Views
No views yet
1from transformers import AutoTokenizer, AutoModelForCausalLM, LlamaForCausalLM,AddedToken
2import sys
3
4mode_id = "p208p2002/bloom-1b1-zh-error-correction-dpo"
5model: LlamaForCausalLM = AutoModelForCausalLM.from_pretrained("p208p2002/bloom-1b1-zh-error-correction-dpo")
6tokenizer = AutoTokenizer.from_pretrained("p208p2002/bloom-1b1-zh-error-correction-dpo")
7
8test_texts = [
9 "為了潔約能源請隨守關閉沒有使用的電器",
10 "今天新情很好",
11 "你快樂我也很高心",
12 "但不再算再找實習生了",
13 "今天太陽很大要注意篩傷",
14 "你要不要和我依起去台北",
15 "清晨六點終太陽會升起",
16 "傾城六點鐘太陽會升起",
17 "鍋馬路時你應該要注意虹綠燈",
18 "他正在學學彈吉他",
19 "下樓梯請注意階梯",
20 "此信件為系統自動發送之通知",
21 "此信件為系統自動發送知通知",
22 "如為誤傳也請立即刪除本郵件並通知寄件者"
23]
24for text in test_texts:
25 inputs = tokenizer(
26 f"{tokenizer.bos_token}{text} {tokenizer.eos_token}\n {tokenizer.bos_token}",
27 return_tensors="pt",
28 add_special_tokens=False
29 )["input_ids"]
30
31 out = model.generate(
32 inputs,
33 max_new_tokens=20,
34 )
35 decode_out = tokenizer.decode(out[0])
36
37 input_text,output_text = decode_out.split("\n")
38 input_text = input_text.strip()
39 output_text = output_text.strip()
40
41 print("input :",input_text)
42 print("output:",output_text)
43 print('-'*30)input: <s>為了潔約能源請隨守關閉沒有使用的電器 </s>
output: <s>為了節約能源請隨時關閉沒有使用的電器 </s>
------------------------------
input: <s>今天新情很好 </s>
output: <s>今天心情很好 </s>
------------------------------
input: <s>你快樂我也很高心 </s>
output: <s>你快樂我也很高興 </s>
------------------------------
input: <s>但不再算再找實習生了 </s>
output: <s>但不再去找實習生了 </s>
------------------------------
input: <s>今天太陽很大要注意篩傷 </s>
output: <s>今天太陽很大要注意一下 </s>
------------------------------
input: <s>你要不要和我依起去台北 </s>
output: <s>你要不要和我一起去台北 </s>
------------------------------
input: <s>清晨六點終太陽會升起 </s>
output: <s>清晨六點鐘太陽會升起 </s>
------------------------------
input: <s>傾城六點鐘太陽會升起 </s>
output: <s>凌晨六點鐘太陽會升起 </s>
------------------------------
input: <s>鍋馬路時你應該要注意虹綠燈 </s>
output: <s>過馬路時你應該要注意紅綠燈 </s>
------------------------------
input: <s>他正在學學彈吉他 </s>
output: <s>他正在學習彈吉他 </s>
------------------------------
input: <s>下樓梯請注意階梯 </s>
output: <s>下樓梯請注意階梯 </s>
------------------------------
input: <s>此信件為系統自動發送之通知 </s>
output: <s>此信件為系統自動發送之通知 </s>
------------------------------
input: <s>此信件為系統自動發送知通知 </s>
output: <s>此信件為系統自動發送通知 </s>
------------------------------
input: <s>如為誤傳也請立即刪除本郵件並通知寄件者 </s>
output: <s>如為誤傳也請立即刪除本郵件並通知寄件者 </s>
------------------------------
input : <s>為了潔約能源請隨守關閉沒有使用的電器 </s>
output: <s>為了節約能源請隨時關閉沒有使用的電器 </s>
------------------------------
input : <s>今天新情很好 </s>
output: <s>今天心情很好 </s>
------------------------------
input : <s>你快樂我也很高心 </s>
output: <s>你快樂我也很高興 </s>
------------------------------
input : <s>但不再算再找實習生了 </s>
output: <s>但不再去找實習生了 </s>
------------------------------
input : <s>今天太陽很大要注意篩傷 </s>
output: <s>今天太陽很大要注意一下 </s>
------------------------------
input : <s>你要不要和我依起去台北 </s>
output: <s>你要不要和我一起去台北 </s>
------------------------------
input : <s>清晨六點終太陽會升起 </s>
output: <s>清晨六點鐘太陽會升起 </s>
------------------------------
input : <s>傾城六點鐘太陽會升起 </s>
output: <s>凌晨六點鐘太陽會升起 </s>
------------------------------
input : <s>鍋馬路時你應該要注意虹綠燈 </s>
output: <s>過馬路時你應該要注意紅綠燈 </s>
------------------------------
input : <s>他正在學學彈吉他 </s>
output: <s>他正在學習彈吉他 </s>
------------------------------
input : <s>下樓梯請注意階梯 </s>
output: <s>下樓梯請注意階梯 </s>
------------------------------
input : <s>此信件為系統自動發送之通知 </s>
output: <s>此信件為系統自動發送之通知 </s>
------------------------------
input : <s>此信件為系統自動發送知通知 </s>
output: <s>此信件為系統自動發送通知 </s>
------------------------------
input : <s>如為誤傳也請立即刪除本郵件並通知寄件者 </s>
output: <s>如為誤傳也請立即刪除本郵件並通知寄件者 </s>
------------------------------