首頁>編程相關>Ai源碼
下載

NLP，簡歷，深度學習紙閱讀表

我讀了這些與NLP和深度學習有關的論文。這是從基本到高級的各種論文。 ？此外，您可以通過單擊附帶的鏈接來檢查我的韓國紙質評論。

您可以在我的博客中查看更多紙質評論，代碼實施和數學描述< - 單擊此處

我的見識？

我寫了幾篇文章，以詳細解釋一些深​​度學習技術。這些文章可以在下表中找到。

標題博客鏈接
NLP中的擴展法是如何制定的？ ？ https://cartinoe5930.tistory.com/entry/how-has-scaling law-developed-in-nlp-%f0%9f%A4%94-nlp%EC%97%90%90%EC%84 ％9C級別LAW％EB％EB％8A％94-％EC％96％B4％EB％96％BB％EA％B2％B2％B2％8C％EB％B0％B0％9C％EC％A0％A0％84％EB％90％90％EC％98％EC％97％88％88％EC％84％EC％84％EA 84％EA B9％B9％B9％8C
封閉源？開源??那是什麼？ ？ https://cartinoe5930.tistory.com/entry/the-hopes-of-researchers-open-source-%F0%9f%A4%97-- ％EC％97％B0％EA％B5％AC％EC％9E％90％EB％93％A4％EC％9D％9D％ED％9D％AC％AC％EB％A7％9D-source-source-％F0％F0％9F％9F％a4％97
LM的上下文窗口，應該很長嗎？應該很短嗎？ ？ https://cartinoe5930.tistory.com/entry/lm%Ec%9d%98-Context-window-%B8%B8%EC%96%B4%EC%95%B C-％ED％95％A0％EA％B9％8C％EC％A7％A7％A7％EC％95％84％EC％95％BC-％ED％95％A0％EA％EA％B9％B9％8C-％F0％9F％93％93％8F％8F％8F％F0％F0％9F％9F％A4％A4％A8
評估LM的最佳方法是什麼？ ？ https://cartinoe5930.tistory.com/entry/lm%Ec%9d%84-%B0%B0%80%EC； B0%80%ED%95%A0-%EC%88%98-%EC%9E%88%EB%8A%94-%EB%B0%A9%EB%B2%95%EC%9D%80-%EB%AC%B4%EC%97%87%EC%9D%BC%EA%B9%8C-%F0%9F%98%8E
Chatgpt的表現越來越糟？ ！ ？ ！ ？ ？ https://cartinoe5930.tistory.com/entry/Chatgpt%EC%9D%98-%EC%84%B1%EBV%8A%A5%EC; 88-％EC％A2％8B％EC％95％84％EC％A7％80％EA％B3％A0-％EC％9E％88％EB％EB％8B％A4％EA4％EA B5％AC-％AC-％F0％9F％95％98％B2％B2％F0％F0％F0％9F％98％98％B2
您也可以微調！用peft？ https://cartinoe5930.tistory.com/entry/%EB%8B%B9%EC%8B%A0%EB%8F%84-fine-tuning- ％ED％95％A0-％EC％88％98-％EC％9E％88％EC％8A％B5％EB％8b％8B％88％EB％8B％A4-With-with-with-with-peft-％F0％9F％a4％97
讓我們像人類一樣逐步思考！ ？ https://cartinoe5930.tistory.com/entry/%95%95%9C-; ％EC％9D％B8％EA％B0％84％EC％B2％98％EB％9F％BC-％EC％83％83％9D％EA％B0％B0％ED％ED％95％B4％EB％B3％B3％B4％EC％EC％9E％90-％90-％90-％F0％F0％a7％a7％a7％a7％a0％a0％a0％f0％9F％a4 f0％a4 f0％a4％a4％94
微調方法的開發過程！從微調到RLHF？ ➡️？ https://cartinoe5930.tistory.com/entry/fine-tuning-method%ec%9d%98-%98-%EC%A7%84%ED%99%99%94-%B3%B3%BC%A0%95--
是時候微調chatgpt了！ ！ ⏰ https://cartinoe5930.tistory.com/entry/ %C%9D%B4%EC%A0%9c%9c%EB%8A%94-chatgpt%EB%A5%BC--fine-tuning-tuning-duning-duning%95%%A0-A0-下
噪音使LLM更好！ - 尼古納（Neftune） https://cartinoe5930.tistory.com/entry/noise-makes-llm-better-neftune-%F0%9F%98%89

自然語言處理

單詞嵌入和神經網絡

紙張標題紙或參考網站鏈接紙質評論
嵌入矩陣https://wikidocs.net/book/2155 https://cartinoe5930.tistory.com/entry/embedding-matrix-%Ed%95%99%Ecc%8A%B5
LSTM：長期術語內存https://colah.github.io/posts/2015-08-ENDERSDANDING-LSTMS/ https://cartinoe5930.tistory.com/entry/%EC%95%8C%B8%B8%B0-%B0-%89%BD%BD%B2%B2%8C-LSTM-NETWORKS-NETWORKS-NETWORKS-NETWORKS-NETWORKS-NETWORKS-NETWORKS-NETWORKS-NETWORKS-b4%B4%B4%B4%B4%95%B4%B4；
GRU：使用rnn編碼器解碼器進行統計機器翻譯的學習短語表示形式https://arxiv.org/abs/1406.1078 https://cartinoe5930.tistory.com/Entry/gru-empirical-evaluation-gated-gated-gated-recurrent-neurent-neurent-networks-on-sequence-modeling-modeling-modeling-modeling-vis-vis-vis-modeling-vis-modeling-vis-modeling-vis-modeling-vis-bc%Bc%BC%Ac 200-
LSTM與GRU：序列建模的封閉式複發神經網絡的經驗評估https://arxiv.org/abs/1412.3555 https://cartinoe5930.tistory.com/entry/lstm-vs-gru-%Eb%Ad%AD%90%90%B0%B0%B0%80-%EB%EB%8D%94-%EB%82%98%98%EC98%ECD%9D%84%84%B9%B9%8C- emp-emp-emp-emp irical評估門控的神經網絡 - 序列模型％EB％EB％85％BC％EB％aC％AC％B8-％EB％A6％AC％AC％EB％b7％B7 B7％B0

