Traditional definition of a language model (LM): a language model is a probability distribution over sentences, that is, it assigns probabilities to sentences. Language models can usually compute the probability of the next word given a sequence of words (autoregressive language models), or in the case of masked language models, the probability of a word given a surrounding context.

Note: unlike autoregressive language models, masked language models usually can't be used to compute the probability of a sentence, and so they aren't really “language models” in the traditional sense.

To experiment with an autoregressive language model or masked language model, see online demos below.

• Introductory Material
  • Basic intro, and n-gram language modeling
    • Language modeling by Mike Collins
    • Probabilistic Language Models by Noah Smith
    • Chapter 3 of Speech and Language Processing
  • Neural language models
    • Section 7.5 of Chapter 7 of Speech and Language Processing
  • Large language models
    • 2021 - AMMUS : A Survey of Transformer-based Pretrained Models in Natural Language Processing Comprehensive overview at the time
    • Wei et al 2023 - An Overview on Language Models: Recent Developments and Outlook
    • Liu et al 2024 - Understanding LLMs: A Comprehensive Overview from Training to Inference Wow, really good
    • Naveed et al 2024 - A Comprehensive Overview of Large Language Models
    • For another nice introduction, see related work of Taylor 2022 (p. 3)
    • Bowman 2023 - Eight Things to Know about Large Language Models
    • Minaee et al 2024 - Large Language Models: A Survey
    • Wan et al 2023 - Efficient Large Language Models: A Survey
    • Zheng et al 2023 - GPT-Fathom: Benchmarking Large Language Models to Decipher the Evolutionary Path towards GPT-4 and Beyond
    • Zhao et al 2023 - A Survey of Large Language Models
    • Weng 2024 - Navigating the Landscape of Large Language Models: A Comprehensive Review and Analysis of Paradigms and Fine-Tuning Strategies
    • 2025 - International AI Safety Report (Has a good non-technical overview of AI, ML & LLMs)
• Language models in the news, etc
  • Wired - GPT-3
  • Twitter GPT-3 code example (Sharif Shameem) “I only had to write 2 samples to give GPT-3 context for what I wanted it to do. It then properly formatted all of the other samples… If I wanted it to write output plain HTML/CSS instead of JSX, all I would have to do would be to re-write my 2 initial samples in HTML/CSS. Then all of GPT-3's outputs would be in plain HTML/CSS.”
  • We Might See A 100T Language Model In 2022 archive.org Nice overview of some large language models in 2022
• Bibliographies
  • Awesome-LLM

• n-gram Models: Old classic papers, and recent papers
  • Teh 2006 - A Bayesian Interpretation of Interpolated Kneser-Ney
  • Malagutti et al 2024 - The Role of n-gram Smoothing in the Age of Neural Networks
• Fill-In-the-Middle
  • Bavarian et al 2022 - Efficient Training of Language Models to Fill in the Middle
  • See also Starcoder Fill-In-The-Middle
• Memory, Cache and Retrieval-Augmented Language Models
  • Khandelwal et al 2019 - Generalization through Memorization: Nearest Neighbor Language Models
  • Yogatama et al 2021 - Adaptive Semiparametric Language Models
  • 2021 - Improving language models by retrieving from trillions of tokens (blog)
  • Wu et al 2022 - Memorizing Transformers Uses k-NN lookup with fixed embeddings to retrieve relevant examples
• Clark et al 2022 - Unified Scaling Laws for Routed Language Models

See also Ecosystem Graphs for a more complete list.

This is a list of large, GPT-style autoregressive LMs. See also pretraining for another list of large language models and GPT-3 alternatives.

