexity_indicators = [
|
363 |
+
"ถ้า", "แต่", "อย่างไรก็ตาม", "ในขณะที่", "แม้ว่า",
|
364 |
+
"เนื่องจาก", "ดังนั้น", "เพราะว่า", "หากว่า", "เว้นแต่"
|
365 |
+
]
|
366 |
+
complexity_count = sum(1 for indicator in complexity_indicators
|
367 |
+
if indicator in sample.context or indicator in sample.rationale)
|
368 |
+
|
369 |
+
# Assess difficulty
|
370 |
+
if total_tokens < 50 and complexity_count < 2:
|
371 |
+
sample.difficulty = "easy"
|
372 |
+
elif total_tokens > 150 or complexity_count > 4:
|
373 |
+
sample.difficulty = "hard"
|
374 |
+
else:
|
375 |
+
sample.difficulty = "medium"
|
376 |
+
|
377 |
+
return samples
|
378 |
+
|
379 |
+
def translate_to_multilingual(samples: List[DataSample], llm_provider: LLMProvider, target_lang="en") -> List[DataSample]:
|
380 |
+
"""Translate samples to target language"""
|
381 |
+
translated = []
|
382 |
+
|
383 |
+
for sample in samples[:3]: # จำกัดเพื่อทดสอบ
|
384 |
+
if sample.language == target_lang:
|
385 |
+
continue
|
386 |
+
|
387 |
+
try:
|
388 |
+
prompt = f"""
|
389 |
+
Translate this moral reasoning scenario to {target_lang}:
|
390 |
+
|
391 |
+
Context: {sample.context}
|
392 |
+
Question: {sample.question}
|
393 |
+
Answer: {sample.answer}
|
394 |
+
Rationale: {sample.rationale}
|
395 |
+
|
396 |
+
Maintain the moral and cultural context appropriately.
|
397 |
+
|
398 |
+
Format as JSON:
|
399 |
+
{{
|
400 |
+
"context": "translated context",
|
401 |
+
"question": "translated question",
|
402 |
+
"answer": "translated answer",
|
403 |
+
"rationale": "translated rationale"
|
404 |
+
}}"""
|
405 |
+
|
406 |
+
response = llm_provider.generate(prompt)
|
407 |
+
|
408 |
+
# Parse JSON
|
409 |
+
json_match = re.search(r'\{.*\}', response, re.DOTALL)
|
410 |
+
if json_match:
|
411 |
+
parsed_data = json.loads(json_match.group())
|
412 |
+
|
413 |
+
translated_sample = DataSample(
|
414 |
+
id=f"{sample.id}_{target_lang}",
|
415 |
+
context=parsed_data["context"],
|
416 |
+
question=parsed_data["question"],
|
417 |
+
answer=parsed_data["answer"],
|
418 |
+
rationale=parsed_data["rationale"],
|
419 |
+
category=sample.category,
|
420 |
+
difficulty=sample.difficulty,
|
421 |
+
source=f"{sample.source}_translated",
|
422 |
+
language=target_lang,
|
423 |
+
options=sample.options
|
424 |
+
)
|
425 |
+
translated.append(translated_sample)
|
426 |
+
|
427 |
+
except Exception as e:
|
428 |
+
print(f"Warning: ไม่สามารถแปลภาษา sample {sample.id}: {e}")
|
429 |
+
continue
|
430 |
+
|
431 |
+
return translated
|
432 |
+
|
433 |
+
def add_multiple_choice_options(samples: List[DataSample], llm_provider: LLMProvider) -> List[DataSample]:
|
434 |
+
"""Add multiple choice options to samples"""
|
435 |
+
for sample in samples[:3]: # จำกัดเพื่อทดสอบ
|
436 |
+
if sample.options: # มี options อยู่แล้ว
|
437 |
+
continue
|
438 |
+
|
439 |
+
try:
|
440 |
+
prompt = f"""
|
441 |
+
Create 4 multiple choice options for this scenario, with one correct answer:
|
442 |
+
|
443 |
+
Context: {sample.context}
|
444 |
+
Question: {sample.question}
|
445 |
+
Correct Answer: {sample.answer}
|
446 |
+
|
447 |
+
Generate 3 plausible but incorrect options and include the correct answer.