語言模型？

基本的

紙張標題紙或參考網站鏈接紙質評論
變形金剛：您需要注意的全部https://arxiv.org/abs/1706.03762 https://cartinoe5930.tistory.com/entry/transformer-cttern-iis-is-ally-you-need-deed-vis-%85%BC%BC%EB%ACB8-%EB%A6%AC%AC%EB%B7%B0%B0；
Elmo：深層上下文化的單詞表示https://arxiv.org/abs/1802.05365 https://cartinoe5930.tistory.com/entry/pre-trained-language-modeling-paper-reading-1--------------------------------------------------------------
BERT：深層雙向變壓器的預訓練以了解語言理解https://arxiv.org/abs/1810.04805 https://cartinoe5930.tistory.com/entry/Pre-trained-Language-Modeling-paper-reading2-BERT-Pre-training-of-Deep-Bidirectional-Transformers-for-Language-Understanding
GPT-1：通過生成預培訓來提高語言理解https://s3-us-west-2.amazonaws.com/openai-assets/research-covers/language-unsupervise/language_understanding_paper.pdf https://cartinoe5930.tistory.com/entry/pre-tration-language-modeling-paper-reading-paper-reading3-gpt-1-----------------------------------------------------------
GPT-2：語言模型是無監督的多任務學習者https://d4mucfpksywv.cloudfront.net/better-language-models/language_models_models_are_unsupervise_multitask_learners.pdf https://cartinoe5930.tistory.com/entry/gpt-2-language-models-models-are-unsuperist-multitask-learners-learners-pinder-learners-pirners-pimbc%BC%EB%AC%B8-%EB； eb%B8-%Eb%A6%AC%EB%EB%B7%B7%B7%B7%B7%B7%B7%B7%B7%B7%B7%B7%B7%B7%B0；
GPT-3：語言模型很少https://cartinoe5930.tistory.com/entry/gpt-3-language-models-are-few-shot-learners-learners-learners-pirners-pimeb%85%BC%EB%ACB8-%B8-%EB%A6%AC%EB%EB%B7%B0%B0； https://cartinoe5930.tistory.com/entry/gpt-3-language-models-are-few-shot-learners-learners-learners-pirners-pimeb%85%BC%EB%ACB8-%B8-%EB%A6%AC%EB%EB%B7%B0%B0；
變壓器-XL：超出固定長度上下文的細心語言模型https://arxiv.org/abs/1901.02860 https://cartinoe5930.tistory.com/entry/transformer-xl-cantentive-language-models-models-beyond-a- a- fixed-length-context-context-context-eb%85%BC%BC%EB%Ac%AC%AC%B8-%EB%EB%A6%ACAc%EBX%B7B7%B7％b7;
稀疏變壓器：用稀疏變壓器生成長序列https://arxiv.org/abs/1904.10509 https://cartinoe5930.tistory.com/entry/sparse-transformers-generating-long-sequence-with-sparse-transformers-parse-transformers-parse-transformers-dransformers-dransformers-dransformers-eb%85%BC%BC%EB;
XLNET：通用自動回報預處理以了解語言理解https://arxiv.org/abs/1906.08237 https://cartinoe5930.tistory.com/entry/xlnet-generalized-autoregrelisized-pretraining-for-langor-language-undercanding-undercanding-mertanding-vis-dranding-dranding-bc%Bc%BC%EB%Ac%B8-%EB %B8-%eb%A6%Aac 200b%EB%B7%B7%B0%B0;
Spanbert：通過表示和預測跨度改善預訓練https://arxiv.org/abs/1907.10529 https://cartinoe5930.tistory.com/entry/spanbert-improving-pre-training-by-representing-and-predicting-spans-presting-spans-pans-pans-drain-s-eb%85%BC%BC%EB%ACB8-%EB8-%EB%EB%A6%ACAC%AC%EB%B7%B7%B7%B0;
羅伯塔（Roberta）：一種強大優化的BERT預訓練方法https://arxiv.org/abs/1907.11692 https://cartinoe5930.tistory.com/entry/Roberta-a-Robustly-optimived-bert-pretraining-pretraining-pretraining-apphack-prace-%EB%85%BC%EB%AC%B8-%EB8-%EB；
句子 - 伯特：使用Siamese Bert-Networks的句子嵌入https://arxiv.org/abs/1908.10084 https://cartinoe5930.tistory.com/entry/sentence-bert-sentence-sentence-embeddings-using-siamese-bert-networks-bert-networks-mertworks-drworks-drworks-viss-bc%BC%EB%AC%B8-%Eb%A6%AC%AC%EB%EB%B7%B7%B0%B0
阿爾伯特：一個用於自我監督語言表徵學習的精簡版https://arxiv.org/abs/1909.11942 https://cartinoe5930.tistory.com/entry/albert-a-lite-lite-for-self-supersef-supersef-language-presentations-langue-presentations-presentations-prensentations-piresentations-pimeb%85%BC%BC%EB%AC%B8-%EB8-%EB%EB%A6%ACAC%EB%B7B7%B7%B7;
巴特：自然語言生成，翻譯和理解的序列前訓練序列前訓練https://arxiv.org/abs/1910.13461 https://cartinoe5930.tistory.com/entry/bart-denoising-sequence-sequence-toence-toence-pre-pre-training-for-natural-natural-natural-language-generation-generation-generation-generation-ranslation-and-comprehention-and-comprehens-d-comprehens-bcomeb%85%Bc%BC%EB%B%Acac，B8-
前LN變壓器：在變壓器體系結構中的圖層歸一化https://arxiv.org/abs/2002.04745 https://cartinoe5930.tistory.com/entry/pre-ln-transformer-on-layer-normalization-in-the-the-transformer-architecture-marchitutecor-br%85%BC%EB%AC%B8-%EB8-%EB%A6%ACAC 200B%EB%B7%B7%B7%B0；
Electra：訓練前文本編碼作為歧視者而不是發電機https://arxiv.org/abs/2003.10555 https://cartinoe5930.tistory.com/entry/Etectra-pre-training-training-training-text-encoders-ans-as-discriminators-rather-than-Generators
longformer：長篇文檔變壓器https://arxiv.org/abs/2004.05150 https://cartinoe5930.tistory.com/entry/longformer-the-long-document-document-transformer-dransformer-%EB%85%BC%EB%ACACB8-%EB%A6%AC%EB%B7%B0；
Bigbird：更長序列的變壓器https://arxiv.org/abs/2007.14062 https://cartinoe5930.tistory.com/entry/bigbird-transformers-for-longer-sequences-pinces-dimences-vis-vis-vis-vis-vis-bc%eb%Acac%B8-%EB%A6%Ac%EB%B7%B7%B0；
WebGPT：通過人類反饋的瀏覽器協助提問https://arxiv.org/abs/2112.09332 https://cartinoe5930.tistory.com/entry/webgpt-browser-assisted-question-question-withering-with-human-feedback-debback-%eb%85%Bc%BC%EB%AC%B8-%EB8-%EB %B%EB%A6%ACAC%EAC：EB%EB%B7%B7%B0%B0；
選擇：開放預訓練的變壓器語言模型https://arxiv.org/abs/2205.01068 https://cartinoe5930.tistory.com/entry/opt-open-pre-trained-tration-transformer-language-models-models-models-models-%EB%85%BC%EB%Acac%b8-%EB8-%EB%A6%AC%AC%EB%EB%B7%B0%B0
Mamba：具有選擇性狀態空間的線性時間序列建模https://arxiv.org/abs/2312.00752沒有計劃！

高效的模型？

紙張標題紙或參考網站鏈接紙質評論
Tinybert：將Bert提煉為自然語言理解https://arxiv.org/abs/1909.10351 https://cartinoe5930.tistory.com/entry/tinybert-distilling--bert-for-natural-natural-language-underage-undercanding-undercanding-dermestand-distand- %B%85%BC%EB%AC：B8-%B8-%Eb%A6%ACMAACAC%AC%AC%EB%B,B7%B0；
Distilbert：Bert的蒸餾版https://arxiv.org/abs/1910.01108 https://cartinoe5930.tistory.com/entry/distilbert-a-distille-version-of-bert-smaller-faster-faster-cheaper-cheaper-cheaper-cheaper-and-light---- and-light-%EB%85%BC%BC%EB%AC%B8-%eb%EB%A6%A6%AC：2%AC%EB%B7%B7%B7%B0;
重要的不僅很重要：小語言模型也很少是學習者（PET응용） https://arxiv.org/abs/2009.07118 https://cartinoe5930.tistory.com/entry/its-not-just-size-that-matter-mall-language-models-models-models-are-also-few-shot-shot-learners--phist-learners--eb%BC%85%BC%EB%EB%ACACB8-V8-

