NLP Wiki

User Tools

Site Tools

Trace: • paraphrase • question_answering • commonsense_reasoning • information_retrieval • chatgpt • domain_adaptation • intelligent_tutor • fairness
ml:scaling_laws

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
ml:scaling_laws [2024/06/21 22:04] – jmflanigml:scaling_laws [2025/06/01 23:09] (current) – [Related Pages] jmflanig
Line 5: Line 5:
   * [[https://arxiv.org/pdf/2203.15556.pdf|Hoffmann et al 2022 - Training Compute-Optimal Large Language Models]]   * [[https://arxiv.org/pdf/2203.15556.pdf|Hoffmann et al 2022 - Training Compute-Optimal Large Language Models]]
   * [[https://arxiv.org/pdf/2207.10551.pdf|Tay et al 2022 - Scaling Laws vs Model Architectures: How Does Inductive Bias Influence Scaling?]]   * [[https://arxiv.org/pdf/2207.10551.pdf|Tay et al 2022 - Scaling Laws vs Model Architectures: How Does Inductive Bias Influence Scaling?]]
 +  * **[[https://arxiv.org/pdf/2405.10938|Ruan et al 2024 - Observational Scaling Laws and the Predictability of Language Model Performance]]** Does a multi-dimensional regression (fitting a sigmoid) to predict the performance across model "families" (LLaMA, GPT-3, etc)
 +  * [[https://arxiv.org/pdf/2406.19146|Porian et al 2024 - Resolving Discrepancies in Compute-Optimal Scaling of Language Models]]
 +  * [[https://arxiv.org/pdf/2410.11840|Choshen et al 2024 - A Hitchhiker's Guide to Scaling Law Estimation]]
  
 ==== Training LLMs ==== ==== Training LLMs ====
Line 11: Line 14:
  
 ==== Emergent Abilities ==== ==== Emergent Abilities ====
 +See also [[nlp:Language Model#Origin of Capabilities|Language Model - Origin of Capabilities]].
 +
 +  * GPT-3: [[https://arxiv.org/pdf/2005.14165.pdf|Brown et al 2021 - Language Models are Few-Shot Learners]] GPT-3 showed emergent abilities.  See for example Fig 3.10.
 +  * [[https://arxiv.org/pdf/2206.07682|Wei et al 2022 - Emergent Abilities of Large Language Models]]
   * [[https://arxiv.org/pdf/2304.15004|Schaeffer et al 2023 - Are Emergent Abilities of Large Language Models a Mirage?]]   * [[https://arxiv.org/pdf/2304.15004|Schaeffer et al 2023 - Are Emergent Abilities of Large Language Models a Mirage?]]
 +  * **[[https://arxiv.org/pdf/2310.03262|Hu et al 2023 - Predicting Emergent Abilities with Infinite Resolution Evaluation]]** Does bootstrap resampling to get a very fine-grained measure of model capabilities by resampling until they get the desire behavior a certain number of times.  Can be used to predict emergent capabilities from very small models that rarely exhibit the desired behavior.
 +  * [[https://arxiv.org/pdf/2403.15796|Du et al 2024 - Understanding Emergent Abilities of Language Models from the Loss Perspective]]
  
 ===== Related Pages ===== ===== Related Pages =====
   * [[Hyperparameter Tuning]]   * [[Hyperparameter Tuning]]
   * [[nlp:Language Model]]   * [[nlp:Language Model]]
 +  * [[nlp:Language Model#Origin of Capabilities|Language Model - Origin of Capabilities]]
   * [[nlp:pretraining#Pretraining Methodology]]   * [[nlp:pretraining#Pretraining Methodology]]
ml/scaling_laws.1719007441.txt.gz · Last modified: 2024/06/21 22:04 by jmflanig
Donate Powered by PHP Valid HTML5 Valid CSS Driven by DokuWiki