• Jozefowicz et al 2016 - Exploring the Limits of Language Modeling It's interesting to see how far we've come since 2016.
• GPT: Radford et al 2018 - Improving Language Understanding by Generative Pre-Training old link
• GPT-2: Radford et al 2019 - Language Models are Unsupervised Multitask Learners original github Annotated GPT-2 Illustrated GPT-2 Interestingly, GPT-2 does not include a bias term in the final linear layer for the vocab, see here and here.
• GPT-3: Brown et al 2019 - Language Models are Few-Shot Learners OpenAI cookbook
• Turing-NLG: A 17-billion-parameter language model by Microsoft
• Fedus et al 2021 - Switch Transformers: Scaling to Trillion Parameter Models with Simple and Efficient Sparsity
• Gopher: Rae et al 2021 - Scaling Language Models: Methods, Analysis & Insights from Training Gopher blog
• Jurassic-1: Lieber et al 2022 - Jurassic-1: Technical Details and Evaluation model blog
• Megatron-Turing NLG: Smith et al 2022 - Using DeepSpeed and Megatron to Train Megatron-Turing NLG 530B, A Large-Scale Generative Language Model Microsoft blog NVidia blog Researcher access code models
• Chinchilla: Hoffmann et al 2022 - Training Compute-Optimal Large Language Models Says most LLMs are undertrained, and trains a compute budget optimal size language model using the same dataset as Gopher. blog1 blog2
• PaLM: Chowdhery et al 2022 - PaLM: Scaling Language Modeling with Pathways blog
• GPT-NeoX-20B: Black et al 2022 - GPT-NeoX-20B: An Open-Source Autoregressive Language Model Has an interesting description of the hardware they used
• OPT: Zhang et al 2022 - OPT: Open Pre-trained Transformer Language Models blog models
• Bloom: model card Training readme Tensorboard log
• Zeng et al 2022 - GLM-130B: An Open Bilingual Pre-trained Model
• Li et al 2023 - FLM-101B: An Open LLM and How to Train It with $100K Budget

• ChatGPT
  • For ChatGPT, see also ChatGPT.
  • Laskar et al 2023 - A Systematic Study and Comprehensive Evaluation of ChatGPT on Benchmark Datasets
  • Ronan & Schneider - Can Chat GPT solve a Linguistics Exam?
  • Frieder et al 2023 - Mathematical Capabilities of ChatGPT
• Creativity
  • Zhao et al 2024 - Assessing and Understanding Creativity in Large Language Models
  • Tian et al 2024 - MacGyver: Are Large Language Models Creative Problem Solvers?
• Self-Correction
  • Liu et al 2024 - Large Language Models have Intrinsic Self-Correction Ability
• Use of Context
  • Khandelwal et al 2018 - Sharp Nearby, Fuzzy Far Away: How Neural Language Models Use Context (Old, no longer applies to transformer models)
  • Liu 2023 - Lost in the Middle: How Language Models Use Long Contexts
    • Hutter et al 2025 - Lost but Not Only in the Middle

• Madabushi et al 2025 - Neither Stochastic Parroting nor AGI: LLMs Solve Tasks through Context-Directed Extrapolation from Training Data Priors
• Machine Translation
  • Briakou et al 2023 - Searching for Needles in a Haystack: On the Role of Incidental Bilingualism in PaLM's Translation Capability
  • Balashov 2025 - Translation in the Wild

• Overviews
  • Chang et al 2023 - A Survey on Evaluation of Large Language Models
• lm-evaluation-harness: LM Evaluation Harness (EleutherAI) (Released May 2021)
• Mizrahi et al 2024 - State of What Art? A Call for Multi-Prompt LLM Evaluation
• lm-eval: Biderman et al 2024 - Lessons from the Trenches on Reproducible Evaluation of Language Models
• Small-scale Evaluations
  • Polo et al 2024 - tinyBenchmarks: evaluating LLMs with fewer examples
• Effects of Length and Irrelevant Context
  • Levy et al 2024 - Same Task, More Tokens: the Impact of Input Length on the Reasoning Performance of Large Language Models

• Shaikh et al 2025 - Navigating Rifts in Human-LLM Grounding: Study and Benchmark

• Bender et al 2021 - On the Dangers of Stochastic Parrots: Can Language Models Be Too Big?
• Rogers & Luccioni 2023 - Position: Key Claims in LLM Research Have a Long Tail of Footnotes