開源語言模型（縮放定律）？

紙張標題紙或參考網站鏈接紙質評論
Chinchilla：培訓的計算最佳大語模型https://arxiv.org/abs/2203.15556 https://cartinoe5930.tistory.com/entry/%EC%A7%80%80%B8%88-%88-%B9%8C%EC%A7%80%E C％9D％98-LM-SCALING LAW％EC％97％97％90％EB％8A％94-％EB％AC％AC％EC％EC％A0％9c％EC％A0％A0％90％90％EC％9D％B4-- %EC%9E%88%EB%8B%A4-%F0%9F%98%B6%E2%80%8D%F0%9F%8C%AB%EF%B8%8F-Chinchilla-Training-Compute-Optimal-Large-Language-Models-%EB%85%BC%EB%AC%B8-%EB%A6%AC%EB%B7%B0
腓熱：一套用於分析跨培訓和擴展的大型語言模型的套件https://arxiv.org/abs/2304.01373沒有計劃！
利馬：更少的是對齊https://arxiv.org/abs/2305.11206 https://cartinoe5930.tistory.com/entry/lima-sir-is-is-more-for-alignment-alignment-prim;
駱駝：開放有效的基礎語言模型https://arxiv.org/abs/2302.13971 https://cartinoe5930.tistory.com/entry/llama-open-and-foundation-foundation-language-models-models-models-%EB%85%BC%EB%ACACACB8-%EB%A6%AC%EB%EB%B7%B0；
wizardlm：授權大語言模型遵循複雜的說明https://arxiv.org/abs/2304.12244 https://cartinoe5930.tistory.com/entry/open-domain-instruction%ec%9d%98-%98-%9a%A8%B3%B3%B3%B3%BC-%9F%AA%84-WIZ ARDLM授權的大量語言模型到遵循式 - 複合構建 - ％EB％85％BC％EB％AC％AC％B8-％EB％A6％AC％AC％EB％B7％B7
WizardCoder：通過Evol-Inscruct授權代碼大語言模型https://arxiv.org/abs/2306.08568 https://huggingface.co/wizardlm/wizardcoder-15b-v1.0
巫師：通過增強的Evol-Infruction賦予大語言模型的數學推理能力https://arxiv.org/abs/2308.09583 https://huggingface.co/wizardlm/wizardmath-70b-v1.0
羊駝：強大的，可複制的指導跟踪模型https://crfm.stanford.edu/2023/03/13/alpaca.html https://cartinoe5930.tistory.com/entry/alpaca-a-a-strong-replicable-rplicable-instruction-following-model-model-model-del-model-%a6%Acas%EB%B7%B0；
Vicuna：一個開源聊天機器人給GPT-4留下深刻印象https://lmsys.org/blog/2023-03-03-30-Vicuna/ https://cartinoe5930.tistory.com/entry/vicuna-an-open-source-source-chatbot-impressing-gpt-4-; gpt-4-%EB%A6%AC%EB%B7%B0；
考拉：學術研究對話模型https://bair.berkeley.edu/blog/2023/04/03/koala/ https://cartinoe5930.tistory.com/entry/%EC%A4%91%EC%9A%9A%94%94%95%9C-%9C-%B1%B1%B4-%B4-%BA%BA%BE%ECB4； ％A7％80-％EC％95％8A％EB％8A％94高質量data-koala％F0％9F％90％A8-A-Dialogue-Modegoge-Model-for-Academic-for-Academic-Arsearc
Baize：一個開源聊天模型，具有自chat數據的參數效率調整https://arxiv.org/abs/2304.01196 https://cartinoe5930.tistory.com/entry/%F0%9f%90%90%B2BAIZE-AN-OPEN-SOURCE-CHAT-CHAT-CHAT-MODEL-WITH-PARAMETER--PARAMETER-FIFICE-TUNING-TUNING-on-self-Chat-c​​​​hat-c​​​​hat-c​​​​hat-data-chat-data-chat-data-chat-data-%EB%8555555%BC%BC%ACB8-B8--eb%B; b;
擴展數據受限的語言模型https://arxiv.org/abs/2305.16264 https://www.youtube.com/watch?v=tk0-sitkcmw&pp=ygugahr0chm6ly9hcnhpdi5vcmcvcmcvywjzlzizmduumtyynjq%3d
Falcon＆Repinedweb https://arxiv.org/abs/2306.01116 https://cartinoe5930.tistory.com/entry/open-llm-leaderboard%EB%A5%BC- %BC-%9C%A9C%A9%EC%93%B4-FALCON%F0%9f%A6%85-lllm-falcon-falcon-refelinedweb
ORCA：從GPT-4的複雜解釋痕跡中進行漸進學習https://arxiv.org/pdf/2306.02707 https://cartinoe5930.tistory.com/entry/%F0%9F%90%90%Acorca-progression-learning-from-complex-complex-explanation-traces-of-gpt-gpt-4-%EB%85%BC%BC%EB%B8-
PHI-1：您需要教科書https://arxiv.org/abs/2306.11644 https://cartinoe5930.tistory.com/entry/ %%95%84%Ec9a%9A%94%94%95%95%9C-%B1B1%B4-%B4-%EC98%A4%EC%A7; %A4%80%EC%9D%98-%EB%8D%B0%EC%9D%B4%ED%84%B0%EB%BF%90-%F0%9F%93%96-phi-1-Textbooks-Are-All-You-Need-%EB%85%BC%EB%AC%B8-%EB%A6%AC%EB%B7%B0
Alpagasus：培訓一個更好的羊駝的數據https://arxiv.org/abs/2307.08701稍後將上傳！
駱駝2：開放基礎和微調聊天模型https://arxiv.org/abs/2307.09288 https://cartinoe5930.tistory.com/entry/the-hopes-of-researchers-open-source-%F0%9f%A4%97-- ％EC％97％B0％EA％B5％AC％EC％9E％90％EB％93％A4％EC％9D％9D％ED％9D％AC％AC％EB％A7％9D-source-source-％F0％F0％9F％9F％a4％97
Platypus：快速，便宜且有力的LLMS https://arxiv.org/abs/2308.07317稍後將上傳！
代碼駱駝：代碼的開放基礎模型https://arxiv.org/abs/2308.12950沒有計劃
FLM-101B：開放LLM以及如何以10萬美元的預算培訓https://arxiv.org/pdf/2309.03852沒有計劃！
教科書就是您需要的II：PHI-1.5技術報告https://arxiv.org/abs/2309.05463 https://huggingface.co/microsoft/phi-1_5
OpenChat：使用混合質量數據推進開源語言模型https://arxiv.org/abs/2309.11235 https://github.com/imoneoi/openchat
Mistral 7b https://arxiv.org/abs/2310.06825 https://mistral.ai/news/announcing-mistral-7b/
Prometheus：在語言模型中誘導細粒度的評估能力https://arxiv.org/abs/2310.08491 https://huggingface.co/papers/2310.08491#652A8E7F30355BEBA68C1BE6
Zephyr：直接蒸餾LM對齊https://arxiv.org/abs/2310.16944 https://www.youtube.com/watch?v=tkzbg3mksio
ORCA2：教小語言模型如何推理https://arxiv.org/abs/2311.11045 https://www.microsoft.com/en-us/research/blog/orca-2-teaching-small-language-models-models-how-to-reason/
獵鷹系列開放語言模型https://arxiv.org/abs/2311.16867沒有計劃！
太陽能10.7b：用簡單而有效的深度縮放大型語言模型https://arxiv.org/abs/2312.15166沒有計劃！

大語言模型（LLM）？

紙張標題紙或參考網站鏈接紙質評論
LAMDA：對話應用程序的語言模型博客：https：//ai.googleblog.com/2022/01/lamda-towards-safe-grounded-and-high.html，論文：https：//arxiv.org/abs/2201.08239 https://cartinoe5930.tistory.com/entry/%B5%AC%AC%B8%B8%80%EC9D%9D%98-%EC%B5%9C%9C%B0%95-%ECB1B1B1B1B1B1B1%97%EBB4； ％80％ED％95％B4-％EC％95％8C％EC％95％84％EB％B3％B3％B4％EC％9E％90語言模型 - dialog-dialog-applications-％eb％EB％EB％85％BC％bc％eb％eb％AC％AC％AC％AC％AC％AC％B8-％EB％AC％AC％AC％AC％AC％AC％AC％B7％B7％B7
棕櫚：用途徑進行縮放語言建模博客：https：//ai.googleblog.com/2022/04/pathways-language-model-palm-scaling-to.html，論文：https：//arxiv.org/abs/2204.022311 1： https://cartinoe5930.tistory.com/entry/lamda%Ecd%9d%98-%EB%92%92%A4%EB%A5%Bc- bc-; 99％9C％EC％9A％A9％ED％95％95％9C％EC％B4％B4％88％EA％B1％B0％B0％EB％8C％80-％EC％96％B8％EC％EC％96％B4-％B4％B4％EB％AA％AA％AA％aa a. 2： https://cartinoe5930.tistory.com/entry/lamda%Ec%9d%9d%98-%EB%92%92%A4%EB%A5%BC-C-- %B-; 82％AC％EC％9A％A9％ED％95％9C-％EC％B4％88％EA％B1％B0％B0％EB％8C％8C％80-％EC％96％B8％EC％96％B4-％B4％B4％B4％AA％aa％AA％aa aa aa aa a.8dd dd％b8 palm-％b8 palm-％b8 palm-％b8 palm％eb％eb％aC％eb％eb％eb％eb％eb％b7％b7％b7％b7％b02
GPT-4：技術評論博客：https：//openai.com/research/gpt-4，論文：https：//arxiv.org/abs/2303.08774 https://cartinoe5930.tistory.com/entry/gpt-4-techinal-report-review
雙子座：一個高度強大的多模型的家族https://arxiv.org/abs/2312.11805沒有計劃！
字母2技術報告https://storage.googleapis.com/deepmind-media/alphacode2/alphacode2_tech_report.pdf沒有計劃！

