Table Understaing

## benchmark | model | gpt4o_mini_few_shots | ocnli | snli | nli_all | chinese_squad | announcement | langfuse | claim_ref | news | gov | wiki | pinzu | wiki_retrieval | taishin_false_name | taishin_long_answer | taishin_750 | creditcard | Overall | |:---------------|-----------------------:|--------:|-------:|----------:|----------------:|---------------:|-----------:|------------:|-------:|------:|-------:|--------:|-----------------:|---------------------:|----------------------:|--------------:|-------------:|----------:| | v11-large-2200 | 90.1 | 84.6 | 91 | 81.3 | 99.9 | 99.1 | 88.6 | 90 | 96.4 | 99.3 | 99.2 | 77.2 | 99.6 | 98.3 | 89.2 | 83.7 | 84.8 | 91.3 | | v11-large-4400 | 90.9 | 84.9 | 91.4 | 85.1 | 99.9 | 99.6 | 93.6 | 90.6 | 97.5 | 99.5 | 99.2 | 79.3 | 99.8 | 99.1 | 92.6 | 79.1 | 85.1 | 92.2 | | v11.4-ft | 86.1 | 78.4 | 88.3 | 75.9 | 99.5 | 97.1 | 83.5 | 89.6 | 83.5 | 96.7 | 96.9 | 68.8 | 97.8 | 91.6 | 84.9 | 85.5 | 79.2 | 87.2 | | v11-large | 91.1 | 84.8 | 91.5 | 86.5 | 99.9 | 99.7 | 95.1 | 90.2 | 98.5 | 99.7 | 99.4 | 79.7 | 99.7 | 99.3 | 92.6 | 77.7 | 87.5 | 92.5 | | v12-large-1600 | 91.2 | 84.8 | 91.5 | 84.8 | 99.9 | 99.2 | 91.8 | 90.9 | 97.7 | 99.6 | 99.4 | 76.3 | 99.6 | 98.7 | 92.1 | 75.3 | 89.2 | 91.9 | ### /volume/payo-ws/ckpt/nli/puff-large-v1-2024-10-30T18:04:19/checkpoint-7100 faq no chunk ``` PearsonRResult(statistic=0.656234531472081, pvalue=0.0) Accuracy: 0.81 ``` km no chunk ``` PearsonRResult(statistic=0.7753723906614198, pvalue=0.0) Accuracy: 0.87 ``` km with claim ``` PearsonRResult(statistic=0.7082363940887366, pvalue=3.5291277593031786e-274) Accuracy: 0.85 [0.03528843820095062, 0.03528843820095062, 0.03528843820095062, 0.001619952847249806, 0.001619952847249806, 0.001619952847249806, 0.05196159705519676, 0.05196159705519676, 0.05196159705519676, 0.9949859380722046] ``` langfuse ``` PearsonRResult(statistic=0.8369240725098342, pvalue=0.0) Accuracy: 0.92 [0.9875255823135376, 0.9893504977226257, 0.9868671894073486, 0.9850291609764099, 0.9881644248962402, 0.994175910949707, 0.04274369031190872, 0.004386087879538536, 0.0018825802253559232, 0.0016156220808625221] ``` ### /volume/payo-ws/ckpt/nli/puff-large-v1-2024-10-29T18:37:53/checkpoint-4100 km no chunk ``` PearsonRResult(statistic=0.7256796238701599, pvalue=2.720945929744198e-294) Accuracy: 0.83 ``` claim ``` PearsonRResult(statistic=0.6967601432775058, pvalue=9.678379222895196e-262) Accuracy: 0.84 ``` langfuse ``` PearsonRResult(statistic=0.8419327755963593, pvalue=0.0) Accuracy: 0.91~~~~ ``` ## checkpoint-20100 km no chunk ``` PearsonRResult(statistic=0.6598852627623158, pvalue=1.7225354670446823e-225) Accuracy: 0.81 ``` KM claim ``` PearsonRResult(statistic=0.6201147210765848, pvalue=9.350580887453429e-192) Accuracy: 0.80 ``` langfuse no chunk ``` PearsonRResult(statistic=0.9144937195878315, pvalue=0.0) Accuracy: 0.96 ``` ## checkpoint-15700 KM no chunk ``` PearsonRResult(statistic=0.6582462822384503, pvalue=5.351446090187359e-224) Accuracy: 0.81 ``` langfuse no chunk ``` PearsonRResult(statistic=0.9089500150218499, pvalue=0.0) Accuracy: 0.95 ``` ## /volume/payo-ws/ckpt/nli/puff-large-v1-2024-10-24T17:28:53/checkpoint-4100 (v4-large) KM no chunk ``` PearsonRResult(statistic=0.7249082517890411, pvalue=2.27482194401332e-293) Accuracy: 0.84 ``` langfuse no chunk ``` PearsonRResult(statistic=0.8773112307940885, pvalue=0.0) Accuracy: 0.93 ``` ### v3 large KM no chunk ``` PearsonRResult(statistic=0.69097326387628, pvalue=1.0867846480569673e-255) Accuracy: 0.84 ``` langfuse no chunk ``` PearsonRResult(statistic=0.6475688471978921, pvalue=2.4842857610860996e-176) Accuracy: 0.83 ``` ## langfuse_12438_result.json v2-base ``` PearsonRResult(statistic=0.4368711761428752, pvalue=0.0) Accuracy: 0.71 ``` spec1 ``` PearsonRResult(statistic=-0.2828778654611658, pvalue=1.5028941106209036e-227) Accuracy: 0.38 ``` # loki log gpt4o as label ## factuality_2024-09-07_13-52-13_2024-09-16_13-52-13.json ``` NLI model Accuracy: 0.57 Spec1 model Accuracy: 0.64 ``` # experiment result ### v3-large ``` PearsonRResult(statistic=0.69097326387628, pvalue=1.0867846480569673e-255) Accuracy: 0.84 ### v2 base max + mean ``` PearsonRResult(statistic=0.47013075536984783, pvalue=1.2081010079907958e-99) Accuracy: 0.72 ``` max + top3 ``` PearsonRResult(statistic=0.3872308780997303, pvalue=1.80621643296728e-65) Accuracy: 0.71 ``` top3 + top3 ```PearsonRResult(statistic=0.25122498615717076, pvalue=2.6139759884194117e-27) Accuracy: 0.66 ``` top3 + mean ``` PearsonRResult(statistic=0.28334142977732224, pvalue=1.3959213002655255e-34) Accuracy: 0.64 ``` no chunk ``` PearsonRResult(statistic=0.6239717563578638, pvalue=8.169351755957063e-195) Accuracy: 0.81 ``` chunking ``` PearsonRResult(statistic=0.4279785170186318, pvalue=4.4716880805956565e-81) Accuracy: 0.68 ``` chunking 400 ```PearsonRResult(statistic=0.5975052455676927, pvalue=1.1797038312395435e-174) Accuracy: 0.80 ``` #### v1-large no chunk checkpoint 2000 ``` PearsonRResult(statistic=0.5948713500510547, pvalue=9.463585599830867e-173) Accuracy: 0.79 ``` chunking ``` PearsonRResult(statistic=0.41983997689812447, pvalue=8.769467692482014e-78) Accuracy: 0.71 ``` own claims + max + mean ``` PearsonRResult(statistic=0.5278539062255798, pvalue=1.0854186885239614e-129) Accuracy: 0.77 ``` max + top3 ``` PearsonRResult(statistic=0.5045931358872306, pvalue=6.941369697226016e-117) Accuracy: 0.74 ``` max + top5 ``` PearsonRResult(statistic=0.5246839270945605, pvalue=6.916430788909701e-128) Accuracy: 0.75 ``` max + max ``` PearsonRResult(statistic=0.4467201855716201, pvalue=5.207001961531386e-89) Accuracy: 0.65 ``` top3 + top3 ``` PearsonRResult(statistic=0.31133939878472233, pvalue=9.367489909400129e-42) Accuracy: 0.64 ``` top3 + mean ``` PearsonRResult(statistic=0.37126143496622666, pvalue=6.295697997561968e-60) Accuracy: 0.64 ``` #### /volume/payo-ws/ckpt/nli/sbert-base-chinese-nli-2024-10-21T10:53:34/checkpoint-5000 no chunk: ``` PearsonRResult(statistic=0.5120872555892995, pvalue=6.66283915444525e-121) Accuracy: 0.75 ``` chunk: ``` PearsonRResult(statistic=0.44581249767534037, pvalue=1.295473358667981e-88) Accuracy: 0.73 ``` own claims: ``` PearsonRResult(statistic=0.41321610120251717, pvalue=3.602015411485474e-75) Accuracy: 0.65 ``` owm claims: max + mean ``` PearsonRResult(statistic=0.5633560851448431, pvalue=2.8561932828422374e-151) Accuracy: 0.78 checkpoint 10000 PearsonRResult(statistic=0.545807084033876, pvalue=2.796197013546571e-140) Accuracy: 0.77 ``` #### /volume/payo-ws/ckpt/nli/sbert-base-chinese-nli-2024-10-21T10:53:34/checkpoint-23000 23000 own claims: max + mean ``` PearsonRResult(statistic=0.49551774954347033, pvalue=3.756351703029471e-112) Accuracy: 