• Xie et al 2023 - Data Selection for Language Models via Importance Resampling

• Language Adaptive Fine-Tuning (LAFT):
  • Artetxe et al 2019 - On the Cross-lingual Transferability of Monolingual Representations
  • Pfeiffer et al 2020 - UNKs Everywhere: Adapting Multilingual Language Models to New Scripts
  • Recycling: Vries & Nissim 2021 - As Good as New. How to Successfully Recycle English GPT-2 to Make Models for Other Languages This one works the best so far. Retrains the embeddings keeping the Transformer layers fixed
  • Zhao & Schütze - Discrete and Soft Prompting for Multilingual Models
  • Multi-lingual version: Alabi et al 2022 - Adapting Pre-trained Language Models to African Languages via Multilingual Adaptive Fine-Tuning
  • Lin et al 2022 - Few-shot Learning with Multilingual Generative Language Models
  • Shi et al 2022 - Language Models are Multilingual Chain-of-Thought Reasoners
  • Marchisio et al 2022 - Mini-Model Adaptation: Efficiently Extending Pretrained Models to New Languages via Aligned Shallow Training
  • Michal Štefánik et al 2023 - Resources and Few-shot Learners for In-context Learning in Slavic Languages (Dataset)
  • Zhao et al 2024 - LLaMA Beyond English: An Empirical Study on Language Capability Transfer

• Lazaridou et al 2021 - Mind the Gap: Assessing Temporal Generalization in Neural Language Models
• Li & Flanigan 2024 - Future Language Modeling from Temporal Document History
• Drinkall et al 2024 - Time Machine GPT

See also Dense Document Retrieval with LLMs.

• Extracting Training Data
  • Carlini et al 2020 - Extracting Training Data from Large Language Models github
• Membership Inference for Training Data
  • (Decide if some sample data is in the training data or not)
  • Related page: Membership Inference Attacks
  • Song & Shmatikov 2018 - Auditing Data Provenance in Text-Generation Models
• Language Models as Knowledge Bases?
• Feldman et al 2019 - Commonsense Knowledge Mining from Pretrained Models
• Jiang et al 2020 - How Can We Know What Language Models Know?
• Cao et al 2021 - Knowledgeable or Educated Guess? Revisiting Language Models as Knowledge Bases
• Liu et al 2022 - Generated Knowledge Prompting for Commonsense Reasoning
• Huang et al 2022 - Language Models as Zero-Shot Planners: Extracting Actionable Knowledge for Embodied Agents
• Akyürek et al 2022 - Tracing Knowledge in Language Models Back to the Training Data
• Schwarzschild et al 2024 - Rethinking LLM Memorization through the Lens of Adversarial Compression

See Knowledge Editing and Model Editing and Unlearning.

• Salemi et al 2023 - LaMP: When Large Language Models Meet Personalization
• Li et al 2024 - Personal LLM Agents: Insights and Survey about the Capability, Efficiency and Security

• Personality
  • Jiang et al 2023 - PersonaLLM: Investigating the Ability of Large Language Models to Express Personality Traits
  • Pan & Zeng 2023 - Do LLMs Possess a Personality? Making the MBTI Test an Amazing Evaluation for Large Language Models
  • Huang et al 2023 - Who is ChatGPT? Benchmarking LLMs' Psychological Portrayal Using PsychoBench
  • Mao et al 2023 - Editing Personality for Large Language Models
  • Suhr et al 2023 - Challenging the Validity of Personality Tests for Large Language Models
• Vocabulary Overuse
  • Kobak et al 2024 - Delving into ChatGPT usage in academic writing through excess vocabulary

See also Fake News Detection.

• Rodriguez - Cross-Domain Detection of GPT-2-Generated Technical Text
• Mitchell et al 2023 - DetectGPT: Zero-Shot Machine-Generated Text Detection using Probability Curvature
• Mireshghallah et al 2023 - Smaller Language Models are Better Black-box Machine-Generated Text Detectors

• Mu et al 2023 - Can LLMs Follow Simple Rules?

• Wu et al 2025 - AxBench: Steering LLMs? Even Simple Baselines Outperform Sparse Autoencoders
• Wu et al 2025 - Improved Representation Steering for Language Models

• Evaluation, see Evaluation with Large Language Models
• Creating Data or replacement for crowdsourcing, see Data Augmentation (Synthetic Data Generation)

See Copyright Issues.