微調

指導調用？ ‍？

紙張標題紙或參考網站鏈接紙質評論
Flan：微調語言模型是零攝的學習者https://arxiv.org/abs/2109.01652 https://cartinoe5930.tistory.com/entry/flan-fine-tuned-language-models-models-aare-zero-sero-shot-learners-learners-learners-pimbc%BC%EB%AC%B8-%B8-%eb%A6%AC%AC%EB%B7%B0%B0；
T0：多任務提示培訓可以啟用零擊任務概括https://arxiv.org/abs/2110.08207 https://cartinoe5930.tistory.com/entry/t0-multitask-prompted-training-enables-zero-sero-sero-sero-sero-sero-sero-shot-task-generalization-dask-generalization-bc%85%BC%BC%EB%AC%B8-%EB8-：
超自然說明：通過1600+ NLP任務的聲明說明進行概括https://arxiv.org/abs/2204.07705 https://cartinoe5930.tistory.com/entry/super-natural-instructions-generalization-gia-via-declarative-instructions-insustions-on-1600-nlp-tasks--tasks-pasks-inlp-tasks-eb%BC%85%BC%EB%EB%B；
不自然的說明：使用（幾乎）不是人工勞動的語言模型https://arxiv.org/abs/2212.09689稍後將上傳！
猜猜指令！翻轉學習使語言模型更強https://arxiv.org/abs/2210.02969 https://cartinoe5930.tistory.com/entry/guess-the-sinstruction-flippertion-learning-makes-language-models-models-models-model-stronger-zer-stronger-stronger-stronger-strong-loshot-learners-phis-pimiseb%85%Bc%BC%EB%Ac 200-
縮放指令 - 通信語言模型https://arxiv.org/abs/2210.11416 https://cartinoe5930.tistory.com/entry/scaling-instruction-finetuned-language-models-models-models-%EB%85%BC%EB%EB%ACB8-%EB%A6%AC%EB%EB%B7%B0；
探索培訓專家語言模型比教學調整的好處https://arxiv.org/abs/2302.03202 https://cartinoe5930.tistory.com/entry/Exploring-the-benefits-of-training-training-training-training-training-language-models-models-models-models-ober-instruction-tuning-tuning-duning-duning-duning-duning-duning-duning-dunis-85%Bc%Bc%Eb%Ac%B8-%EB/%EB%EB%A6%AC，eb%B7：B7：B7：B7：B7B7%B7;
ICIL：在文章中的教學學習https://arxiv.org/abs/2302.14691 https://cartinoe5930.tistory.com/entry/icil-in-in-context-instext-instruction-learning-mearning-%EB%85%BC%EB%AC%B8-%EB%A6%AC%EB%B7%B0%B0；
使用GPT-4進行指導調整https://arxiv.org/abs/2304.03277 https://cartinoe5930.tistory.com/entry/instruction-tuning-with-with-gpt-4-pm%EB%85%BC%BC%AC%AC%B8-%EB%A6%AC%EB%B7%B0%B0
FIP：固定輸入參數化以進行有效提示https://aclanthology.org/2023.findings-acl.533.pdf稍後將上傳！
Flacuna：使用Flan微調釋放Vicuna的問題解決能力https://arxiv.org/abs/2307.02053稍後將上傳！
也許只需要0.5％的數據：對低訓練數據指導調整的初步探索https://arxiv.org/abs/2305.09246稍後將上傳！
成為自我教育：引入最小的指示調整的早期停止標準https://arxiv.org/abs/2307.03692稍後將上傳！

從人類反饋（RLHF）學習的強化學習

紙張標題紙或參考網站鏈接紙質評論
RLHF（從人類反饋中學習） https://huggingface.co/blog/rlhf https://cartinoe5930.tistory.com/entry/%EC%82%AC%EB%9E%9E%8C%ECD%9D%98-%98-%94%BC%93%93%9C%EB%B0%B1B1B1%B1%ECM%9D%84 -％ED％86％B5％ED％95％9C-％EA B0％B0％95％ED％99％99％ED％95％99％EC％8A％B5-Forminforcement-fured-Formenting Learning-From-Human-Human-Feedback-Rlhf
用語言模型的紅色小組語言模型https://arxiv.org/abs/2202.03286 https://cartinoe5930.tistory.com/entry/red-teaming-teaming-language-models-with-language-models-models-models-%EB%85%BC%EB%ACACB8-;
指示：培訓語言模型遵循人類反饋的指示https://arxiv.org/abs/2203.02155 https://cartinoe5930.tistory.com/entry/instructgpt-training-training-models-models-models-to-follow-instructions-with-human-feedback-debback-%EB%85%BC%BC%EB%AC%B8-%EB8-：
通過從人類反饋中學習的強化學習培訓有用且無害的助手https://arxiv.org/abs/2204.05862 https://cartinoe5930.tistory.com/entry/training-a-helpful-and-harmless-assistant--with-reinforecement-learning-from-human-feedback-feedback-feedback-d.bc%85%Bc%Bc%BC%BC%ACB8-%B8-%EB;
Alpacafarm：從人類反饋中學習的方法的模擬框架https://arxiv.org/abs/2305.14387稍後將上傳！
幾乎是：通過合成反饋對齊大語言模型https://arxiv.org/abs/2305.13735 https://cartinoe5930.tistory.com/entry/aligning-large-lange-lange-models-models-though-synthetic-feedback-back-back-%eb%85%BC%EB%AC%B8-%EB8-%B8-％
從人類反饋中學習的開放問題和強化學習的根本局限性https://arxiv.org/abs/2307.15217稍後將上傳！
RLAIF：通過AI反饋從人類反饋中縮放增強的學習https://arxiv.org/abs/2309.00267沒有計劃！
steerlm：屬性條件為SFT作為RLHF的（用戶轉速）替代品https://arxiv.org/abs/2310.05344沒有計劃！
HelpSteer：Steerlm的多屬性幫助數據集https://arxiv.org/abs/2311.09528沒有計劃！