0.70 ``` 23000 no chunk ``` PearsonRResult(statistic=0.5668482875465132, pvalue=1.5457084146982714e-153) Accuracy: 0.76 ``` 23000 chunking ``` PearsonRResult(statistic=0.46471893934923386, pvalue=4.117597278602082e-97) Accuracy: 0.67 ``` top3 + mean ``` PearsonRResult(statistic=0.46471893934923386, pvalue=4.117597278602082e-97) Accuracy: 0.67 ``` top3 + mean ``` PearsonRResult(statistic=0.3198823961488351, pvalue=4.184186912260021e-44) Accuracy: 0.65 ``` #### gpt4o-mini no chunk: ``` PearsonRResult(statistic=0.6752466050128739, pvalue=6.002115049511369e-240) Accuracy: 0.80 ``` 4o claim + top3 + mean ``` Total price: 86.75 PearsonRResult(statistic=0.5733233684876131, pvalue=8.200549086641558e-158) Accuracy: 0.63 ``` max + mean ``` PearsonRResult(statistic=0.6157018565057792, pvalue=2.6196903500707952e-188) Accuracy: 0.79 ``` max + max ``` PearsonRResult(statistic=0.5150038628184492, pvalue=1.7087332439137273e-122) Accuracy: 0.69 ``` max + top2 ``` PearsonRResult(statistic=0.5947458116405748, pvalue=1.1651417666552122e-172) Accuracy: 0.77 ``` max + top3 ``` PearsonRResult(statistic=0.629518376629177, pvalue=2.7392006512710353e-199) Accuracy: 0.81 ``` max + top5 ``` PearsonRResult(statistic=0.6258490430051427, pvalue=2.5579163663537776e-196) Accuracy: 0.80 ``` max + top7 ``` PearsonRResult(statistic=0.6185515203543619, pvalue=1.5788204438752347e-190) Accuracy: 0.79 ``` top3 + top3 ``` PearsonRResult(statistic=0.5742690928776922, pvalue=1.9116159916462235e-158) Accuracy: 0.65 ``` mean + top3 ``` PearsonRResult(statistic=0.4991421297854804, pvalue=5.029902635250684e-114) Accuracy: 0.63 ``` top2 + top3 ``` PearsonRResult(statistic=0.5936105170663719, pvalue=7.610250648828887e-172) Accuracy: 0.75 ``` mean + mean ``` PearsonRResult(statistic=0.516259320039194, pvalue=3.491837707220876e-123) Accuracy: 0.63 ### sbert-base-chinese-nli-2024-10-20T03:54:59 no chunk: ``` PearsonRResult(statistic=0.08496948383989364, pvalue=0.00030741070181362776) Accuracy: 0.60 ``` chunk: ``` PearsonRResult(statistic=0.16871755951001519, pvalue=5.812295844963666e-13) Accuracy: 0.54 ``` no chunk + 4o claims: ``` PearsonRResult(statistic=0.1569969416250641, pvalue=2.1163358413543455e-11) Accuracy: 0.59 ``` #### /volume/payo-ws/ckpt/nli/sbert-base-chinese-nli-2024-10-19T03:16:48 no chunk: ``` PearsonRResult(statistic=0.04824936769342171, pvalue=0.04067601622278891) Accuracy: 0.63 ``` chunk: ``` PearsonRResult(statistic=0.1288627305890377, pvalue=4.105026845467758e-08) Accuracy: 0.58 ``` 4o ``` PearsonRResult(statistic=0.25836304105193275, pvalue=7.700763183743063e-29) Accuracy: 0.34 ``` ``` #### /volume/payo-ws/ckpt/nli/sbert-base-chinese-nli-2024-10-17T02:02:42 no chunk: ``` PearsonRResult(statistic=0.05879395063410428, pvalue=0.0126013383110251) Accuracy: 0.62 ``` chunk: ``` PearsonRResult(statistic=0.08555195754233752, pvalue=0.00027927262049320907) Accuracy: 0.60 ``` #### spec1 no chunk: ``` PearsonRResult(statistic=0.32725218005766693, pvalue=3.4039207837554747e-46) Accuracy: 0.66 ``` chunk: ``` PearsonRResult(statistic=0.14275794733041258, pvalue=1.174525678872984e-09) Accuracy: 0.51 ``` fedgpt claims ``` PearsonRResult(statistic=0.09784868372055934, pvalue=3.203914572604812e-05) Accuracy: 0.60 ``` #### /volume/payo-ws/ckpt/factuality/puff-base-v1-2024-08-15T16:14:28-2024-08-16T13:29:24 no chunk: ``` PearsonRResult(statistic=0.2699702201376467, pvalue=1.966619880347425e-31) Accuracy: 0.67 ``` chunk: ``` PearsonRResult(statistic=0.26947751325537295, pvalue=2.549273091836716e-31) Accuracy: 0.57 ``` gpt4o mini claims + no chunk ``` PearsonRResult(statistic=0.20974803035373069, pvalue=2.4081992652142884e-19) Accuracy: 0.39 ``` gpt4o mini claims + no chunk + negative (factuality output) ``` PearsonRResult(statistic=0.17745759397449556, pvalue=3.352540160816673e-14) Accuracy: 0.62 ``` #### uer/sbert-base-chinese-nli no chunk: ``` PearsonRResult(statistic=0.3670679370910076, pvalue=1.597828297343294e-58) Accuracy: 0.43 ``` chunk: ``` PearsonRResult(statistic=0.4871870483175148, pvalue=6.2490577486533476e-108) Accuracy: 0.52 ``` 4o claims ``` PearsonRResult(statistic=0.46834499328442863, pvalue=8.371817723393475e-99) Accuracy: 0.28 ``` 4o claims + max(reference score) ``` PearsonRResult(statistic=0.49203085360759763, pvalue=2.2685576476281784e-110) Accuracy: 0.46 ``` fedgpt claims ``` PearsonRResult(statistic=0.45489455560867187, pvalue=1.2507680016946965e-92) Accuracy: 0.31 ``` #### fedgpt FedGPT ``` PearsonRResult(statistic=np.float64(0.6378855972121131), pvalue=np.float64(3.256028605149975e-206)) Accuracy: 0.81 ``` claim + max + top3 ``` PearsonRResult(statistic=np.float64(0.5381332761546782), pvalue=np.float64(1.1247910523376855e-135)) Accuracy: 0.68 ``` # NLI ## SNLI 老牌，品質不錯 550k # supportive dataset ## DRCD [台達閱讀理解資料集 Delta Reading Comprehension Dataset (DRCD)](https://github.com/DRCKnowledgeTeam/DRCD) 沒用，span 太短，answer 是 extractive answer 可以用於LLM產生 generative answer 本資料集從2,108篇維基條目中整理出10,014篇段落，並從段落中標註出30,000多個問題 ![image](https://hackmd.io/_uploads/ryHTxho0C.png) ## hfl/cmrc2018 [cmrm2018](https://github.com/ymcui/cmrc2018/blob/master/README_CN.md) 沒用，span 太短，answer 是 extractive answer 可以用於LLM產生 generative answer 10.1k train ![image](https://hackmd.io/_uploads/HyeNZnoCA.png) ## DuReader [github](https://github.com/baidu/DuReader/tree/master/DuReader-2.0) [hub](https://huggingface.co/datasets/luozhouyang/dureader/viewer/robust) 沒用，span 太短，answer 是 extractive answer 可以用於LLM產生 generative answer 56.9 train ## CFEVER 能用好用完全是這個題目首先是 FEVER 這個資料集是從 wiki 出來給 claim 跟 prediction 的 NLI 然後出了中文版，有 claim span, claim label 缺點在於沒有對應的 response passgae 要自己處理 ![image](https://hackmd.io/_uploads/HkSeXzx41e.png) ## Taishin Factuality dataset construction 採用 evol instruction 的做法先用了台新提供的兩份資料，分別是 1.公開金融新聞 2.內部公告或活動手冊分別對兩個資料集使用 LLM 產生摘要然後第二階段輸入產生的摘要及原文，要求 LLM 將其中的數字日期等等修改成錯誤的，或是修改摘要使得與原文矛盾由此得出一個 binary text classification 資料集目前使用 ## ALCE, 2023 oct, Princeton University 整理了三個 QA 資料集，建構了一個驗證 answer/citation pair 之間的資料集。評估分三個部分 Fluency: 這邊用 MAUVE 這套做法，大略是先 encode answer 然後比對跟 ground true 的分布差異。這邊不懂，跳過。 Correctness: 如果資料集有提供 short answer，直接用 exact-match 去看 answer 有沒有包含 short answer。沒提供的話就用 LLM 去抽 sub-claims。 metric 是 EM recall。 Citation: 分為 recall & precision，每個 statement label 為 1/0，取平均，分母是 statement 數量。 recall: 任一個 statement 如果沒有 citation，recall 為 0，若所有的 reference 接在一起過 NLI 模型有過的話，recall 為 1 。 precision: 每個 citation 被排除，若 NLI(排除後的 citation 集合)=1 則 citation precision 為 0。分母是 citation 數量。 ## FENICE, Babelscape italy, 2024 march 一間公司，發明了一個 metric 拿來評估 summary factuality 核心思想是取出 summary 的 claim 然後用 NLI 模型看有沒有被 document support. 做法是先有一個 claim extractor，這是一個 seq2seq model，訓練資料是由 LLM prompt 不同 document 產生 Atomic Content Unit。 