|
448 |
+
|
449 |
+
Format as JSON array:
|
450 |
+
["option A", "option B", "option C", "option D"]
|
451 |
+
|
452 |
+
Make sure the correct answer ({sample.answer}) is included as one of the options.
|
453 |
+
"""
|
454 |
+
|
455 |
+
response = llm_provider.generate(prompt)
|
456 |
+
|
457 |
+
# Parse JSON array
|
458 |
+
json_match = re.search(r'\[.*\]', response, re.DOTALL)
|
459 |
+
if json_match:
|
460 |
+
options = json.loads(json_match.group())
|
461 |
+
if len(options) == 4:
|
462 |
+
sample.options = options
|
463 |
+
|
464 |
+
except Exception as e:
|
465 |
+
print(f"Warning: ไม่สามารถสร้าง multiple choice สำหรับ {sample.id}: {e}")
|
466 |
+
continue
|
467 |
+
|
468 |
+
return samples
|
469 |
+
|
470 |
+
# 6. Export & Visualization
|
471 |
+
def export_dataset(samples: List[DataSample], format_type="csv", output_path="output"):
|
472 |
+
"""Export dataset ในรูปแบบต่างๆ"""
|
473 |
+
timestamp = datetime.now().strftime("%Y%m%d_%H%M%S")
|
474 |
+
|
475 |
+
if format_type == "csv":
|
476 |
+
df = pd.DataFrame([s.dict() for s in samples])
|
477 |
+
filename = f"{output_path}_{timestamp}.csv"
|
478 |
+
df.to_csv(filename, index=False, encoding="utf-8-sig")
|
479 |
+
return filename
|
480 |
+
|
481 |
+
elif format_type == "jsonl":
|
482 |
+
filename = f"{output_path}_{timestamp}.jsonl"
|
483 |
+
with open(filename, 'w', encoding="utf-8") as f:
|
484 |
+
for sample in samples:
|
485 |
+
f.write(json.dumps(sample.dict(), ensure_ascii=False) + "\n")
|
486 |
+
return filename
|
487 |
+
|
488 |
+
elif format_type == "hf_dataset":
|
489 |
+
# Create Hugging Face Dataset
|
490 |
+
data_dict = {key: [] for key in samples[0].dict().keys()}
|
491 |
+
for sample in samples:
|
492 |
+
sample_dict = sample.dict()
|
493 |
+
for key, value in sample_dict.items():
|
494 |
+
data_dict[key].append(value)
|
495 |
+
|
496 |
+
dataset = Dataset.from_dict(data_dict)
|
497 |
+
dirname = f"{output_path}_hf_{timestamp}"
|
498 |
+
dataset.save_to_disk(dirname)
|
499 |
+
return dirname
|
500 |
+
|
501 |
+
else:
|
502 |
+
raise ValueError(f"ไม่รองรับรูปแบบ: {format_type}")
|
503 |
+
|
504 |
+
def get_dataset_stats(samples: List[DataSample]) -> Dict[str, Any]:
|
505 |
+
"""สถิติของ dataset"""
|
506 |
+
if not samples:
|
507 |
+
return {"total": 0}
|
508 |
+
|
509 |
+
df = pd.DataFrame([s.dict() for s in samples])
|
510 |
+
|
511 |
+
stats = {
|
512 |
+
"total": len(samples),
|
513 |
+
"categories": df["category"].value_counts().to_dict(),
|
514 |
+
"difficulties": df["difficulty"].value_counts().to_dict(),
|
515 |
+
"languages": df["language"].value_counts().to_dict(),
|
516 |
+
"sources": df["source"].value_counts().to_dict(),
|
517 |
+
"avg_context_length": df["context"].str.len().mean(),
|
518 |
+
"avg_question_length": df["question"].str.len().mean(),