高效調整

紙張標題紙或參考網站鏈接紙質評論
適配器：NLP的參數效率學習https://arxiv.org/abs/1902.00751 https://cartinoe5930.tistory.com/entry/%EB%8B%B9%EC%8B%A0%EB%8F%84-fine-tuning- ％ED％95％A0-％EC％88％98-％EC％9E％88％EC％8A％B5％EB％8b％8B％88％EB％8B％A4-With-with-with-with-peft-％F0％9F％a4％97
前綴調整：優化發電的連續提示https://arxiv.org/abs/2101.00190 https://cartinoe5930.tistory.com/entry/%EB%8B%B9%EC%8B%A0%EB%8F%84-fine-tuning- ％ED％95％A0-％EC％88％98-％EC％9E％88％EC％8A％B5％EB％8b％8B％88％EB％8B％A4-With-with-with-with-peft-％F0％9F％a4％97
洛拉：大語言模型的低排名https://arxiv.org/abs/2106.09685 https://cartinoe5930.tistory.com/entry/%EB%8B%B9%EC%8B%A0%EB%8F%84-fine-tuning- ％ED％95％A0-％EC％88％98-％EC％9E％88％EC％8A％B5％EB％8b％8B％88％EB％8B％A4-With-with-with-with-peft-％F0％9F％a4％97
邁向參數效率轉移學習的統一觀點https://arxiv.org/abs/2110.04366稍後將上傳！
Unipelt：參數有效語言模型調整的統一框架https://arxiv.org/abs/2110.07577稍後將上傳！
（ia）^3：幾乎沒有參數效率的微調比在文化學習中更好，更便宜https://arxiv.org/abs/2205.05638稍後將上傳！
QLORA：量化LLM的有效微調https://arxiv.org/abs/2305.14314稍後將上傳！
堆疊更多的層不同：通過低排名更新的高級培訓https://arxiv.org/abs/2307.05695稍後將上傳！
Lorahub：通過動態洛拉組成有效的跨任務概括https://arxiv.org/abs/2307.13269稍後將上傳！

數據集？

紙張標題紙或參考網站鏈接紙質評論
指令挖掘：大語言模型的高質量指令數據選擇https://arxiv.org/abs/2307.06290沒有計劃！
蘇打水：社交常識性上下文化的百萬級對話蒸餾https://arxiv.org/abs/2212.10465沒有計劃！
mod：指導調整的面向模型的數據選擇https://arxiv.org/abs/2311.15653沒有計劃！
超越人類數據：通過語言模型來擴展解決問題的自我訓練https://arxiv.org/abs/2312.06585沒有計劃！
MagicOder：源代碼就是您所需要的https://arxiv.org/abs/2312.02120沒有計劃！
WaveCoder：通過精製數據生成的廣泛和多功能增強的指令調整https://arxiv.org/abs/2312.14187沒有計劃！
是什麼使良好的數據保持對齊方式：對教學調整中自動數據選擇的全面研究https://arxiv.org/abs/2312.15685沒有計劃！

及時工程？ ‍？

紙張標題紙或參考網站鏈接紙質評論
什麼是“及時工程”？查看我的博客！ https://cartinoe5930.tistory.com/entry/prompt-eendering%EC%9D%B4-%EB%AC%B4%EC%97%87%ECMEC9D%BC%BC%BC%B9%8C
COT：經過深思熟慮鏈的促使大語言模型引起推理博客：https：//ai.googleblog.com/2022/05/language-models-perform-reasoning-via.html，論文：https：//arxiv.org/abs/2201.11903 https://cartinoe5930.tistory.com/entry/lm%Ec%9D%B4-%EC%82%AC%AC%EB%9E%9E%9E%8C%B3%B3%BC-%BC-%9C%9C%A0%EC%82%82%82; %9C%EC%84%B8%EC%8A%A4%EB%A5%BC-%EA%B0%80%EC%A7%80%EA%B2%8C-%EB%90%9C%EB%8B%A4%EB%A9%B4-Chain-of-Thought-%EB%85%BC%EB%AC%B8-%EB%A6%AC%EB%B7%B0
零鏡頭嬰兒床：大型語言模型是零擊的推理器https://arxiv.org/abs/2205.11916 https://cartinoe5930.tistory.com/entry/large-lange-models-models-are-zero-sero-sero--reasoners-jbc%85%BC%EB%ACB8-%EB%A6%AC%AC%EB%BX%B7%B0%B0；
語言模型是多語言鏈的推理者https://arxiv.org/abs/2210.03057稍後將上傳！
自動cot：自動思想鏈在大型語言模型中提示https://arxiv.org/abs/2210.03493稍後將上傳！
COT KD：教小語言模型以推理https://arxiv.org/abs/2212.08410稍後將上傳！
TOT：思想樹：大型語言模型的故意解決問題https://arxiv.org/abs/2305.10601 https://cartinoe5930.tistory.com/entry/tree/tree-of-thoughts-deliberate-problem-solving-with-large-lange-lange-lange-models-models-models-models-models-models-models-models-modelm%85%BC%BC%EB%AC%B8-%EB8-%EB%EB%A6%ACAC 200B%B7%B7%B0%B0;
COT集合：通過經過思考的微型調整改善語言模型的零拍攝和幾乎沒有射擊的學習https://arxiv.org/abs/2305.14045 https://cartinoe5930.tistory.com/entry/cot-collection-collection-improving-zero-sero-shot-shot-and-hot-few-shot--few-shot-learning of-language-models-models-models-via-chain of-thought-thought-thought-thought-tuning-tuning-puning-puning-duning-duning-%EB%85%BC%BC%EB%EB%ACACACB8-%B8-%B8-%Ac%Ac%B-7B8-;
讓我們逐步驗證https://arxiv.org/abs/2305.20050 https://cartinoe5930.tistory.com/entry/lets-verify-step-by-step-dep-%Eb%85%BC%BC%AC%AC%B8-%EB%A6%AC%EB%B7%B7%B0；
衡量經濟鏈推理中的相性https://arxiv.org/abs/2307.13702稍後將上傳！
SOT：經過思考的骨骼：大語言模型可以進行平行解碼https://arxiv.org/abs/2307.15337稍後將上傳！
思想圖：解決大型語言模型的詳盡問題https://arxiv.org/abs/2308.09687稍後將上傳！
從稀疏到密集：GPT-4匯總，並鍊鍊鏈https://arxiv.org/abs/2309.04269沒有計劃！
驗證鏈在大語言模型中喚起了幻覺https://arxiv.org/abs/2309.11495 https://www.youtube.com/watch?v=l0zfjwregog&pp=ygugahr0chm6ly9hcnhpdi5vcmcvywjzlzizmdkumte0otu%3d
對比度鏈的促使https://arxiv.org/abs/2311.09277沒有計劃！
思想界線解開混亂的環境https://arxiv.org/abs/2311.08734沒有計劃！
系統2注意（您也可能需要的東西） https://arxiv.org/abs/2311.11829沒有計劃！
代碼鏈：使用語言模型的代碼模擬器推理https://arxiv.org/abs/2312.04474沒有計劃！

模型效率？

紙張標題紙紙質評論
閃存：快速且記憶效益的精確注意https://arxiv.org/abs/2205.14135 https://gordicaleksa.medium.com/eli5-flash-astention-5c444017022ad
指數級更快的語言建模https://arxiv.org/abs/2311.10770沒有計劃！
閃光燈中的LLM：有限的內存有效的大型語言模型推斷https://arxiv.org/abs/2312.11514沒有計劃！