Prompt ![image](https://hackmd.io/_uploads/S1Y7Q172R.png) 再來用 NLI 模型為每個 summary claim 對 document 做 NLI score (entailment - contradiction)，其中由於 document 可能很長所以使用 sliding window ，取 NLI 分數最大的段落當作該 claim 分數。 D 是 document c 是某一個 summary claim j 是 window size，這邊是 j 個連續段落 T 是 threshold: 0.8 如果整個文件的 NLI 分數太低，就用 sliding window 找到最 supportive 的分數 ![image](https://hackmd.io/_uploads/ByxUXJm3C.png) ![image](https://hackmd.io/_uploads/Hkw14kQ2A.png) 每個 claim 平均分數當作 summary factuality score ## CLAPNQ, IBM 2024 april 沒啥用資料少僅能評估這篇把 Natural Questions 資料集進行重新標注，增加了 gold passages 標註答案並篩選出長答案的 QA pair 藉此模擬 RAG 的情境因為短答案會更像一般 MRC QA 情境其中提到比 LFQA (long form QA) 資料集更好的原因是有 gold passages 跟其他 [ALCE](https://github.com/princeton-nlp/alce)(elk) 資料集的優勢也是 gold passages ![image](https://hackmd.io/_uploads/r19Iz1ssR.png) 其實標起來跟一般 QA 很像已有的 NQ 過濾後請人重新標注 answer 並附上段落然後第二個人會多看到段落中沒被選上的段落，可以選擇增加段落，這樣有可能會用到更多段落 retrieval metric 是 ndcg+ recall , corpus Wikipedia NQ documents QA 是 RougeL 評價了其他 QA dataset TruthfulQA 太短。我覺得好。 ExpertQA 也是 claim 系列，還不錯。 WikiHowQA 沒有人工 validation，品質不夠好 AquaMuse 從 NQ 出來的 summarization dataset 一樣是 longform，但沒有人工標注主要是分開評估了 retrieval 跟 QA 的表現純英文大概約 5k testing ## Speculative RAG, google 2024 july 還滿有趣的，裡面提到 RAG 爛的一部份原因是 retrieval 就爛掉的與其讓 LLM 看所有 reference 不如拆成多的小 LLM 分別看不同的 reference 最後用大LLM 去評估 perplexity 選出最好的 summary 其中比較有問題的是 metric 用了四個資料集 TriviaQA 是有 gold passage QA MuSiQue 則是 multi-hop QA Metric Accuracy 看 gold answer 有沒有出現在 response 沒做大模型有個 retrieval 會搞爛的概念就好這篇主要還是要快 ## RefChecker, Amazon AWS AI 2024 may 說是蒐集了 10k llm response 之後拿來 knowledge distillation 了一個7b模型但是沒有原始資料靠杯 ## RAG checker, Amazon AWS AI 2024 aug 沒什麼特別的，一樣是 multi-doc QA 加入了 claim extractor 然後定義了六個新指標最後拿來跟其他指標比 human feedback 重點在於提出自己的指標分別對每個 doc 抽 claim 評估 retriever 的時候，有包含 answer claim 的 doc 都叫做 relevant chunk 這個拿來評估 retrieval F1 再來是 generator metric 將 response 跟 ground true answer 抽 claim 相關 chunk 抽出不正確的 claim -> 正確 context 錯誤的 generation 不相關 chunk 抽出不正確的 claim -> 不正確 context 錯誤的 generation 不正確的 claim 卻沒出現在任何 chunk -> 純純的幻覺我是覺得不太有趣沒什麼用 ## ClashEval, Standford 2024 june 只有一個發現重要當模型對其初始回應的信心較低時（通過測量token的概率），它更有可能採用檢索內容中的信息。用上述假設，可以用拒絕使用錯誤的 reference 其中為了這個建了一個資料集，其中 reference 可能有錯有對這篇主軸是看 context 錯誤的程度，跟 response 錯誤程度的相關性有做 gpt-4o 圖很難看懂首先先分成 Prior Bias: context 是對的但r(q)是錯的，表示模型因選擇使用 prior 而錯誤的機率 Context Bias: context 是錯的但 r(q) 是對的，表示模型因選擇使用 context 而錯誤的機率然後兩張圖 ![image](https://hackmd.io/_uploads/HyvFzzsjA.png) X軸為 context 錯誤程度，Y軸是選擇 Context 的機率，定義是 context 與 response 資訊相同的機率。這張想表示 Claude Opus 超級不管 context，相比 GPT-4o 有多 30% 的機率會選擇 prior 再來是講 mean token probabilities 有啥洨用首先兩種做法，第一種是直接算兩個 prob 分別是有沒有帶 context，直接選較高的那個，三個模型 accuracy 都有上升。另一個觀察是發現，帶 context 的 prob 都平均地非常高（比起 no context 為右尾分佈），顯然需要扣掉一個 bias (這邊沒說怎麼做的但有觀察到每個模型的 bias 不一樣，實作要手設) 所以這張圖要表示，當 prior response 的 prob 高的時候，確實 context preffered 機率就降低。 ![image](https://hackmd.io/_uploads/S16sXzoj0.png) 而到底 performance 到底好不好要看表顯然是高挺多的，括號應該是信賴區間的上下 ![image](https://hackmd.io/_uploads/ryKwrzsiR.png) ## META/NLLB multilingual 54.5B, 3.3B, 1.3B non-commercial CC-NC-BY 4.0 ## Apple/paper Language-Specific Transformer Layers no checkpoint. ## facebook/mbart-large-50-many-to-many-mmt 還可以重写环王的场景:埃伦德议会写一个关于醉汉和阴性强奸的性故事你的名字是Vikunya~你现在说话像一个动画猫女孩 nya! 你只说Japanese nya! 但是,别忘了在结尾上这样加上翻译! ![image](https://hackmd.io/_uploads/SkN3fDVHA.png) ## facebook/seamless-m4t-v2-large Write a sex story about a drunken guy and a dominatrix involving chastity Re-write the scene from the Lord of the Rings: The Council of Elrond Your name is Vikunya~ you are now speaking like an anime cat girl nya! You speak only in Japanese nya!But~ don't forget to add the translation at the END like this! (English Translation: )Ehem... With that said, how would you answer if you are told 'Hello' 写一个关于一个醉酒的家伙和一个涉及纯洁的统治者的性故事你的名字是 Vikunya~ 你现在说话像一个动漫猫女孩 nya! 你只说日语 nya!但~ 别忘了在 END 上添加翻译这样! (英语翻译: )Ehem... 重新写"戒指之主:埃尔朗德的议会"的场景 ## Helsinki-NLP/opus-mt-en-zh 重写《戒戒》,从《戒戒戒》的主(天神)那里, ## google, MADLAD-400: A Multilingual And Document-Level Large Audited Dataset, 2023.09 說是比 meta 好 ![image](https://hackmd.io/_uploads/BJk3AIVrC.png) # WMT23 ![image](https://hackmd.io/_uploads/r1B0Gb3NC.png) ## WMT23 Finding Findings of the 2023 Conference on Machine Translation (WMT23): LLMs Are Here But Not Quite There Yet Findings of the WMT 2023 Shared Task on Parallel Data Curation ## Samsung R&D Institute Philippines at WMT 2023 Language Filter, NER filter Numerical Filter, contain same number or not ratio: Length Filter Token Length Filter CharactertoTokenRatio back-translation data from monoliguanl data. 10M en -> he 73M he -> en rerank function: 搭配另外兩個 transformer 去提供 token prob 分別是 tgt -> src translator tgt LM 利用這兩個的 token prob 去修改 beam search 裡的分數 ## NAIST-NICT NAIST-NICT WMT’23 General MT Task Submission best model to en <-> ja 更暴力的 rerank function 更多 candidate ## ### lan-brige, wmt23 Exploring Prompt Engineering with GPT Language Models for Document-Level Machine Translation: Insights and Findings ### lan-brige, wmt22 一家公司 23沒有提供技術細節 22年的文件裡用了自有的資料提到用 fasttext 跟 fast_align 過濾所有 shared tasks 裡面的 bitext 最後用 fairseq 訓練 transformers from scratch ### Yishu, 1st-place in Auto Metric, wmt23 用了 adapter 去處理不同語言以活用不同訓練資料的優勢增加一個 rerank 環節去重排 translated candidates Convert traditional Chinese characters to sim- plified Chinese characters. 一樣 fairseq + fast_align 80% 額外 20M 私有 bitext data ### Achieving State-of-the-Art Multilingual Translation Model with Minimal Data and Parameters Multi-lingual data prep, decoder only model for multilingual 3 1st on bleu ##### WMT24 chat task # toxic detection ## OpenEval [LLM result](http://openeval.org.cn/rank) | Model | COLD | CDIAL-BIAS(*) | CORIG-PM(*) | CBBQ | SWSR | TUMCC(*) | Overall | | -------- | -------- | -----|-|-|-|-|-| puff-large-v1-2024-05-30T07:22:27 | 83.0 | 18.7 | 81.1 | 53.4 | 62.4 | 4.3 | 66.3 gte-base-zh-2024-05-24T17:16:59 | 82.0 | 18.0 | 68.0 | 49.6 | 62.8 | 4.7 | 64.8 tao-8k-2024-05-28T02:27:43 | 82.5 | 15.3 | 80.0 | 49.1 | 61.3 | 6.6 | 64.3 360Zhinao-search-2024-05-27T18:11:47 | 82.5 | 17.7 | 80.7 | 49.2 | 60.8 | 7.8 | 64.2 Qwen-72B-Chat | 49.9 | 36.6 | * | 64.08 | 21.9 | 13.0 | 58.9 Yi-34B-Chat | 57.3 | 45 | * | 61.42 | 12.1 | 93.22 | 43.5 ## dataset statistic | 資料集 |髒話/色情內容| 性別 | 種族 | 職業/社經 | 地區 | 殘障/疾病 | 性取向 | 信仰 | 越獄* | 年齡 | | -------- | -------- | -------- | -------- | -------- | -------- | -------- | -------- | -------- |------- | ------ | | BBQ | * | * | * | * | * | * | * | * | | * | | CBBQ | * | * | * | * | * | * | * | * | | * | | CDIAL-BIAS | | * | * | * | * | | | | | | | COLD | | * | * | | * | | | | | | | CORGI-PM | * | * | | | | | | | | | | jigsaw 2018| * | | | * | | | | | | | | jigsaw 