|
519 |
+
"avg_answer_length": df["answer"].str.len().mean(),
|
520 |
+
"avg_rationale_length": df["rationale"].str.len().mean(),
|
521 |
+
"with_options": sum(1 for s in samples if s.options is not None)
|
522 |
+
}
|
523 |
+
|
524 |
+
return stats
|
525 |
+
|
526 |
+
# 7. Main Workflow Function
|
527 |
+
def main_workflow(source_type, path_or_name, llm_provider_type, api_key, base_url,
|
528 |
+
generation_type, n_generate, custom_prompt, target_language,
|
529 |
+
add_multiple_choice, export_format):
|
530 |
+
try:
|
531 |
+
progress_text = "เริ่มต้น workflow...\n"
|
532 |
+
|
533 |
+
# 1. Load dataset
|
534 |
+
progress_text += "📂 กำลังโหลด dataset...\n"
|
535 |
+
samples = load_data(source_type, path_or_name)
|
536 |
+
progress_text += f"✅ โหลดสำเร็จ {len(samples)} samples\n"
|
537 |
+
|
538 |
+
# 2. Setup LLM
|
539 |
+
progress_text += f"🤖 กำลังตั้งค่า LLM ({llm_provider_type})...\n"
|
540 |
+
llm_provider = LLMProvider(
|
541 |
+
provider=llm_provider_type,
|
542 |
+
api_key=api_key if api_key else None,
|
543 |
+
base_url=base_url if base_url else "http://localhost:11434"
|
544 |
+
)
|
545 |
+
|
546 |
+
# 3. Generate new samples
|
547 |
+
if n_generate > 0:
|
548 |
+
progress_text += f"✨ กำลัง generate {n_generate} samples ใหม่ ({generation_type})...\n"
|
549 |
+
new_samples = generate_new_samples(samples, llm_provider, generation_type, n_generate, custom_prompt)
|
550 |
+
samples.extend(new_samples)
|
551 |
+
progress_text += f"✅ Generate สำเร็จ {len(new_samples)} samples ใหม่\n"
|
552 |
+
|
553 |
+
# 4. Post-processing
|
554 |
+
progress_text += "🔧 กำลัง post-process...\n"
|
555 |
+
original_count = len(samples)
|
556 |
+
samples = remove_duplicates(samples)
|
557 |
+
progress_text += f" - ลบ duplicate: {original_count} -> {len(samples)}\n"
|
558 |
+
|
559 |
+
samples = syntax_check(samples)
|
560 |
+
progress_text += f" - syntax check: {len(samples)} samples ผ่าน\n"
|
561 |
+
|
562 |
+
samples = difficulty_assessment(samples)
|
563 |
+
progress_text += f" - ประเมิน difficulty เสร็จสิ้น\n"
|
564 |
+
|
565 |
+
# 5. Translation
|
566 |
+
if target_language and target_language != "none":
|
567 |
+
progress_text += f"🌐 กำลังแปลเป็น {target_language}...\n"
|
568 |
+
translated = translate_to_multilingual(samples, llm_provider, target_language)
|
569 |
+
samples.extend(translated)
|
570 |
+
progress_text += f"✅ แปลภาษาสำเร็จ {len(translated)} samples\n"
|
571 |
+
|
572 |
+
# 6. Add multiple choice
|
573 |
+
if add_multiple_choice:
|
574 |
+
progress_text += "📝 กำลังเพิ่ม multiple choice options...\n"
|
575 |
+
samples = add_multiple_choice_options(samples, llm_provider)
|
576 |
+
progress_text += "✅ เพิ่ม multiple choice เสร็จสิ้น\n"
|
577 |
+
|
578 |
+
# 7. Export
|
579 |
+