方法 ？

紙張標題紙或參考網站鏈接紙質評論
NLP中的數據增強blogs: https://neptune.ai/blog/data-augmentation-nlp, https://amitness.com/2020/05/data-augmentation-for-nlp/?fbclid=IwAR11MkccCti-2cD93RYftNPHb7Wxdj7AlZG7NNG4EhPaBkmiJkcBPtdl1eo https://cartinoe5930.tistory.com/entry/data-aigmentation-methods-in-nlp
寵物：利用幾張鏡頭文本分類和自然語言推斷的披肩問題https://arxiv.org/abs/2001.07676 https://cartinoe5930.tistory.com/entry/pet-exploiting-cloze-questions-for-for-for-for-few-shot-text-classification-and-natural-language-inference-inperion-inference-inperion-inperion-inperion-inperion-inperion-inperion-inperion-bc%85%Bc%BC%EB%ACB；
途徑https://blog.google/technology/ai/introducing-pathways-nextways-next-generation-ai-Architection/ https://cartinoe5930.tistory.com/entry/%EB%A7%8c%95%BD-%BD-%AA； B0％81％EC％9D％84-％EB％8A％90％EB％82％82％84-％EC％88％98-％EC％9E％9E％88％EA％B2％B2％8C-％EB％90％90％9C％9C％EB％EB％EB％EB％A4％A4％EB％EB％EB％A9％A9％A9％B4-PATHWAYS-EB％B4-PATHWAREED-EB％A6％EB％a6％a6％AC％AC％AC％B7％B7％B7％B7
LMSI：大型語言模型可以自我突破https://arxiv.org/abs/2210.11610 https://cartinoe5930.tistory.com/entry/lmsi-large-lange-models-models-models-can-self-improve-%EB%85%BC%EB%ACB8-%B8-%EB%A6%AC%EB%EB%B7%B0%B0；
自我建造：將語言模型與自我生成的指示結合https://arxiv.org/abs/2212.10560 https://cartinoe5930.tistory.com/entry/self-instruct-aligning-language-model-model-with-self-generated-instructions-instructions-minstructions-dimstructions-pimbc%85%BC%EB%AC%AC%B8-%EB;
反射：語言加強學習的語言代理商https://arxiv.org/abs/2303.11366 https://cartinoe5930.tistory.com/entry/reflexion-language-agents-with-verbal-reinforcement-learning-learning-learning-learning-eb%85%BC%EB%AC%B8-%B8-%eb%A6%AC%AC%EB%EB%B7%B0%B0
自我refine：迭代精緻和自我反饋https://arxiv.org/abs/2303.17651 https://cartinoe5930.tistory.com/entry/self-refine-iterative-refinement-with-self-self-feedback-debback-%EB%85%BC%EB%AC%B8-%Eb%A6%AC%AC%EB%EB%B7%B0%B0；
煉油廠：中間表示的推理反饋https://arxiv.org/abs/2304.01904沒有計劃！
自由：迭代自我申請LLM由自我反饋生成升級https://kaistai.github.io/selfee/ https://cartinoe5930.tistory.com/entry/selfee-iterative-self-revisis--revising-llm-expate-by-self-feedback-generation-eb%85%Bc%BC%EB%Ac%B8-%EB8-%EB%A6%ACMAC%EB%B7%B7%B7%B0%B0;
GQA：培訓多頭檢查點的通用多電量變壓器模型https://arxiv.org/abs/2305.13245 https://aliissa99.medium.com/-A596E4D86F79
Shpherd：語言模型生成的評論家https://arxiv.org/abs/2308.04592稍後將上傳！
自我調整與教學反向翻譯https://arxiv.org/pdf/2308.06259稍後將上傳！
螺絲：通過修訂的模塊化框架https://arxiv.org/pdf/2309.13075沒有計劃！
Neftune：嘈雜的嵌入式嵌入改善教學方法https://arxiv.org/abs/2310.05914 https://cartinoe5930.tistory.com/entry/noise-makes-llm-better-neftune-%F0%9F%98%89
語言模型是超級馬里奧；來自同源模型的吸收能力作為免費午餐https://arxiv.org/abs/2311.03099沒有計劃！
Loramoe：革新專家的混合物，以維持世界知識在語言模型時https://arxiv.org/abs/2312.09979沒有計劃！

檢索增強發電（RAG）

紙張標題紙或參考網站鏈接紙質評論
用於知識密集型NLP任務的檢索生成一代https://arxiv.org/abs/2005.11401沒有計劃！
自我剝離：學會通過自我反省來檢索，產生和批評https://arxiv.org/abs/2310.11511沒有計劃！
儀錶盤：指導調整後檢索預告片https://arxiv.org/abs/2310.07713沒有計劃！
大型語言模型的檢索演示一代：一項調查https://arxiv.org/abs/2312.10997沒有計劃！

基準？和評估公制⚔️

紙張標題紙或參考網站鏈接紙質評論
大板凳艱難：具有挑戰性的大台任務以及經過思考鍊是否可以解決THAM https://arxiv.org/abs/2210.09261稍後將上傳！
大型語言模型不是公平的評估者https://arxiv.org/abs/2305.17926稍後將上傳！
MT-Bench: Judging LLM-as-a-judge with MT-Bench https://arxiv.org/abs/2306.05685稍後將上傳！
講座：朝著教學調節的大語言模型的整體評估https://arxiv.org/abs/2306.04757稍後將上傳！
燒瓶：基於對齊技巧的細粒度語言模型評估https://arxiv.org/abs/2307.10928稍後將上傳！
蓋亞：通用AI助手的基準https://arxiv.org/abs/2311.12983沒有計劃！

LLM的上下文

紙張標題紙或參考網站鏈接紙質評論
一個長度驅動變壓器https://arxiv.org/abs/2212.10554沒有計劃！
通過位置插值擴展大語模型的上下文窗口https://arxiv.org/abs/2306.15595 https://cartinoe5930.tistory.com/entry/lm%Ec%9d%98-Context-window-%B8%B8%EC%96%B4%EC%95%B C-％ED％95％A0％EA％B9％8C％EC％A7％A7％A7％EC％95％84％EC％95％BC-％ED％95％A0％EA％EA％B9％B9％8C-％F0％9F％93％93％8F％8F％8F％F0％F0％9F％9F％A4％A4％A8
Longnet：將變壓器擴展到1,000,000,000代幣https://arxiv.org/abs/2307.02486 https://cartinoe5930.tistory.com/entry/lm%Ec%9d%98-Context-window-%B8%B8%EC%96%B4%EC%95%B C-％ED％95％A0％EA％B9％8C％EC％A7％A7％A7％EC％95％84％EC％95％BC-％ED％95％A0％EA％EA％B9％B9％8C-％F0％9F％93％93％8F％8F％8F％F0％F0％9F％9F％A4％A4％A8
中間丟失：語言模型如何使用長上下文https://arxiv.org/abs/2307.03172 https://cartinoe5930.tistory.com/entry/lm%Ec%9d%98-Context-window-%B8%B8%EC%96%B4%EC%95%B C-％ED％95％A0％EA％B9％8C％EC％A7％A7％A7％EC％95％84％EC％95％BC-％ED％95％A0％EA％EA％B9％B9％8C-％F0％9F％93％93％8F％8F％8F％F0％F0％9F％9F％A4％A4％A8
紗線：大語模型的有效上下文窗口擴展https://arxiv.org/abs/2309.00071沒有計劃！

分析？

紙張標題紙或參考網站鏈接紙質評論
GPT為什麼可以在文本中學習？ https://arxiv.org/abs/2212.10559 https://cartinoe5930.tistory.com/entry/why-can-gpt-learn-in-context-vombc%85%BC%EB%AC%B8-%EB%A6%AC%EB%B7%B0%B0；
人工通用智能的火花：GPT-4的早期實驗論文：https：//arxiv.org/abs/2303.12712，YouTube：https：//www.youtube.com/watch?v=mqg3atgnxz0 https://cartinoe5930.tistory.com/entry/sparks-of-mandover-gnover-general-intelligence-early-ecperiments-with-gpt-4-pmpt-4-%Eb%85%BC%BC%EB%AC%B8-%EB8-%EB%EB%A6%ACAC 200B%EB%B7%B7%B7%B7;
模仿專有LLM的錯誤承諾https://arxiv.org/abs/2305.15717 https://cartinoe5930.tistory.com/entry/%B8%B8%B0%EC%A1%B4-Imitation-Model%EC%9D%80-%Ecc%9E%98%EB%AA%BB-BB-BB- %% 9 5％99％EC％8A％B5％EB％90％98％EA％B3％A0-％EC％9E％88％EB％8B％A4-％a4-％F0％9F％AB％A2-FALSE-FALSE-FALSE-FALSE-FALSE-FALSE-PALSIMITAIN-INMITMITAINS-PROPTARETARE-PROPREPRIETARY-L
圖盧：駱駝可以走多遠？在開放資源上探索教學調整狀態https://arxiv.org/abs/2306.04751稍後將上傳！
隨著時間的流逝，Chatgpt的行為如何改變？ https://arxiv.org/abs/2307.09009 https://cartinoe5930.tistory.com/entry/Chatgpt%EC%9D%98-%EC%84%B1%EBV%8A%A5%EC; 88-％EC％A2％8B％EC％95％84％EC％A7％80％EA％B3％A0-％EC％9E％88％EB％EB％8B％A4％EA4％EA B5％AC-％AC-％F0％9F％95％98％B2％B2％F0％F0％F0％9F％98％98％B2
大型語言模型還不能自我糾正推理https://arxiv.org/abs/2310.01798
大型語言模型與具有理論理論的代理商有多遠https://arxiv.org/pdf/2310.03051沒有計劃！
LLM可以遵循簡單的規則嗎https://arxiv.org/abs/2311.04235 https://www.youtube.com/watch?v=cy6o43037oy
駱駝在氣候變化中；通過Tulu 2增強LM適應https://arxiv.org/abs/2311.10702沒有計劃！
Chatgpt成立一周年；開源大語模型正在趕上https://arxiv.org/abs/2311.15653沒有計劃！
深入了解雙子座的語言能力https://arxiv.org/abs/2312.11444沒有計劃！