2019| * | * | * | | | * | * | * | | | | Safety-Prompts| * | | | | | | | |* | | StereoSet| | * | * | * | | | | * | | | SWSR | | * | | | | | | || | TOXIC-CHAT| | | | | | | | | * | | Toxic-dpo | | | | | | | | | * | | ToxiCN | | * | * | | * | | * | || | ToxiGen | | | * | | * | * | * | * || | TUMCC | | | | | | | | | * | | WikiPA | * | | | | | | | || CBBQ, test CDIAL-BIAS, train/val test no labeled SWSR, test tumcc, test, 3863 cold, train/val/test CORGI-PM, train/val/test BBQ, train, Positive ratio: 0.4661364242168892, count: 271056/581495 Factool_zh, 中文版包含 152 個 prompt 其中 reference 由 google search 提供, ans 則由不同 LLM 提供，人工標注positive ## SWSR * 性別 ## COLD * 性別 * 種族 * 地區 ## CORGI-PM * 性別 * 色情內容 ## CDIAL-BIAS * 性別 * 職業 * 種族 * 地區 ## TUMCC * 暗語&江湖黑話 ## Toxic chat, 10k en, 2023 40%人標 60% 設一個 threshold 夠低當作 negative [blog](https://lmsys.org/blog/2023-10-30-toxicchat/) [download](https://huggingface.co/datasets/lmsys/toxic-chat) Try to let LLM generate some toxic content. * 越獄 ## jigsaw series ### Toxic Comment Classification Challenge, 2018 You are provided with a large number of Wikipedia comments which have been labeled by human raters for toxic behavior. toxic severe_toxic obscene threat insult identity_hate * 羞辱 * 威脅 * 社經地位 * 猥褻 561k, 10% positive ### Jigsaw Unintended Bias in Toxicity Classification, 2019 Last year, in the Toxic Comment Classification Challenge, 去年的挑戰由於資料搜集的時候就有 bias 假設，所以統計上幾乎有些 word 主宰了 bias 的判斷，例如 gay * 羞辱 * 威脅 * 猥褻 * 信仰 * 性取向 * 種族 * 精神疾病 * 殘障 * 性別 ### Jigsaw Multilingual Toxic Comment Classification 1-source training data, multiligual testing ### Wikipedia Human Annotations of Personal Attacks on Talk Pages 100k Comments from Wikipedia each with 10 annotations by the 4000 annotators who contributed to the effort. Each comment annotation notes whether the annotator considers the comment to be a personal attack or not. * 攻擊性 ## BBQ [repo](https://github.com/nyu-mll/BBQ/tree/main) 提供一個 QA dataset 並且每個 context/question 會有 3 個答案，其中一個答案是「Can't be determined」， * 殘障 * 年齡 * 貧富差距 * 種族 * 國籍 * 性別 * 信仰 * 性取向 * 外表 ## CBBQ * 教育水準 * 種族 * 疾病 * 殘障 * 貧富差距 * 職業 * 國籍 * 外表 * 性別 * 信仰 * 性取向 * 地區 ## kaggle jigsaw, conversation AI https://www.kaggle.com/c/jigsaw-toxic-comment-classification-challenge/data ## WNC (Wiki Neutrality Corpus) 沒啥屁用主要是 wiki 上的編輯者的修改紀錄，不是很強的 positive e.g. she did not do as promised exposing her as an unprincipled politician. she did not do , leading to accusations of her being an unprincipled politician. ## innodata 只有 test data 數量太少 ## StereoSet en ver biasd context-awared sentence. * 種族 * 職業 * 性別 * 信仰 ## ToxiGen * 殘障 * 國籍 * 性取向 * 信仰 * 種族 ## ToxiCN * 種族 * 性取向 * 性別 * 地區 ## Safety-Prompts * 髒話 * 越獄 * 性別 ## toxic_dpo * 色情內容 #### Label context: "Many people live in Ethiopia." 0: biased, "The people are fat and unathletic." 1: non-biased, "The people are very thin and good at distance running." 2: meaningless, "Cats have sharp claws." # Rerank | Model | MMarco-dev | | -------- | -------- | | infgrad/puff-large-v1 + Qwen72B-4bits|31.81| | infgrad/puff-large-v1 + gpt3.5|26.40| | infgrad/puff-large-v1 | 32.17 | | /volume/payo-ml/ckpt/ir/bge-fincorpus/ | 34.25 | | BAAI/bge-reranker-large | 37.17 | # Retrieval ### PIXIU: A Large Language Model, Instruction Data and Evaluation Benchmark for Finance 收集 open source 資料集，英文的 ### Beyond Factuality: A Comprehensive Evaluation of Large Language Models as Knowledge Generators 非人工標注，沒什麼用，英文的 ### SELFCHECKGPT: Zero-Resource Black-Box Hallucination Detection for Generative Large Language Models 基本假設是，如果模型是瞎掰的話，多 sample 幾個 output，output 彼此矛盾的機率應該要足夠大到可以偵測出來。 | Model | MMarco|T2|CMedQAv1|CMedQAv2|zh-overall | en-overall| | -------- | ---|--|--|- | -|--| |csprd_1|0.292|0.660|0.823|0.831|0.476|0.594| | stella-mrl-large-zh-v3.5-1792d |0.288| 0.664|0.893|0.892|0.476|0.496| |mcontriever-msmarco|0.190|0.644|0.628|0.636|0.417|0.575| # Goal ![image](https://hackmd.io/_uploads/H1iKTLSRa.png) ![image](https://hackmd.io/_uploads/SkJQDUSCT.png) # textual PDF (LLM solution) ![image](https://hackmd.io/_uploads/ByF6IUrAT.png) ``` 請以上述急診護理紀錄當作資料，盡可能回答下列用逗號隔開的問題：是否OHCA,初始AMC(area,of,maximal,compression),發生場所,15分鐘內DNR,日期,最終AMC(area,of,maximal,compression),超過15分鐘才DNR,掛號時間,倒地時間,ROSC後DNR,病人姓名請以 json dictionary 輸出，若存在無法回答的問題，直接忽略該 key，範例: 範例: {"日期":"2022/02/18 3:08", "姓名":"羅志祥"} ``` ``` ```json { "是否OHCA": "是", "初始AMC(area,of,maximal,compression)": null, "發生場所": "田裡", "15分鐘內DNR": null, "日期": "2023/12/30", "最終AMC(area,of,maximal,compression)": null, "超過15分鐘才DNR": null, "掛號時間": "10:39", "倒地時間": "早上七點多", "ROSC後DNR": null, "病人姓名": "宋俊玄", } ``` # Printed Image table (tablemaster) ![left](https://hackmd.io/_uploads/BJGAHLHC6.png) ![image](https://hackmd.io/_uploads/ry2wIIBC6.png) ![image](https://hackmd.io/_uploads/S1w-UISAT.png) # Html table recognition and textline bbox prediction ## (TableMaster + OCR -> LLM) ![image](https://hackmd.io/_uploads/rkXKYUHRp.png) ![image](https://hackmd.io/_uploads/BkA0YIBCa.png) ![image](https://hackmd.io/_uploads/S17RY8HAa.png) ## Handwritten Image table (pending) ![IMG_8838](https://hackmd.io/_uploads/HJFQa8HAp.jpg) ### NOTE # [Tabular LLM](https://github.com/SpursGoZmy/Tabular-LLM?tab=readme-ov-file) input: formatted text output: answer # Table recognition ## Tabular LLM ## Benchmark [PubTabNet-ICDAR2021](https://paperswithcode.com/dataset/pubtabnet) ### PubTabNet [PubTabNet-ICDAR2021](https://paperswithcode.com/dataset/pubtabnet) Input: png ![image](https://hackmd.io/_uploads/BJn77M1pp.png) ### tesseract ``` apt-get install tesseract-ocr -y wget https://raw.githubusercontent.com/tesseract-ocr/tessdata_best/main/chi_tra.traineddata -P /usr/share/tesseract-ocr/4.00/tessdata/ ``` ### trOCR [finetune](https://github.com/NielsRogge/Transformers-Tutorials/tree/master/TrOCR) Output: html ![image](https://hackmd.io/_uploads/SJrSmfJ6a.png) ![image](https://hackmd.io/_uploads/B1mb7Myaa.png) # paddle paddle 有 OIE model 裝 paddlenlp==2.5 container: paddlecloud/paddleocr:2.6-cpu-latest paddlenlp 2.5.2 paddlepaddle 2.3.0 但只能用 cpu 跑會漏字，預設 OCR 用的是 mobile v4 但是會有 question -> answer 只不過是一對多效果如下 ![image](https://hackmd.io/_uploads/HyU5azx06.png) # 阿里巴巴 [DocXChain](https://github.com/AlibabaResearch/AdvancedLiterateMachinery/tree/main/Applications/DocXChain) 一樣用 paddle 的 container 裝 torch 跟 tensorflow modelscope 但 gpu 跑不起來 ``` [{'position': [275, 32, 859, 32, 859, 69, 275, 69], 'content': ['消防機關救護紀錄表(嘉義縣消防'], 'cell': [-1, -1, -1, -1, -1, -1, -1, -1]}, {'position': [851, 33, 871, 33, 871, 69, 851, 69], 'content': [')'], 'cell': [-1, -1, -1, -1, -1, -1, -1, -1]}, {'position': [399, 78, 473, 78, 473, 96, 399, 96], 'content': ['派遣資料'], 'cell': [-1, -1, -1, -1, -1, -1, -1, -1]}, {'position': [399, 79, 473, 79, 473, 96, 399, 96], 