• Henderson et al 2023 - Foundation Models and Fair Use
• Karamolegkou et al 2023 - Copyright Violations and Large Language Models

See also Analysis of In-Context-Learning and Language Model - Origin of Capabilities.

See also Scaling Laws - Emergent Abilities

• Lu et al 2023 - Are Emergent Abilities in Large Language Models just In-Context Learning?

See paper list here. See also Model Compression.

• Overviews
  • Xu & McAuley et al 2022 - A Survey on Model Compression and Acceleration for Pretrained Language Models
  • Wan et al 2023 - Efficient Large Language Models: A Survey Updated continuously. See paper list here

• Howell et al 2023 - The Economic Trade-offs of Large Language Models: A Case Study

• Meta 2024 - Large Concept Models: Language Modeling in a Sentence Representation Space
  • Ahmad & Goel 2025 - The Future of AI: Exploring the Potential of Large Concept Models

• Overviews
  • Chen et al 2025 - Exploring Consciousness in LLMs: A Systematic Survey of Theories, Implementations, and Frontier Risks

Historical papers that may or may not be applicable today.

• Teh 2003 - A Bayesian Interpretation of Interpolated Kneser-Ney
• Melis et al 2017 - On the State of the Art of Evaluation in Neural Language Models Shows that LSTMS, when properly tuned, outperform other models (as of 2017, so before the Transformer)
• Khandelwal et al 2018 - Sharp Nearby, Fuzzy Far Away: How Neural Language Models Use Context

• Standard benchmark datasets
  • Wikitext 103
  • Penn Treebank
  • PG-19 Uses books from before 1919. Good for long sequences.
• Large datasets
  • Bookcorpus, also reproduced in the Pile, see here. reproduction original dataset
  • Common Crawl
  • WebText and OpenWebText:
    • WebText: Introduced in GPT-2 (paper).
    • OpenWebText: Various implementations here and here and here (on HuggingFace here). Used in MegatronLM here.
    • OpenWebText2 Open re-implementation, widely used. Use this one. On HuggingFace here.
  • Colossal Clean Crawled Corpus (C4): paper AI2 reimplementation atHuggingFace
  • The Pile Diverse set of data for building language models. Individual components, see readme. Older individual components Paper: The Pile: An 800GB Dataset of Diverse Text for Language Modeling Pile check tool
    • The pile has been removed due to this letter, see here and here
  • BigScienceCorpus: 2022 - The BigScience Corpus: A 1.6TB Composite Multilingual Dataset
  • RedPajama-Data An Open Source Recipe to Reproduce LLaMA training dataset
• Code datasets
  • The Stack Used in StarCoder. Has two membership test websites: Am I in the stack and DataPortraits
• Small datasets
  • BabyLM Challenge
  • Tinystories: paper dataset
  • Minipile: paper Huggingface

• Training and/or Inference Frameworks for LLMs
  • For an overview, see table 4 and section 3.7 of Liu et al 2024 - Understanding LLMs: A Comprehensive Overview from Training to Inference
• (Historical) n-gram LM toolkits
  • The best, highly optimized toolkit: KenLM
  • Industry standard toolkit with many options: SRILM
  • NLTK also implements n-gram LMs
• Deep learning toolkits
  • GPT Neo An open-source implementation of model & data parallel GPT3-like models using the mesh-tensorflow library.
  • NVidia's Megatron-LM Used for example, by BLOOM
• Online demos
  • AI2's Jurassic language model Jurassic-1
  • GPT-3: web interface is free after signing up
  • Hugging Face Fill Mask demo: Fill-Mask Demo Text Generation Demo

• Autonomous Language Agents
• BERT and Friends
• ChatGPT
• Dense Document Retrieval with LLMs
• Hallucination and Factivity
• Instruction-Tuning
• Large Reasoning Models (such as OpenAI o1 or DeepSeek R1)
• Mixture of Expert Models
• Perplexity
• Pretraining
• Prompting
• Scaling Laws
• Supertasks