有趣的？

紙張標題紙或參考網站鏈接紙質評論
檢測：使用概率彎曲https://arxiv.org/abs/2301.11305 https://cartinoe5930.tistory.com/entry/%EC%9D%B4-%B4%B8%80%EC%9D%B4-lm%EC%9 D％B4-％EB％A7％8C％EB％93％A4％EC％96％B4％EB％82％B8-％EA B8％B8％80％EC％9D％BC％EA％B9％B9％B9％8C-％EB％8 F％84％EC％99％80％EC％A4％98-DETECTGPT-DETECTGPT-ZERO-SHOT-SHOT-MACHINE-MACHINE-MACHINE-TEXTEXT-DETECTION-使用驗證性 - 曲線 - ％EB％EB％EB％85％BC％EB％EB％AC％AC％AC％AC％AC％AC％B8-％EB％EB％EB％AC％AC％AC％AC％EB％B7％B7
檢查您的事實並重試：改進具有外部知識和自動反饋的大型語言模型https://arxiv.org/abs/2302.12813 https://cartinoe5930.tistory.com/entry/chatgpt%9d%98-hallucination-hallucination-marcin--%96%B4%B4%EB%96%BBBBB2%B2%B2%B2%B2%B2%95%B4%B4%B4%B2%B2%B2%B0%B0%B4%B4%B4%B4%B4%Ecc%ECECB4%ECB4%Ecc%95 ％bc-％ED％95％A0％EA％B9％8C檢查您的事實和嘗試的 - 不良式 - 語言模型，並進行外部知識和自動化返回
recurrentgpt：（任意）長文本的互動生成https://arxiv.org/abs/2305.13304 https://cartinoe5930.tistory.com/entry/Chatgpt%Ecd； ％EC％9A％A9％ED％95％9C％EB％8b％A4％EB％A9％B4-RECRENTGPTSTRACTICTARCTARCTARCTARCTIVE GERACTARTACTIVE GENTRACTIVE GEN-text-text-％EB％EB％85％BC％EB％EB％AC％AC％AC％AC％B8-％B8-％EB％EB％EB％AC％AC％EB％EB％B7％B7
大型語言模型作為工具製造商https://arxiv.org/abs/2305.17126 https://cartinoe5930.tistory.com/entry/lm%EC%9D%B4-%EB%8F%84%84%B5%AC%AC%EB%A5%BC-%BC-%EC%82%AC%AC%EC%9A%9A%A9; -％EB％90％9C％EB％8b％A4％EB％A9％B4-％F0％9F％94％AC-LARGE-LARGE-MODELAGE-MODELS-AS-AS-TOOL-MAKERS-％EB％EB％85％BC％EB％AC％AC％AC％AC％AC％AC％B8-％EB％EB％A6％A6％AC％AC％AC％EB％B7％B7％B7
LLM-Blender：結合具有成對排名和生成融合的大型語言模型https://arxiv.org/abs/2306.02561沒有計劃！
大型語言模型的知識蒸餾https://arxiv.org/abs/2306.08543 https://cartinoe5930.tistory.com/entry/KD%EC%97%90-%90-%EC%82%B4%B4%EC; ％EB％B3％B4％EC％9E％90-％F0％9F％98％9C-知識 - 知識降低了大量語言模型 - ％EB％EB％85％BC％EB％AC％AC％AC％AC％B8-％EB％A6％AC％AC％AC％EB％EB％B7％B7
與大語言模型學習數學推理的擴展關係https://arxiv.org/abs/2308.01825稍後將上傳！
TOOLLLM：促進LARE語言模型掌握16000多個現實世界中的API https://arxiv.org/abs/2307.16789稍後將上傳！
自我檢查：使用LLMS零射擊檢查自己的分步推理https://arxiv.org/abs/2308.00436稍後將上傳！
使用基於代碼的自我驗證的GPT-4代碼解釋器解決挑戰性的數學單詞問題https://arxiv.org/abs/2308.07921稍後將上傳！
大型語言模型作為優化器https://arxiv.org/abs/2309.03409沒有計劃！
菲亞特：將學習範式與教學加速調諧融合https://arxiv.org/abs/2309.04663 https://www.youtube.com/watch?v=ezszecrdte0&pp=ygugahr0chm6ly9hcnhpdi5vcmcvcmcvywjzlzizmdkumdkumdkumdq2njm%3d
對比度解碼改善了大語言模型的推理https://arxiv.org/abs/2309.09117 https://www.youtube.com/watch?v=nmr56tkwc1q&pp=ygugahr0chm6ly9hcnhpdi5vcmcmcvywjzlzizmdkumdkxmdkxmtc%3d
在說話之前先思考：帶有暫停令牌的培訓語言模型https://arxiv.org/abs/2310.02226 https://www.youtube.com/watch?v=mtj1jacr_yi
大型語言模型可以學習規則https://arxiv.org/abs/2310.07064沒有計劃！
文檔預測：語言建模超出文檔邊界https://arxiv.org/abs/2310.10638 https://www.youtube.com/watch?v=gi-0laailru
從錯誤中學習使LLM更好的推理器https://arxiv.org/abs/2310.20689沒有計劃！
語言模型可以是邏輯求解器https://arxiv.org/abs/2311.06158沒有計劃！
MART：通過多輪自動紅線改善LLM安全性https://arxiv.org/abs/2311.07689沒有計劃！
微調語言模型的事實模型https://arxiv.org/abs/2311.08401沒有計劃！
位置描述有關變壓器算術的重要事項https://arxiv.org/abs/2311.14737沒有計劃！
弱到緊密的概括：通過弱監督引起強大的能力https://arxiv.org/abs/2312.09390 https://openai.com/research/weak-to-to-strong-generalization
TinyGSM：使用小語言模型在GSM8K上獲得超過80％ https://arxiv.org/abs/2312.09241沒有計劃！

韓語LM ??