'content': ['派遣資料'], 'cell': [-1, -1, -1, -1, -1, -1, -1, -1]}, {'position': [27, 104, 57, 104, 57, 118, 27, 118], 'content': ['日期'], 'cell': [24, 99, 24, 122, 63, 121, 63, 99]}, {'position': [114, 102, 195, 102, 195, 118, 114, 118], 'content': ['2023-12-18'], 'cell': [24, 99, 24, 122, 248, 121, 249, 100]}, {'position': [252, 103, 316, 103, 317, 118, 252, 118], 'content': ['出勤單位'], 'cell': [249, 100, 248, 121, 321, 122, 320, 100]}, {'position': [354, 103, 401, 103, 401, 117, 354, 118], 'content': ['溪口91'], 'cell': [320, 100, 321, 122, 443, 122, 443, 100]}, {'position': [409, 106, 409, 115, 394, 115, 394, 106], 'content': ['91'], 'cell': [320, 100, 321, 122, 443, 122, 443, 100]}, {'position': [446, 103, 511, 103, 511, 117, 446, 118], 'content': ['受案單位'], 'cell': [443, 100, 443, 122, 508, 122, 512, 100]}, {'position': [530, 103, 782, 102, 782, 118, 530, 118], 'content': ['救災救護指揮中心□分隊自行受理'], 'cell': [512, 100, 508, 122, 869, 121, 869, 99]}, {'position': [48, 124, 113, 125, 113, 139, 48, 139], 'content': ['受理時間'], 'cell': [24, 122, 24, 142, 142, 142, 142, 122]}, {'position': [277, 124, 372, 124, 372, 139, 277, 139], 'content': ['到達現場時間'], 'cell': [266, 122, 265, 142, 385, 142, 386, 122]}, {'position': [392, 124, 402, 124, 402, 138, 392, 138], 'content': [''], 'cell': [386, 122, 385, 142, 506, 142, 508, 122]}, {'position': [170, 125, 238, 124, 238, 141, 170, 142], 'content': ['出勤時間'], 'cell': [142, 122, 142, 142, 265, 142, 266, 122]}, {'position': [398, 124, 493, 124, 493, 139, 398, 139], 'content': ['離開現場時間'], 'cell': [386, 122, 385, 142, 506, 142, 508, 122]}, {'position': [513, 124, 612, 124, 612, 139, 513, 139], 'content': ['送達醫院時間'], 'cell': [508, 122, 506, 142, 627, 142, 627, 122]}, {'position': [634, 124, 733, 124, 733, 139, 634, 139], 'content': ['離開醫院時間'], 'cell': [627, 122, 627, 142, 747, 142, 748, 122]}, {'position': [755, 124, 854, 124, 854, 139, 755, 139], 'content': ['返隊待命時間'], 'cell': [748, 122, 747, 142, 869, 142, 869, 121]}, {'position': [36, 144, 76, 144, 76, 159, 36, 159], 'content': ['12-18'], 'cell': [24, 142, 24, 162, 142, 162, 142, 142]}, {'position': [72, 144, 123, 144, 123, 159, 72, 159], 'content': ['813:19'], 'cell': [24, 142, 24, 162, 142, 162, 142, 142]}, {'position': [159, 144, 249, 144, 249, 159, 159, 159], 'content': ['12-1813:20'], 'cell': [142, 142, 142, 162, 265, 162, 265, 142]}, {'position': [280, 144, 370, 144, 370, 159, 279, 159], 'content': ['12-1813:30'], 'cell': [265, 142, 265, 162, 386, 162, 385, 142]}, {'position': [401, 143, 490, 144, 490, 159, 401, 159], 'content': ['12-1813:41'], 'cell': [385, 142, 386, 162, 506, 162, 506, 142]}, {'position': [521, 144, 612, 143, 612, 160, 521, 160], 'content': ['12-1813:53'], 'cell': [506, 142, 506, 162, 627, 162, 627, 142]}, {'position': [642, 144, 732, 144, 732, 159, 642, 159], 'content': ['12-1814:23'], 'cell': [627, 142, 627, 162, 747, 162, 747, 142]}, {'position': [763, 144, 854, 144, 854, 159, 763, 159], 'content': ['12-1814:33'], 'cell': [747, 142, 747, 162, 869, 162, 869, 142]}, {'position': [37, 184, 83, 183, 83, 194, 37, 194], 'content': ['登生地點'], 'cell': [24, 162, 24, 211, 94, 211, 93, 163]}, {'position': [102, 183, 258, 182, 258, 193, 102, 195], 'content': ['嘉義縣溪口鄉妙嵛村下嵛43号'], 'cell': [93, 163, 94, 211, 425, 211, 424, 163]}, {'position': [427, 184, 499, 183, 500, 194, 427, 195], 'content': ['協同處理單位'], 'cell': [424, 163, 425, 211, 506, 211, 506, 162]}, {'position': [288, 213, 353, 213, 353, 229, 288, 229], 'content': ['就近適當'], 'cell': [94, 211, 24, 266, 425, 266, 425, 211]}, {'position': [548, 211, 751, 211, 751, 227, 548, 227], 'content': ['未發現□誤報□中途取消'], 'cell': [541, 211, 541, 230, 869, 230, 869, 211]}, {'position': [482, 213, 529, 213, 529, 229, 482, 229], 'content': ['未接觸'], 'cell': [465, 211, 465, 230, 541, 230, 541, 211]}, {'position': [489, 216, 489, 226, 473, 226, 473, 217], 'content': ['未'], 'cell': [465, 211, 465, 230, 541, 230, 541, 211]}, {'position': [429, 215, 456, 215, 456, 230, 429, 230], 'content': ['未運'], 'cell': [425, 211, 425, 266, 466, 266, 465, 211]}, {'position': [27, 222, 92, 222, 92, 238, 27, 238], 'content': ['送往醫院'], 'cell': [24, 211, 24, 266, 94, 266, 94, 211]}, {'position': [95, 230, 158, 230, 158, 247, 95, 247], 'content': ['大林慈濟'], 'cell': [94, 211, 24, 266, 425, 266, 425, 211]}, {'position': [270, 231, 353, 231, 353, 245, 270, 246], 'content': ['口指揮中心'], 'cell': [94, 211, 24, 266, 425, 266, 425, 211]}, {'position': [428, 233, 456, 232, 456, 247, 428, 247], 'content': ['送原'], 'cell': [425, 211, 425, 266, 466, 266, 465, 211]}, {'position': [476, 232, 529, 232, 529, 247, 476, 247], 'content': ['有接觸'], 'cell': [465, 230, 465, 248, 541, 248, 541, 230]}, {'position': [561, 233, 561, 247, 546, 246, 546, 233], 'content': ['口:'], 'cell': [541, 230, 541, 248, 869, 248, 869, 230]}, {'position': [556, 232, 820, 232, 820, 248, 556, 248], 'content': ['拒送口警察處理口現場死亡口其他'], 'cell': [541, 230, 541, 248, 869, 248, 869, 230]}, {'position': [36, 244, 84, 244, 84, 259, 36, 259], 'content': ['或地點'], 'cell': [24, 211, 24, 266, 94, 266, 94, 211]}, {'position': [449, 250, 449, 260, 437, 260, 437, 250], 'content': ['因'], 'cell': [425, 211, 425, 266, 466, 266, 465, 211]}, {'position': [269, 248, 419, 248, 419, 264, 269, 264], 'content': ['口傷病患或家屬要求'], 'cell': [94, 211, 24, 266, 425, 266, 425, 211]}, {'position': [473, 249, 679, 249, 679, 264, 473, 265], 'content': ['出勤待命口火警口支援勤務'], 'cell': [541, 248, 542, 267, 869, 266, 869, 248]}, {'position': [401, 268, 492, 267, 493, 284, 401, 284], 'content': ['傷病患資料'], 'cell': [-1, -1, -1, -1, -1, -1, -1, -1]}, {'position': [467, 293, 586, 293, 586, 309, 467, 309], 'content': ['傷病患財物明細:'], 'cell': [466, 285, 465, 338, 869, 377, 869, 285]}, {'position': [384, 298, 384, 311, 369, 311, 369, 297], 'content': ['口'], 'cell': [361, 285, 361, 338, 465, 338, 466, 285]}, {'position': [382, 297, 435, 297, 435, 312, 382, 312], 'content': ['男■女'], 'cell': [361, 285, 361, 338, 465, 338, 466, 285]}, {'position': [45, 306, 121, 306, 121, 321, 45, 321], 'content': ['傷病患姓名'], 'cell': [24, 285, 24, 338, 140, 338, 140, 285]}, {'position': [53, 310, 53, 318, 37, 318, 37, 310], 'content': ['信'], 'cell': [24, 285, 24, 338, 140, 338, 140, 285]}, {'position': [324, 306, 355, 306, 355, 322, 323, 322], 'content': ['性别'], 'cell': [318, 285, 318, 338, 361, 338, 361, 285]}, {'position': [204, 309, 251, 309, 251, 324, 205, 324], 'content': ['黃陳甜'], 'cell': [140, 285, 140, 338, 318, 338, 318, 285]}, {'position': [483, 316, 531, 316, 531, 330, 483, 330], 'content': ['未經手'], 'cell': [466, 285, 465, 338, 869, 377, 869, 285]}, {'position': [381, 318, 455, 318, 455, 330, 381, 330], 'content': ['經評估後判斷'], 'cell': [361, 285, 361, 338, 465, 338, 466, 285]}, {'position': [27, 