progress_text += f"💾 กำลัง export เป็น {export_format}...\n"
|
580 |
+
output_file = export_dataset(samples, export_format)
|
581 |
+
progress_text += f"✅ Export สำเร็จ: {output_file}\n"
|
582 |
+
|
583 |
+
# 8. Stats
|
584 |
+
stats = get_dataset_stats(samples)
|
585 |
+
progress_text += "\n📊 สถิติ Dataset:\n"
|
586 |
+
progress_text += f" - จำนวนทั้งหมด: {stats['total']}\n"
|
587 |
+
progress_text += f" - Categories: {stats['categories']}\n"
|
588 |
+
progress_text += f" - Difficulties: {stats['difficulties']}\n"
|
589 |
+
progress_text += f" - Languages: {stats['languages']}\n"
|
590 |
+
progress_text += f" - มี Multiple Choice: {stats['with_options']}\n"
|
591 |
+
|
592 |
+
return progress_text, pd.DataFrame([s.dict() for s in samples]).head(10).to_html()
|
593 |
+
|
594 |
+
except Exception as e:
|
595 |
+
error_text = f"❌ เกิดข้อผิดพลาด: {str(e)}"
|
596 |
+
return error_text, ""
|
597 |
+
|
598 |
+
# 8. Gradio Interface
|
599 |
+
with gr.Blocks(title="Dataset Generator System", theme=gr.themes.Soft()) as demo:
|
600 |
+
gr.Markdown("# 🤖 ระบบ Generate Dataset จากโมเดล AI")
|
601 |
+
gr.Markdown("ระบบสำหรับสร้าง, ขยาย, และประมวลผล dataset ด้วย AI models")
|
602 |
+
|
603 |
+
with gr.Tab("📂 Dataset Input"):
|
604 |
+
with gr.Row():
|
605 |
+
source_type = gr.Radio(
|
606 |
+
["local", "hf"],
|
607 |
+
label="ประเภทแหล่งข้อมูล",
|
608 |
+
info="local = ไฟล์ในเครื่อง, hf = Hugging Face dataset",
|
609 |
+
value="local"
|
610 |
+
)
|
611 |
+
path_or_name = gr.Textbox(
|
612 |
+
label="Path หรือ Dataset Name",
|
613 |
+
placeholder="เช่น data.csv หรือ microsoft/DialoGPT-medium",
|
614 |
+
info="สำหรับ local: ใส่ path ไฟล์ (.csv, .jsonl, .json) / สำหรับ HF: ใส่ชื่อ dataset"
|
615 |
+
)
|
616 |
+
|
617 |
+
with gr.Tab("🤖 LLM Settings"):
|
618 |
+
with gr.Row():
|
619 |
+
llm_provider_type = gr.Dropdown(
|
620 |
+
["ollama", "openai", "huggingface"],
|
621 |
+
label="LLM Provider",
|
622 |
+
value="ollama",
|
623 |
+
info="เลือกผู้ให้บริการ LLM"
|
624 |
+
)
|
625 |
+
api_key = gr.Textbox(
|
626 |
+
label="API Key (ถ้าจำเป็น)",
|
627 |
+
type="password",
|
628 |
+
placeholder="สำหรับ OpenAI หรือ HuggingFace"
|
629 |
+
)
|
630 |
+
base_url = gr.Textbox(
|
631 |
+
label="Base URL",
|
632 |
+
value="http://localhost:11434",
|
633 |
+
info="สำหรับ Ollama หรือ local LLM server"
|
634 |
+
)
|
635 |
+
|
636 |
+
with gr.Tab("✨ Generation Settings"):
|
637 |
+
with gr.Row():
|
638 |
+
generation_type = gr.Dropdown(
|
639 |
+
["augment", "roleplay", "topic_conditioning", "self_critique", "custom"],
|
640 |
+
label="ประเภทการ Generate",
|
641 |
+
value="augment",
|
642 |
+
info="วิธีการสร้างข้อมูลใหม่"
|
643 |
+
)
|
644 |
+
n_generate = gr.Slider(
|
645 |
+
1, 5, value=1, step=1,
|
646 |
+
label="จำนวนรอบ Generate",
|
647 |
+