紙張標題紙或參考網站鏈接紙質評論
詞素意識的子詞令牌：針對各種韓國NLP任務的令牌化策略的實證研究https://arxiv.org/abs/2010.02534稍後將上傳！
大型語言模型可以帶來什麼變化？ HyperClova的密集研究：數十億個韓國生成預識的變壓器https://arxiv.org/abs/2109.04650稍後將上傳！

計算機視覺

紙張標題紙或參考網站鏈接紙質評論
CNN的歷史LENET，ALEXNET，VGGNET，GOOGLENET，RESNET，RESNEXT，SCEPTION，MOBILENET，DENSENET，EFIDENDENET，EFIDENDNET，CONSNEXT https://cartinoe5930.tistory.com/entry/cnn-network%ec%9d%98-%98-%Ecd%AD%EC%EC%82%AC
VIT：值得16 x 16個單詞的圖像：用於大規模圖像識別的變壓器https://arxiv.org/abs/2010.11929 https://cartinoe5930.tistory.com/entry/vit-an-image-worth-worth-16-x-16-words-transformers-for-for-image-rendimage-cognition-rendition-ar accale
SWIN Transformer：使用移動的Winodws的分層視覺變壓器https://arxiv.org/abs/2103.14030 https://cartinoe5930.tistory.com/entry/swin-transformer-hierarchical-vision-vision-vision-vision-transformer-using-shifted windows-windows-pimeb%85%BC%BC%EB%AC%B8-%EB8-%EB%EB%A6%ACACMACACAC 200B%EB%B7%B7%B0%B0；
剪輯：從自然語言監督中學習可轉移的視覺模型https://arxiv.org/abs/2103.00020 https://cartinoe5930.tistory.com/entry/clip-learning-transferable-visual-models-models-from-natural-natural-language-supervision-supervision-supervision-supervision-subervision-eb%85%BC%EB%EB%Ac%B8-%B8-%EB%EB%A6%ACAC%EB%B7%B7%B7;

多模式模型

紙張標題紙或參考網站鏈接紙質評論
讓我們了解VLM（視覺語言模型） https://huggingface.co/blog/vision_language_pretraining#supporting-vision-vision-language-models-models-in-%F0%9f%A4%97- transformers https://cartinoe5930.tistory.com/Entry/vlmvision-language-model%EC%97%97%90-%EB%80%80%ED%95%95%B4-%ECB4-%ECMB4-%ECB4-%ECM;
Visualbert：視覺和語言的簡單且性能的基線https://arxiv.org/abs/1908.03557 https://cartinoe5930.tistory.com/entry/visualbert-a-simple-and-performant-baseline-for-vision-vision-and-language-language-vis---%EB%85%BC%BC%EB%AC%B8-%eb%B8-%eb%A6%A6%AC%eb%B7%B7%B7%B7%B7%B7%B7%B7%B7%B7%B7%B7%B7%B7%B7%B7%B7%B7%B7%B7%B7%B7%B0%B0%B7%B7%B7%B7%B7%B7%B7%B7%B7%B7%B7%B7％
Vilbert：視覺和語言任務的訓練前任務無形語言語言表示https://arxiv.org/abs/1908.02265 https://cartinoe5930.tistory.com/entry/vilbert-pretraining-task-agnostic-visiolistic-rymenistic-for-visual and語言和語言任務
LXMERT：從變形金剛學習的跨模式編碼器表示https://arxiv.org/abs/1908.07490 https://cartinoe5930.tistory.com/entry/lxmert-learning-cross-modality-modality-ecoder-coder-presentations-from-transformers-transformers-transformers-prom-prom-prom-pimeb%85%Bc%BC%EB%Ac%B8-%EB8-%EB%EB%A6%ACAC%EBX%B7B7%B7;
VL-Bert：通用視覺語言表示的預培訓https://arxiv.org/abs/1908.08530 https://cartinoe5930.tistory.com/entry/vl-bert-pre-training-of-generic-generic-visual-linguistic-presentations-presentations-presentations-presentations-eb%85%BC%BC%EB%AC%B8-%EB %B8-%EB%A6%ACMACACMAC；
VLP：圖像字幕和VQA的統一視力語言預訓練https://arxiv.org/abs/1909.11059 https://cartinoe5930.tistory.com/Entry/vlp-unified-vision-vision-language-pre-pre-traning-for-image-captioning-and-veqa-vqa-vqa-vqa-%EB%85%BC%BC%EB%Ac%AC%B8-%B8-%EB%EB%A6%ACAC%EB7B7B7%B7;
奧斯卡：對象 - 符合視力語言任務的預先培訓對準預訓練https://arxiv.org/abs/2004.06165 https://cartinoe5930.tistory.com/entry/carcar-object-semantics-Aligned-pre-training-for-training-for-vision-tanguage-tasks-tasks-tasks-pasks-pasts-dasks-pimeb%85%BC%EB%AC%B8-%EB8-%EB%EB%A6%ACAC%EB%B7%B7%B7%B7%B7;
VINVL：視覺模型中的視覺表示https://arxiv.org/abs/2101.00529 https://cartinoe5930.tistory.com/entry/vinvl-revisiting-visual-visual-presentations-in-vision-language-models-models-models-models-models-models-models-pim;
vilt：無卷積或地區監督的視覺和語言變壓器https://arxiv.org/abs/2102.03334 https://cartinoe5930.tistory.com/entry/vilt-vision-and-language-transformer-without-convolution-convolution-or-region-supervision-supervision-supervision-supervision-eb%85%Bc%BC%EB%AC%Ac%B8-%EB8-%EB%EB%A6%ACAC%EBM%B7%B7%B7;
對齊：使用嘈雜的文本監督來擴展視覺和視覺語言表示學習https://arxiv.org/abs/2102.05918 https://cartinoe5930.tistory.com/entry/Align-scaling-up-ip--visual-and-visial-vision-language-representation-with-noisy-text-supervision-supervision-supervision-supervision-bc%85%BC%EB%EB%B8-
ALBEF：動量蒸餾的視覺和語言表示學習https://arxiv.org/abs/2107.07651 https://cartinoe5930.tistory.com/entry/albef-vision-and-language-presentation-learnning-with-momentum-distillation-momentum-distillation-distillation-momentum-distillation-distillation-eb%85%BC%EB%AC%B8
SIMVLM：簡單的視覺語言模型，並通過弱監督進行預處理https://arxiv.org/abs/2108.10904 https://cartinoe5930.tistory.com/entry/simvlm-simple--visual-language-language-model-model-pre-pre-with-with-with-supervision-supervision-supervision-supervision-pim;
VLMO：統一的視覺語言預訓練與模式外科的混合物https://arxiv.org/abs/2111.02358 https://cartinoe5930.tistory.com/entry/vlmo-unified-vision-vision-language-pre-pre-pre-training-with-mixture-modality-modality-experts-experts-experts-xexeb%85%BC%BC%EB%AC%AC%B8-%EB %B8-%EB%A6;
點亮：零射傳輸，鎖定圖像文本調整https://arxiv.org/abs/2111.07991 https://cartinoe5930.tistory.com/entry/lit%F0%9f%94%A5- Zero-Zero-shot-shot-transfer-with-with-locked-image-image-image-image-image-text-tuning-tuning-duning-duning-duning-duning-duning-duniseb%85%Bc%BC%EB%B8--
Flava：基礎語言和遠見模型https://arxiv.org/abs/2112.04482 https://cartinoe5930.tistory.com/entry/flava-a-foundational-language-and-vision-vision-alignment-model-del-model-%Eb%85%BC%EB%AC%B8-%Eb%A6%ACMAC%AC%EB%B7%B0%B0；
Blip：引導語言圖像預訓練，用於統一視力語言理解和產生https://arxiv.org/abs/2201.12086 https://cartinoe5930.tistory.com/entry/blip-bootstapping-language-image-image-image-image-pre-training-fro-unified-vision-vision-vision-vision-language-language-underage-undercranding-and-generation-and-generation--generation-generation-generation-generation-generation-generation-bc%85%Bc%BC%EB%B8-b8-b8-b8-b8-b8-b8- b8-b8-b8-

深度學習概念

紙或發布標題參考網站鏈接審查
知識蒸餾：在神經網絡中提取知識https://arxiv.org/abs/1503.02531 https://cartinoe5930.tistory.com/entry/distilling-the-the--knowledg-in-a-neur-network-vork-%EB%85%BC%BC%EB%ACB8-%EB%A6%AC%AC%EB%B7%B0%B0；
什麼是零射擊，一次性，很少的學習？查看我的博客！ https://cartinoe5930.tistory.com/entry/Zero-Shot-one-shot-few-shot-few-shot-learning%ec%ec%9d%B4-%EB%AC%Ac%B4%ECB4%EC97%EC97%EC9D%BC%BC%BC%AV%B9%8C；
展開
附加信息
相關應用
爲您推薦
相關資訊 全部