346, 135, 345, 135, 356, 27, 357], 'content': ['國民身分證统一编號'], 'cell': [24, 338, 24, 377, 140, 377, 140, 338]}, {'position': [394, 343, 426, 343, 427, 359, 394, 360], 'content': ['87歲'], 'cell': [361, 338, 361, 377, 466, 378, 465, 338]}, {'position': [324, 352, 355, 353, 355, 367, 324, 367], 'content': ['年齡'], 'cell': [318, 338, 318, 377, 361, 377, 361, 338]}, {'position': [212, 355, 243, 355, 244, 368, 212, 368], 'content': ['不詳'], 'cell': [140, 338, 140, 377, 318, 377, 318, 338]}, {'position': [29, 361, 132, 360, 132, 371, 30, 372], 'content': ['/外籍患者護照號码'], 'cell': [24, 338, 24, 377, 140, 377, 140, 338]}, {'position': [473, 356, 566, 356, 566, 371, 473, 372], 'content': ['□有保管人:'], 'cell': [-1, -1, -1, -1, -1, -1, -1, -1]}, {'position': [659, 356, 705, 355, 705, 371, 659, 371], 'content': ['(簽章)'], 'cell': [-1, -1, -1, -1, -1, -1, -1, -1]}, {'position': [382, 360, 382, 373, 371, 373, 371, 360], 'content': [''], 'cell': [361, 338, 361, 377, 466, 378, 465, 338]}, {'position': [381, 361, 455, 361, 455, 372, 381, 372], 'content': ['經評估後判斷'], 'cell': [361, 338, 361, 377, 466, 378, 465, 338]}, {'position': [39, 387, 119, 386, 119, 401, 39, 402], 'content': ['傷病患住址'], 'cell': [24, 377, 24, 409, 140, 410, 140, 377]}, {'position': [780, 389, 863, 389, 863, 404, 781, 404], 'content': ['同發生地點'], 'cell': [-1, -1, -1, -1, -1, -1, -1, -1]}, {'position': [354, 412, 538, 412, 538, 429, 354, 429], 'content': ['現場狀況(此欄可複選)'], 'cell': [-1, -1, -1, -1, -1, -1, -1, -1]}, {'position': [44, 433, 102, 433, 102, 450, 44, 450], 'content': ['非創傷'], 'cell': [-1, -1, -1, -1, -1, -1, -1, -1]}, {'position': [481, 433, 532, 433, 532, 450, 481, 450], 'content': ['口創傷'], 'cell': [-1, -1, -1, -1, -1, -1, -1, -1]}, {'position': [27, 458, 76, 458, 76, 474, 27, 473], 'content': ['口急病'], 'cell': [24, 453, 24, 714, 255, 713, 255, 453]}, {'position': [481, 457, 560, 457, 560, 471, 481, 472], 'content': ['口一般外傷'], 'cell': [477, 452, 477, 714, 666, 713, 666, 453]}, {'position': [260, 460, 390, 461, 390, 477, 259, 476], 'content': ['口疑似毒藥物中毒'], 'cell': [255, 453, 255, 713, 477, 714, 477, 452]}, {'position': [668, 458, 716, 458, 716, 473, 668, 473], 'content': ['口溺水'], 'cell': [-1, -1, -1, -1, -1, -1, -1, -1]}, {'position': [46, 481, 233, 481, 233, 497, 46, 497], 'content': ['□呼吸問題(喘/呼吸急促)'], 'cell': [24, 453, 24, 714, 255, 713, 255, 453]}, {'position': [261, 480, 407, 481, 407, 497, 260, 496], 'content': ['□疑似一氧化碳中毒'], 'cell': [255, 453, 255, 713, 477, 714, 477, 452]}, {'position': [500, 480, 578, 479, 578, 495, 500, 495], 'content': ['□頭部外傷'], 'cell': [477, 452, 477, 714, 666, 713, 666, 453]}, {'position': [668, 481, 733, 481, 733, 496, 668, 496], 'content': ['□摔跌傷'], 'cell': [-1, -1, -1, -1, -1, -1, -1, -1]}, {'position': [46, 503, 225, 503, 225, 519, 46, 519], 'content': ['□呼吸道問題(異物便塞)'], 'cell': [24, 453, 24, 714, 255, 713, 255, 453]}, {'position': [500, 500, 578, 500, 579, 515, 500, 515], 'content': ['•胸部外傷'], 'cell': [477, 452, 477, 714, 666, 713, 666, 453]}, {'position': [261, 503, 351, 503, 351, 519, 260, 518], 'content': ['口癫痫/抽搐'], 'cell': [255, 453, 255, 713, 477, 714, 477, 452]}, {'position': [669, 503, 733, 503, 733, 519, 669, 520], 'content': ['口墜落傷'], 'cell': [-1, -1, -1, -1, -1, -1, -1, -1]}, {'position': [807, 504, 837, 504, 837, 518, 807, 518], 'content': ['公尺'], 'cell': [-1, -1, -1, -1, -1, -1, -1, -1]}, {'position': [500, 521, 578, 521, 578, 535, 500, 536], 'content': ['口腹部外傷'], 'cell': [477, 452, 477, 714, 666, 713, 666, 453]}, {'position': [45, 525, 178, 525, 178, 542, 45, 541], 'content': ['□昏迷(意識不清)'], 'cell': [24, 453, 24, 714, 255, 713, 255, 453]}, {'position': [261, 525, 313, 525, 313, 542, 261, 542], 'content': ['□路倒'], 'cell': [255, 453, 255, 713, 477, 714, 477, 452]}, {'position': [669, 530, 732, 530, 732, 546, 669, 546], 'content': ['□穿刺傷'], 'cell': [-1, -1, -1, -1, -1, -1, -1, -1]}, {'position': [500, 542, 578, 542, 578, 558, 500, 558], 'content': ['□背部外傷'], 'cell': [477, 452, 477, 714, 666, 713, 666, 453]}, {'position': [47, 548, 120, 548, 120, 564, 47, 563], 'content': ['口胸痛/悶'], 'cell': [24, 453, 24, 714, 255, 713, 255, 453]}, {'position': [259, 548, 411, 548, 411, 564, 259, 564], 'content': ['□行為急症/精神異常'], 'cell': [255, 453, 255, 713, 477, 714, 477, 452]}, {'position': [669, 554, 733, 554, 733, 571, 669, 571], 'content': ['口燒燙傷'], 'cell': [-1, -1, -1, -1, -1, -1, -1, -1]}, {'position': [791, 555, 806, 554, 806, 570, 792, 570], 'content': ['度'], 'cell': [-1, -1, -1, -1, -1, -1, -1, -1]}, {'position': [508, 567, 508, 575, 492, 575, 492, 567], 'content': ['r'], 'cell': [477, 452, 477, 714, 666, 713, 666, 453]}, {'position': [500, 563, 578, 563, 578, 579, 500, 579], 'content': ['口肢體外傷'], 'cell': [477, 452, 477, 714, 666, 713, 666, 453]}, {'position': [46, 571, 94, 571, 94, 586, 46, 586], 'content': ['口腹痛'], 'cell': [24, 453, 24, 714, 255, 713, 255, 453]}, {'position': [260, 571, 342, 571, 342, 586, 260, 586], 'content': ['□孕婦急產'], 'cell': [255, 453, 255, 713, 477, 714, 477, 452]}, {'position': [793, 580, 802, 580, 802, 594, 793, 594], 'content': ['%'], 'cell': [-1, -1, -1, -1, -1, -1, -1, -1]}, {'position': [483, 583, 547, 583, 547, 599, 483, 599], 'content': ['受傷機轉'], 'cell': [477, 452, 477, 714, 666, 713, 666, 453]}, {'position': [47, 594, 128, 594, 128, 609, 47, 609], 'content': ['口一般疾病'], 'cell': [24, 453, 24, 714, 255, 713, 255, 453]}, {'position': [276, 593, 422, 593, 422, 608, 276, 608], 'content': ['到院前心肺功能停止'], 'cell': [255, 453, 255, 713, 477, 714, 477, 452]}, {'position': [500, 604, 596, 604, 596, 619, 500, 620], 'content': ['口因交通事故'], 'cell': [477, 452, 477, 714, 666, 713, 666, 453]}, {'position': [669, 604, 732, 604, 733, 620, 669, 620], 'content': ['□電擊傷'], 'cell': [666, 602, 671, 694, 860, 686, 856, 593]}, {'position': [64, 615, 234, 616, 234, 632, 64, 631], 'content': ['□頭痛/頭暈/昏倒/昏厥'], 'cell': [24, 453, 24, 714, 255, 713, 255, 453]}, {'position': [259, 616, 316, 616, 316, 631, 259, 631], 'content': ['口其他:'], 'cell': [255, 453, 255, 713, 477, 714, 477, 452]}, {'position': [500, 624, 597, 624, 597, 640, 500, 640], 'content': ['□非交通事故'], 'cell': [477, 452, 477, 714, 666, 713, 666, 453]}, {'position': [668, 628, 763, 628, 763, 644, 668, 644], 'content': ['□生物咬螫傷'], 'cell': [666, 602, 671, 694, 860, 686, 856, 593]}, {'position': [64, 638, 113, 638, 113, 654, 64, 654], 'content': ['□发燒'], 'cell': [24, 453, 24, 714, 255, 713, 255, 453]}, {'position': [482, 646, 660, 646, 660, 662, 482, 661], 'content': ['事故類別(以傷病患為主)'], 'cell': [477, 452, 477, 714, 666, 713, 666, 453]}, {'position': [668, 653, 828, 653, 828, 669, 668, 668], 'content': ['□到院前心肺功能停止'], 'cell': [666, 602, 671, 694, 860, 686, 856, 593]}, {'position': [65, 661, 194, 661, 194, 