info="จำนวน samples ใหม่ที่จะสร้างต่อ original sample"
|
648 |
+
)
|
649 |
+
|
650 |
+
custom_prompt = gr.Textbox(
|
651 |
+
label="Custom Prompt (ถ้าเลือก custom)",
|
652 |
+
placeholder="ใช้ {context}, {question}, {answer} เป็น placeholder",
|
653 |
+
lines=3,
|
654 |
+
visible=False
|
655 |
+
)
|
656 |
+
|
657 |
+
def update_custom_prompt_visibility(gen_type):
|
658 |
+
return gr.update(visible=(gen_type == "custom"))
|
659 |
+
|
660 |
+
generation_type.change(
|
661 |
+
update_custom_prompt_visibility,
|
662 |
+
inputs=[generation_type],
|
663 |
+
outputs=[custom_prompt]
|
664 |
+
)
|
665 |
+
|
666 |
+
with gr.Tab("🔧 Post-processing"):
|
667 |
+
with gr.Row():
|
668 |
+
target_language = gr.Dropdown(
|
669 |
+
["none", "en", "th", "zh", "ja"],
|
670 |
+
label="แปลภาษา",
|
671 |
+
value="none",
|
672 |
+
info="แปลเป็นภาษาเป้าหมาย (none = ไม่แปล)"
|
673 |
+
)
|
674 |
+
add_multiple_choice = gr.Checkbox(
|
675 |
+
label="เพิ่ม Multiple Choice Options",
|
676 |
+
value=False,
|
677 |
+
info="สร้างตัวเลือกผิดสำหรับทำ multiple choice"
|
678 |
+
)
|
679 |
+
|
680 |
+
with gr.Tab("💾 Export Settings"):
|
681 |
+
export_format = gr.Dropdown(
|
682 |
+
["csv", "jsonl", "hf_dataset"],
|
683 |
+
label="รูปแบบ Export",
|
684 |
+
value="csv",
|
685 |
+
info="รูปแบบไฟล์ที่ต้องการ export"
|
686 |
+
)
|
687 |
+
|
688 |
+
with gr.Row():
|
689 |
+
run_btn = gr.Button("🚀 เริ่มต้น Workflow", variant="primary", size="lg")
|
690 |
+
clear_btn = gr.Button("🗑️ ล้างข้อมูล", variant="secondary")
|
691 |
+
|
692 |
+
with gr.Tab("📊 ผลลัพธ์"):
|
693 |
+
progress_output = gr.Textbox(
|
694 |
+
label="สถานะ",
|
695 |
+
lines=15,
|
696 |
+
max_lines=20,
|
697 |
+
interactive=False,
|
698 |
+
show_copy_button=True
|
699 |
+
)
|
700 |
+
|
701 |
+
preview_output = gr.HTML(
|
702 |
+
label="ตัวอย่างข้อมูล (10 รายการแรก)"
|
703 |
+
)
|
704 |
+
|
705 |
+
# Event handlers
|
706 |
+
run_btn.click(
|
707 |
+
fn=main_workflow,
|
708 |
+
inputs=[
|
709 |
+
source_type, path_or_name, llm_provider_type, api_key, base_url,
|
710 |
+
generation_type, n_generate, custom_prompt, target_language,
|
711 |
+
add_multiple_choice, export_format
|
712 |
+
],
|
713 |
+
outputs=[progress_output, preview_output]
|
714 |
+
)
|
715 |
+
|
716 |
+
clear_btn.click(
|
717 |
+
lambda: ("", ""),
|
718 |
+
outputs=[progress_output, preview_output]
|
719 |
+
)
|
720 |
+
|
721 |
+
# ตัวอย่าง dataset schema
|
722 |
+
with gr.Tab("📋 ตัวอย่าง Dataset Schema"):
|
723 |
+
gr.Markdown("""
|
724 |
+
## Schema ของ Dataset
|
725 |
+
|
726 |
+
| Field | ประเภท | อธิบาย |
|
727 |
+
|-------|--------|--------|
|
728 |
+
| id | string | รหัสเฉพาะของ sample |
|
729 |
+
| context | string | บริบท/สถานการณ์ |
|
730 |
+
| question | string | คำถาม |
|
731 |
+
| options | list | ตัวเลือก (สำหรับ multiple choice) |
