NLP Wiki

User Tools

Site Tools

Trace: • data_augmentation • large_reasoning_models • task_oriented_dialog • ensembling
nlp:pretraining

Differences

This shows you the differences between two versions of the page.

Link to this comparison view

Both sides previous revisionPrevious revision
Next revision		Previous revision
nlp:pretraining [2026/01/14 07:40] – [Contextualized Pretrained Models] jmflanignlp:pretraining [2026/08/03 03:13] (current) – [Pretraining (and Post-training) Methodology] jmflanig
Line 10: Line 10:
 For a history, see section 2.4 of [[https://arxiv.org/pdf/2003.08271.pdf|Qiu 2020]] or the related work in the [[https://d4mucfpksywv.cloudfront.net/better-language-models/language-models.pdf|GPT-2 paper]]. For a history, see section 2.4 of [[https://arxiv.org/pdf/2003.08271.pdf|Qiu 2020]] or the related work in the [[https://d4mucfpksywv.cloudfront.net/better-language-models/language-models.pdf|GPT-2 paper]].
   * [[https://arxiv.org/pdf/1103.0398|Collobert et al 2011 - Natural Language Processing (almost) from Scratch]]   * [[https://arxiv.org/pdf/1103.0398|Collobert et al 2011 - Natural Language Processing (almost) from Scratch]]
 +  * [[https://arxiv.org/pdf/1506.06726|Kiros et al 2015 - Skip-Thought Vectors]]
   * [[https://arxiv.org/pdf/1511.01432.pdf|Dai et al 2015 - Semi-supervised Sequence Learning]]   * [[https://arxiv.org/pdf/1511.01432.pdf|Dai et al 2015 - Semi-supervised Sequence Learning]]
 +  * [[https://arxiv.org/pdf/1705.00108|Peters et al 2017 - Semi-supervised Sequence Tagging with Bidirectional Language Models]]
   * [[https://arxiv.org/pdf/1611.02683.pdf|Ramachandran et al 2017 - Unsupervised Pretraining for Sequence to Sequence Learning]]   * [[https://arxiv.org/pdf/1611.02683.pdf|Ramachandran et al 2017 - Unsupervised Pretraining for Sequence to Sequence Learning]]
   * [[https://arxiv.org/pdf/1802.05365.pdf|Peters et al 2018 - Deep Contextualized Word Representations]]   * [[https://arxiv.org/pdf/1802.05365.pdf|Peters et al 2018 - Deep Contextualized Word Representations]]
Line 98: Line 100:
   * [[https://arxiv.org/pdf/2007.06778.pdf|Tu et al 2020 - An Empirical Study on Robustness to Spurious Correlations using Pre-trained Language Models]]   * [[https://arxiv.org/pdf/2007.06778.pdf|Tu et al 2020 - An Empirical Study on Robustness to Spurious Correlations using Pre-trained Language Models]]
  
-====== Pretraining Methodology ======+====== Pretraining (and Post-training) Methodology ======
 See also [[ml:scaling laws]]. See also [[ml:scaling laws]].
   * **Blog posts**:   * **Blog posts**:
Line 121: Line 123:
     * [[https://arxiv.org/pdf/2407.21783|Llama Team 2024 - The Llama 3 Herd of Models]]     * [[https://arxiv.org/pdf/2407.21783|Llama Team 2024 - The Llama 3 Herd of Models]]
     * [[https://arxiv.org/pdf/2412.19437|2024 - DeepSeek-V3 Technical Report]]     * [[https://arxiv.org/pdf/2412.19437|2024 - DeepSeek-V3 Technical Report]]
 +    * [[https://arxiv.org/pdf/2512.02556|2025 - DeepSeek-V3.2: Pushing the Frontier of Open Large Language Models]]
 +    * [[https://arxiv.org/pdf/2606.19348|2026 - DeepSeek-V4: Towards Highly Efficient Million-Token Context Intelligence]]
 +    * [[https://arxiv.org/pdf/2607.02770|Google Team 2026 - Gemma 4 Technical Report]]
 +  * **Closed Model System Cards**
 +    * These sometimes contain some interesting information, especially OpenAI's system cards
 +    * [[https://arxiv.org/pdf/2412.16720|OpenAI 2024 - OpenAI o1 System Card]]
 +    * [[https://deploymentsafety.openai.com/gpt-5-4-thinking/gpt-5-4-thinking.pdf|OpenAI 2026 - GPT-5.4 Thinking System Card]]
 +    * [[https://deploymentsafety.openai.com/gpt-5-5/gpt-5-5.pdf|OpenAI 2026 - GPT-5.5 System Card]]
 +    * [[https://deploymentsafety.openai.com/gpt-5-6-preview/gpt-5-6-preview.pdf|OpenAI 2026 - GPT-5.6 Preview System Card]]
  
 ===== Amount, Selection and Cleaning of Pretraining Data ===== ===== Amount, Selection and Cleaning of Pretraining Data =====
Line 160: Line 171:
   * [[Language Model]]   * [[Language Model]]
   * [[nlp:vision_and_language#Multimodal Pretraining]]   * [[nlp:vision_and_language#Multimodal Pretraining]]
 +  * [[Post-training]]
   * [[Prompting]]   * [[Prompting]]
   * [[ml:Semi-supervised Learning]]   * [[ml:Semi-supervised Learning]]
nlp/pretraining.1768376450.txt.gz · Last modified: 2026/01/14 07:40 by jmflanig
Donate Powered by PHP Valid HTML5 Valid CSS Driven by DokuWiki