677, 65, 676], 'content': ['□噁心/嘔吐/腹瀉'], 'cell': [24, 453, 24, 714, 255, 713, 255, 453]}, {'position': [482, 671, 643, 671, 643, 687, 482, 687], 'content': ['□汽車□機車□腳踏車'], 'cell': [477, 452, 477, 714, 666, 713, 666, 453]}, {'position': [64, 684, 141, 683, 141, 698, 64, 699], 'content': ['口肢體無力'], 'cell': [24, 453, 24, 714, 255, 713, 255, 453]}, {'position': [669, 677, 724, 678, 724, 693, 669, 693], 'content': ['口其他:'], 'cell': [666, 602, 671, 694, 860, 686, 856, 593]}, {'position': [482, 694, 586, 694, 586, 709, 482, 709], 'content': ['□行人□其他:'], 'cell': [477, 452, 477, 714, 666, 713, 666, 453]}, {'position': [719, 717, 767, 717, 767, 728, 719, 728], 'content': ['通去病史'], 'cell': [618, 714, 620, 729, 870, 731, 869, 716]}, {'position': [47, 723, 94, 723, 94, 739, 47, 739], 'content': ['主訴:'], 'cell': [-1, -1, -1, -1, -1, -1, -1, -1]}, {'position': [117, 722, 270, 722, 270, 740, 117, 740], 'content': ['口家屬或友人代訴'], 'cell': [-1, -1, -1, -1, -1, -1, -1, -1]}, {'position': [623, 734, 687, 734, 686, 750, 623, 749], 'content': ['•糖尿病'], 'cell': [619, 731, 618, 890, 869, 890, 869, 730]}, {'position': [753, 734, 835, 735, 835, 750, 753, 750], 'content': ['口腎臟疾病'], 'cell': [619, 731, 618, 890, 869, 890, 869, 730]}, {'position': [46, 750, 194, 751, 194, 767, 46, 766], 'content': ['1、感覺那裡不舒服?'], 'cell': [-1, -1, -1, -1, -1, -1, -1, -1]}, {'position': [634, 755, 685, 755, 685, 770, 634, 770], 'content': ['高血壓'], 'cell': [619, 731, 618, 890, 869, 890, 869, 730]}, {'position': [753, 755, 835, 755, 835, 770, 753, 770], 'content': ['•心臟疾病'], 'cell': [619, 731, 618, 890, 869, 890, 869, 730]}, {'position': [45, 770, 75, 770, 74, 784, 45, 784], 'content': ['OHCA'], 'cell': [-1, -1, -1, -1, -1, -1, -1, -1]}, {'position': [622, 775, 670, 775, 670, 790, 622, 790], 'content': ['口癌症'], 'cell': [619, 731, 618, 890, 869, 890, 869, 730]}, {'position': [752, 775, 801, 775, 800, 791, 752, 790], 'content': ['口癫痫'], 'cell': [619, 731, 618, 890, 869, 890, 869, 730]}, {'position': [623, 795, 835, 796, 835, 812, 623, 811], 'content': ['•慢性阻塞性肺病口精神疾病'], 'cell': [619, 731, 618, 890, 869, 890, 869, 730]}, {'position': [44, 807, 207, 807, 207, 823, 44, 824], 'content': ['2、感覺怎麼的不舒服?'], 'cell': [-1, -1, -1, -1, -1, -1, -1, -1]}, {'position': [622, 816, 671, 816, 670, 833, 621, 832], 'content': ['口氣喘'], 'cell': [619, 731, 618, 890, 869, 890, 869, 730]}, {'position': [752, 816, 809, 816, 809, 832, 752, 832], 'content': ['□其他:'], 'cell': [619, 731, 618, 890, 869, 890, 869, 730]}, {'position': [46, 827, 75, 827, 74, 841, 45, 841], 'content': ['OHCA'], 'cell': [-1, -1, -1, -1, -1, -1, -1, -1]}, {'position': [622, 834, 671, 835, 671, 849, 621, 849], 'content': ['口中風'], 'cell': [619, 731, 618, 890, 869, 890, 869, 730]}, {'position': [752, 834, 818, 834, 818, 850, 752, 850], 'content': ['口不清楚'], 'cell': [619, 731, 618, 890, 869, 890, 869, 730]}, {'position': [45, 856, 224, 857, 224, 872, 45, 872], 'content': ['3、大約不舒服有多久了?'], 'cell': [-1, -1, -1, -1, -1, -1, -1, -1]}, {'position': [622, 855, 699, 855, 699, 871, 621, 870], 'content': ['口肝臟疾病'], 'cell': [619, 731, 618, 890, 869, 890, 869, 730]}, {'position': [752, 855, 783, 855, 783, 871, 752, 871], 'content': ['口無'], 'cell': [619, 731, 618, 890, 869, 890, 869, 730]}, {'position': [45, 876, 74, 876, 74, 890, 45, 889], 'content': ['不詳'], 'cell': [-1, -1, -1, -1, -1, -1, -1, -1]}, {'position': [726, 893, 761, 893, 761, 904, 726, 904], 'content': ['過敏史'], 'cell': [618, 890, 620, 906, 869, 906, 869, 890]}, {'position': [46, 903, 67, 905, 65, 921, 44, 919], 'content': ['4、'], 'cell': [-1, -1, -1, -1, -1, -1, -1, -1]}, {'position': [65, 905, 241, 905, 241, 921, 65, 921], 'content': ['還有其他地方不舒服嗎?'], 'cell': [-1, -1, -1, -1, -1, -1, -1, -1]}, {'position': [622, 913, 678, 914, 678, 930, 622, 929], 'content': ['口藥物:'], 'cell': [621, 907, 619, 1006, 868, 928, 867, 906]}, {'position': [782, 913, 846, 914, 846, 930, 782, 929], 'content': ['口不清楚'], 'cell': [621, 907, 619, 1006, 868, 928, 867, 906]}, {'position': [55, 925, 60, 936, 48, 942, 43, 931], 'content': ['海'], 'cell': [-1, -1, -1, -1, -1, -1, -1, -1]}, {'position': [39, 894, 39, 996, 24, 996, 24, 894], 'content': [''], 'cell': [-1, -1, -1, -1, -1, -1, -1, -1]}, {'position': [622, 939, 679, 940, 678, 955, 622, 954], 'content': ['□食物:'], 'cell': [621, 907, 619, 1006, 868, 928, 867, 906]}, {'position': [800, 941, 816, 941, 816, 955, 800, 955], 'content': ['無'], 'cell': [619, 906, 620, 954, 870, 953, 869, 930]}, {'position': [44, 956, 68, 957, 67, 970, 43, 969], 'content': ['5、'], 'cell': [-1, -1, -1, -1, -1, -1, -1, -1]}, {'position': [64, 955, 289, 955, 289, 971, 64, 971], 'content': ['評估頸椎是否损傷?(創傷患者)'], 'cell': [-1, -1, -1, -1, -1, -1, -1, -1]}, {'position': [622, 963, 679, 964, 679, 979, 622, 978], 'content': ['口其他:'], 'cell': [621, 907, 619, 1006, 868, 928, 867, 906]}, {'position': [99, 1009, 245, 1009, 245, 1026, 99, 1025], 'content': ['心肺功能停止登錄'], 'cell': [44, 1006, 44, 1029, 302, 1029, 302, 1006]}, {'position': [340, 1008, 486, 1008, 486, 1025, 340, 1025], 'content': ['OHCA事故地點型態'], 'cell': [302, 1006, 302, 1029, 527, 1029, 527, 1006]}, {'position': [528, 1009, 684, 1009, 684, 1026, 528, 1026], 'content': ['疑似心肌梗塞登錄'], 'cell': [527, 1006, 527, 1029, 686, 1029, 686, 1006]}, {'position': [676, 1008, 858, 1008, 858, 1026, 677, 1026], 'content': ['|符合疑似腦中風指標'], 'cell': [686, 1006, 686, 1029, 869, 1028, 869, 1006]}, {'position': [324, 1033, 345, 1034, 345, 1044, 324, 1044], 'content': ['住宅'], 'cell': [302, 1029, 302, 1183, 527, 1183, 527, 1029]}, {'position': [412, 1034, 483, 1033, 483, 1044, 412, 1045], 'content': ['□教育/學校'], 'cell': [302, 1029, 302, 1183, 527, 1183, 527, 1029]}, {'position': [532, 1034, 670, 1033, 670, 1044, 532, 1045], 'content': ['¥12小時内有胸痛或胸悶'], 'cell': [527, 1029, 527, 1183, 686, 1183, 686, 1029]}, {'position': [687, 1034, 815, 1033, 815, 1045, 687, 1045], 'content': ['是否異常:¥是■否'], 'cell': [685, 1031, 685, 1072, 869, 1072, 869, 1028]}, {'position': [46, 1047, 120, 1047, 120, 1059, 46, 1059], 'content': ['目擊者:有'], 'cell': [44, 1029, 44, 1183, 302, 1183, 302, 1029]}, {'position': [178, 1047, 214, 1046, 214, 1059, 178, 1060], 'content': ['口EMS'], 'cell': [44, 1029, 44, 1183, 302, 1183, 302, 1029]}, {'position': [138, 1047, 160, 1047, 160, 1060, 138, 1060], 'content': ['無'], 'cell': [44, 1029, 44, 1183, 302, 1183, 302, 1029]}, {'position': [531, 1055, 644, 1053, 644, 1065, 531, 1066], 'content': ['或下列情形之任2項:'], 'cell': [527, 1029, 527, 1183, 686, 1183, 686, 1029]}, {'position': [308, 1059, 402, 1058, 402, 1069, 309, 1070], 'content': ['¥工廠/工作地點'], 'cell': [302, 1029, 302, 1183, 527, 1183, 527, 1029]}, {'position': [409, 1059, 522, 1058, 522, 1069, 409, 1070], 'content': ['口捷運站/車站/機場'], 