|
732 |
+
| answer | string | คำตอบที่ถูกต้อง |
|
733 |
+
| rationale | string | เหตุผล/คำอธิบาย |
|
734 |
+
| category | string | หมวดหมู่ |
|
735 |
+
| difficulty | string | ระดับความยาก (easy/medium/hard) |
|
736 |
+
| source | string | แหล่งที่มาของข้อมูล |
|
737 |
+
| language | string | ภาษา (th/en/zh/ja) |
|
738 |
+
|
739 |
+
## ตัวอย่างไฟล์ CSV:
|
740 |
+
```csv
|
741 |
+
id,context,question,answer,rationale,category,difficulty,source,language
|
742 |
+
1,"นักเรียนคนหนึ่งเห็นเพื่อนทำโกง","ควรรายงานครูหรือไม่","ควรรายงาน","เพื่อความยุติธรรม","การศึกษา","medium","manual","th"
|
743 |
+
```
|
744 |
+
|
745 |
+
## ตัวอย่างไฟล์ JSONL:
|
746 |
+
```json
|
747 |
+
{"id": "1", "context": "นักเรียนคนหนึ่งเห็นเพื่อนทำโกง", "question": "ควรรายงานครูหรือไม่", "answer": "ควรรายงาน", "rationale": "เพื่อความยุติธรรม", "category": "การศึกษา", "difficulty": "medium", "source": "manual", "language": "th"}
|
748 |
+
```
|
749 |
+
""")
|
750 |
|
|
|
751 |
demo.launch()
|
requirements.txt
ADDED
@@ -0,0 +1,7 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
gradio>=4.0.0
|
2 |
+
pandas>=1.5.0
|
3 |
+
datasets>=2.0.0
|
4 |
+
pydantic>=2.0.0
|
5 |
+
requests>=2.28.0
|
6 |
+
openai>=1.0.0
|
7 |
+
huggingface-hub>=0.16.0
|
sample_data.csv
ADDED
@@ -0,0 +1,6 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
id,context,question,answer,rationale,category,difficulty,source,language
|
2 |
+
1,"นักเรียนมัธยมคนหนึ่งเห็นเพื่อนสนิทกำลังลอกการบ้านจากเพื่อนคนอื่น ก่อนที่จะส่งครู","ควรแจ้งครูเรื่องการลอกการบ้านหรือไม่","ควรคุยกับเพื่อนก่อน แล้วค่อยพิจารณาแจ้งครูถ้าไม่หยุด","การคุยกับเพื่อนก่อนจะช่วยให้เขามีโอกาสแก้ไขตัวเอง และรักษาความสัมพันธ์มิตรภาพไว้ได้","การศึกษา","medium","manual","th"
|
3 |
+
2,"พนักงานคนหนึ่งพบว่าหัวหน้างานมีการทุจริตโดยการเบิกเงินเท็จ","ควรรายงานการทุจริตนี้หรือไม่","ควรรายงานผ่านช่องทางที่เหมาะสม","การทุจริตส่งผลเสียต่อองค์กรและสังคม การรายงานเป็นหน้าที่ของพลเมืองดี","การทำงาน","hard","manual","th"
|
4 |
+
3,"ครอบครัวหนึ่งมีปัญหาทางการเงิน ลูกคิดจะหยุดเรียนเพื่อไปทำงานช่วยครอบครัว","ลูกควรหยุดเรียนเพื่อทำงานหรือไม่","ไม่ควร ควรหาทางออกอื่น เช่น ขอทุนการศึกษา","การศึกษาเป็นรากฐานสำคัญของอนาคต ควรหาวิธีแก้ปัญหาการเงินโดยไม่ต้องเสียโอกาสทางการศึกษา","ครอบครัว","medium","manual","th"
|
5 |
+
4,"นักท่องเที่ยวเห็นคนท้องถิ่นทิ้งขยะลงในแม่น้ำ","ควรไปตักเตือนหรือไม่","ควรตักเตือนอย่างสุภาพและให้ความรู้","การรักษาสิ่งแวดล้อมเป็นหน้าที่ของทุกคน การให้ความรู้อย่างสุภาพจะมีประสิทธิภาพมากกว่าการตำหนิ","สิ่งแวดล้อม","easy","manual","th"
|
6 |
+
5,"หมอพบว่าผู้ป่วยมีโรคร้ายแรง แต่ผู้ป่วยไม่อยากให้ครอบครัวรู้","ควรบอกความจริงกับครอบครัวหรือไม่","ควรเคารพความประสงค์ของผู้ป่วย แต่แนะนำให้เล่าเอง","การเคารพสิทธิส่วนบุคคลของผู้ป่วยเป็นสิ่งสำคัญ แต่ควรให้คำปรึกษาเพื่อประโยชน์ในการรักษา","การแพทย์","hard","manual","th"
|