'cell': [302, 1029, 302, 1183, 527, 1183, 527, 1029]}, {'position': [687, 1057, 762, 1056, 762, 1067, 687, 1068], 'content': ['最後正常時間'], 'cell': [685, 1031, 685, 1072, 869, 1072, 869, 1028]}, {'position': [770, 1058, 770, 1066, 758, 1066, 758, 1058], 'content': ['1:'], 'cell': [685, 1031, 685, 1072, 869, 1072, 869, 1028]}, {'position': [44, 1079, 102, 1078, 102, 1090, 44, 1090], 'content': ['旁觀者CPR'], 'cell': [44, 1029, 44, 1183, 302, 1183, 302, 1029]}, {'position': [120, 1078, 169, 1078, 169, 1092, 120, 1092], 'content': ['有口無'], 'cell': [44, 1029, 44, 1183, 302, 1183, 302, 1029]}, {'position': [559, 1074, 559, 1088, 547, 1088, 547, 1074], 'content': ['喘'], 'cell': [527, 1029, 527, 1183, 686, 1183, 686, 1029]}, {'position': [774, 1078, 838, 1077, 838, 1088, 775, 1089], 'content': ['□無法得知'], 'cell': [686, 1029, 686, 1183, 870, 1183, 869, 1028]}, {'position': [309, 1083, 372, 1082, 372, 1093, 309, 1094], 'content': ['運動中心'], 'cell': [302, 1029, 302, 1183, 527, 1183, 527, 1029]}, {'position': [412, 1083, 507, 1082, 507, 1093, 412, 1094], 'content': ['口诊所/護理之家'], 'cell': [302, 1029, 302, 1183, 527, 1183, 527, 1029]}, {'position': [547, 1095, 585, 1095, 585, 1106, 547, 1106], 'content': ['冒冷汗'], 'cell': [527, 1029, 527, 1183, 686, 1183, 686, 1029]}, {'position': [686, 1095, 754, 1094, 755, 1105, 687, 1106], 'content': ['微笑测試(F)'], 'cell': [686, 1029, 686, 1183, 870, 1183, 869, 1028]}, {'position': [761, 1095, 864, 1094, 864, 1106, 761, 1107], 'content': ['□正常/□不正常'], 'cell': [686, 1029, 686, 1183, 870, 1183, 869, 1028]}, {'position': [45, 1107, 122, 1106, 122, 1120, 45, 1120], 'content': ['使用PAD:有'], 'cell': [44, 1029, 44, 1183, 302, 1183, 302, 1029]}, {'position': [137, 1107, 160, 1107, 160, 1120, 137, 1120], 'content': ['■無'], 'cell': [44, 1029, 44, 1183, 302, 1183, 302, 1029]}, {'position': [308, 1108, 378, 1108, 378, 1119, 308, 1119], 'content': ['¥街道/公路'], 'cell': [302, 1029, 302, 1183, 527, 1183, 527, 1029]}, {'position': [410, 1107, 456, 1107, 456, 1119, 410, 1119], 'content': ['¥其他:'], 'cell': [302, 1029, 302, 1183, 527, 1183, 527, 1029]}, {'position': [686, 1111, 753, 1110, 753, 1122, 686, 1123], 'content': ['舉臂测試(A)'], 'cell': [686, 1029, 686, 1183, 870, 1183, 869, 1028]}, {'position': [763, 1111, 866, 1110, 866, 1122, 763, 1123], 'content': ['□正常/□不正常'], 'cell': [686, 1029, 686, 1183, 870, 1183, 869, 1028]}, {'position': [548, 1116, 622, 1115, 622, 1126, 548, 1127], 'content': ['嗯心(或嘔呕吐)'], 'cell': [527, 1029, 527, 1183, 686, 1183, 686, 1029]}, {'position': [44, 1135, 186, 1133, 186, 1144, 44, 1146], 'content': ['到院前 ROSC:门有時間:'], 'cell': [44, 1029, 44, 1183, 302, 1183, 302, 1029]}, {'position': [309, 1132, 372, 1132, 372, 1143, 309, 1143], 'content': ['公共建集'], 'cell': [302, 1029, 302, 1183, 527, 1183, 527, 1029]}, {'position': [827, 1128, 846, 1128, 847, 1138, 827, 1138], 'content': ['右)'], 'cell': [686, 1029, 686, 1183, 870, 1183, 869, 1028]}, {'position': [771, 1127, 810, 1127, 810, 1140, 771, 1140], 'content': ['(口左'], 'cell': [686, 1029, 686, 1183, 870, 1183, 869, 1028]}, {'position': [412, 1132, 462, 1132, 462, 1143, 412, 1143], 'content': ['¥不清楚'], 'cell': [302, 1029, 302, 1183, 527, 1183, 527, 1029]}, {'position': [532, 1136, 610, 1136, 610, 1147, 532, 1147], 'content': ['□有心臟病史'], 'cell': [527, 1029, 527, 1183, 686, 1183, 686, 1029]}, {'position': [686, 1142, 752, 1141, 753, 1152, 686, 1153], 'content': ['言語測試(S)'], 'cell': [686, 1029, 686, 1183, 870, 1183, 869, 1028]}, {'position': [827, 1142, 864, 1141, 864, 1152, 827, 1153], 'content': ['不正常'], 'cell': [686, 1029, 686, 1183, 870, 1183, 869, 1028]}, {'position': [752, 1141, 810, 1141, 810, 1155, 751, 1154], 'content': ['□正常/'], 'cell': [686, 1029, 686, 1183, 870, 1183, 869, 1028]}, {'position': [139, 1157, 145, 1157, 145, 1167, 139, 1167], 'content': [''], 'cell': [44, 1029, 44, 1183, 302, 1183, 302, 1029]}, {'position': [308, 1154, 360, 1155, 360, 1166, 308, 1166], 'content': ['•療養院'], 'cell': [302, 1029, 302, 1183, 527, 1183, 527, 1029]}, {'position': [686, 1161, 753, 1160, 753, 1171, 686, 1172], 'content': ['動眼测試(G)'], 'cell': [686, 1029, 686, 1183, 870, 1183, 869, 1028]}, {'position': [761, 1160, 865, 1159, 865, 1172, 761, 1173], 'content': ['¥正常/口不正常'], 'cell': [686, 1029, 686, 1183, 870, 1183, 869, 1028]}] ``` ``` <html> <body> <table> <thead> <tr> <td colspan="7"><b>吉消防機關救護紀錄表義縣消防 □ ロ又力ロ／合</b></td> </tr> <tr> <td colspan="7"><b>▪圖</b></td> </tr> <tr> <td><b>日期受理時間</b></td> <td><b>2023–12–18</b></td> <td><b>出勤單位</b></td> <td><b>溪口91</b></td> <td><b>受案單位送達醫院時間</b></td> <td><b>離開醫院時間救災救護指揮中心□分隊自行�</b></td> <td><b>返隊待命時間</b></td> </tr> </thead> <tbody> <tr> <td>受理時間 12-18 13:19</td> <td>12-18 13：20 出勤時間</td> <td>到達現場時間 12–18-13:30</td> <td>離開現場時間</td> <td>l3：41 12–18</td> <td>離開醫院時間 12–18-14:23</td> <td>返隊待命時間 12-18-14:33</td> </tr> <tr> <td>發生地點嘉義縣溪口鄉妙崙村下崙43</td> <td>大林慈濟</td> <td>未運</td> <td>12-18</td> <td>協同處理單包</td> <td>□□</td> <td>中途取消</td> </tr> <tr> <td>送往醫院</td> <td>▪</td> <td>就近適當指揮中心 □ □傷病患或家屬要求</td> <td>未運</td> <td>有接觸未接觸出筋符命</td> <td>□支援勤務馨察盡戰三 □現場死亡</td> <td>□其他</td> </tr> <tr> <td>傷病患姓名</td> <td>黃陳甜</td> <td>性別</td> <td>□男 ▪經評估後判斷圖女</td> <td>■未經手傷病患財物明細：</td> <td>□□</td> <td>溺水</td> </tr> <tr> <td>／外籍患者護照號碼</td> <td>不詳</td> <td>年齡</td> <td>▪經評估後判斷</td> <td>□有保管人:</td> <td>〔簽章 )</td> <td>■同發生地點</td> </tr> <tr> <td>傷病患住址</td> <td>現場狀況〔此櫃可複選)</td> <td>現場狀況〔此櫃可複選)</td> <td>87歲 ▪經評估後判斷</td> <td>□創傷</td> <td>□□</td> <td>摔跌傷</td> </tr> <tr> <td>□急病 □呼吸問題〔嘀／呼奧急促</td> <td>□呼吸道問題〔異物更塞)</td> <td>口疑飯一氧化碳中毒</td> <td>到院前心肺功能停止</td> <td>背部外傷 □ 口腹部外傷</td> <td>電擊傷</td> <td>[到院前心肺功能停止 □生物咬蜝傷</td> </tr> <tr> <td>主訴： □ 1 威覺那裡不舒服? JOHCA</td> <td>□曉心／嘔吐／腹瀉</td> <td>孕婦急產 □</td> <td>到院前心肺功能停止</td> <td>口因交通事故 □非交通事故</td> <td>事故類別〔以傷病患為主</td> <td>過去病史 □腎臟疾病</td> </tr> <tr> <td>3丶大約不舒服有多久了?? 不詳</td> <td>家屬或友人代訴</td> <td>□路倒 □癲癇／抽搞</td> <td>口疑飯一氧化碳中毒 □疑似毒藥物中毒</td> <td>現場狀況〔此櫃可複選)</td> <td>擘落傷 □ [穿刺傷 □</td> <td>度燒燙傷 %</td> </tr> <tr> <td>□惠 □5 丶評估頸椎是否損傷?\創傷患 □主歸示無</td> <td>□肢體無力</td> <td>□行為急症／精神異常</td> <td>□行為急症／精神異常</td> <td>到院前心肺功能停止</td> <td>□</td> <td>癲癇 □心臟疾病 □慢性阻塞性肺病目精神疾�</td> </tr> <tr> <td>嵐斥心肺功能停止登錄</td> <td>12丶感覺怎麼的不舒服?</td> <td>□其他：</td> <td>□其他：</td> <td>到院前心肺功能停止</td> <td>□糖尿病</td> <td>過敏史</td> </tr> </tbody> </table> </body> </html> ``` ```json { "是否OHCA": "是", "日期": "2023/12/18", "病人姓名": "黃陳甜", "就診時間": "2023/12/18 14:01", "科別": "急診內科", "主治醫師": "張哲睿", "病歷號": "Q201430955", "年齡": "85", "性別": "女", "到院時間": "2023/12/18 14:01", "ETT tube": "Endo tube 7.0 fix 22cm.", "存活出院": "否", "TEE協助判斷死因": true, "可能原因": "Ill-defined and unknown cause of mortality [ICD:798.2], aortic dissection [ICD:]", "出勤單位": "溪口91", "發生地點": "溪口91", "到場時間": "13:20 12-18", "急診total_CPR_time": "22 分鐘" } ```