Scaling Up to Large Language Models

How to Model NLP Tasks?

• Define the problem clearly: What specific NLP task are you trying to solve? (e.g., sentiment analysis, question answering, machine translation, code generation).

• Analyze the data for challenges:

  • Multilingual? Does it involve multiple languages?

  • Dialectal? Does it contain variations within a language?

  • Domain-specific? Is the language from a particular field (e.g., medical, legal)?

  • Conversational? Is it informal, with slang and colloquialisms?

  • Chronological? Does the order of the data matter (e.g., time-series data)?

  • Core: What is your input/output? What machine learning algorithms are applicable?

• Data Size and Quality:

  • How much data is available? Is it labeled?

  • Is it clean and well-formatted? If not, what preprocessing techniques are applicable?

• Select an appropriate model architecture (e.g., sequence-to-sequence, classification/regression).

• Select appropriate approaches given data (e.g., fine-tuning with labeled data / continual pre-training for unlabeled data).

• Identify suitable metrics for both implicit (e.g., training/validation loss, perplexity) and explicit evaluation.

  • e.g., AUROC, Precision, Recall, F1, Spearman/Pearson’s Correlation, BLEU, ROUGE, BERTScore, TER.

  • To be explained in Lecture 9 on Evaluating model performance comprehensively.

• Example: Categorizing customer feedback from an online retail platform.

  • Customer feedback is a written statement and a review on a scale of 1-5 stars; categories can be positive, negative, or neutral.

NLP Research Background

• Modeling NLP tasks, following NLP research, and publishing your research.

The NLP Research Community

• Research papers on ACL Anthology – 93k+ papers!

• Search for relevant areas by:

  • Topic (to learn which methods are applied).

  • Method (to learn which topics it is applied to).

• Great way to learn about current topics and ongoing research and a good archive of NLP papers.

• ACL Anthology data is also used by research.

• Task Example: Given research papers, authors, paper metadata (including links to codebase) and presentation video, what tasks are possible?

Research Papers

• Most NLP papers are published as 9-page conference papers (long papers) with 3 reviewers (double blind review process).

• Other types of papers:

  • Short papers (5 pages).

  • Findings papers.

  • Systems demo track.

  • Industry track.

  • Journal papers (up to 12 pages, or unlimited).

  • Survey Papers [e.g., IJCAI Survey track].

• Venues: ACL, EMNLP, NAACL, EACL, AACL, COLING, LREC, AAAI, IJCAI, NeurIPS, ICLR, CVPR, ECCV, Interspeech, IWSLT, and various other conferences and workshops.

NLP Research Tracks

• Machine Learning for NLP

• Interpretability & Analysis of Models for NLP

• Resources & Evaluation

• Ethics & NLP

• Phonology, Morphology & Word Segmentation

• Syntax: Tagging, Chunking & Parsing

• Semantics: Lexical

• Semantics: Sentence-level Semantics, Textual Inference & Other areas

• Linguistic Theories, Cognitive Modeling & Psycholinguistics

• Information Extraction

• Information Retrieval & Text Mining

• Question Answering

• Summarization

• Machine Translation & Multilinguality

• Speech & Multimodality

• Discourse & Pragmatics

• Sentiment Analysis, Stylistic Analysis, & Argument Mining

• Dialogue & Interactive Systems

• Language Grounding to Vision, Robotics & Beyond

• Computational Social Science & Cultural Analytics

• NLP Applications

• Often Special themes are introduced.

Follow NLP Research

• arXiv papers.

• Twitter accounts.

• NLP researchers with active accounts (grad students, profs, industry folks).

• Official conference accounts.

• “NLP Highlights” podcast.

• “NLP News” newsletter, ‘NLP People’ (https://nlppeople.com/jobs/).

• Huggingface, Microsoft Research, Google DeepMind, Meta AI (FAIR), Amazon, AI4Bharat, eBay Research, NVIDIA Research, academic labs (StanfordNLP, MBZUAI, CFILT@IITB, SurreyNLP) and many startups (Cohere, Anthropic).

How to Begin NLP Research

• Find a relevant problem and identify recent research.

• Find good or popular tools:

  • Identify papers, ask around, search the web.

• Trying to identify the best tool for your job:

  • Produces appropriate, sufficiently detailed output?

  • Accurate? (on the measure you care about).

  • Robust? (accurate on your data, not just theirs).

  • Fast?

  • Easy and flexible to use? Nice file formats, command line options, visualization?

  • Trainable for new data and languages? How slow is training?

  • Open-source and easy to extend?

Shared Tasks

• Dijkstra would participate in a shared task ☺

• Competitions organized by research community which invite participation.

• Components:

  • Training data (optional).

  • Development data (optional).

  • Test data, for evaluating the final participating systems.

  • Evaluation metric(s) (how well does the system perform).

  • Additional Training/Synthetic data (optional).

  • A prize (optional; with clear rules on what data can be used).

• Easy to write system description paper.

Large Language Models (LLMs)

Natural Language Processing (NLP)

• 'Understand’ Human Language

• A key goal of NLP research is to ensure that machines understand human language.

Analyse Human Language

• Textual analytics, extraction, and retrieval to analyze the information present in human language.

Generate Human Language

• Generation of understandable human language to interface with people.

Different Types of Attention

• Diagram illustrating encoder-decoder architecture with self-attention and masked self-attention.

Pre-trained Transformer Flavors

Encoders

• Examples: BERT, RoBERTa, SciBERT.

• Captures bidirectional context.

Decoders

• Examples: GPT-2, GPT-3, LaMDA

• Also known as: causal or auto-regressive language model

• Natural if the goal is generation, but cannot condition on future words

Encoder-Decoders

• Examples: BART, T5, Meena

• Conditional generation based on an encoded input

Autoregressive Decoders

• A non-auto-regressive model:

  • Inputs and outputs are different.

  • Use case: Assigning labels for each word (e.g., part-of-speech tagging).

• A “causal” or an auto-regressive model:

  • Each output is the next input in the sequence.

  • Use case: Generating tokens (e.g., language modeling).

The GPT Family

• GPT (2018): 117 million parameters - Improving Language Understanding by Generative Pre-Training.

• GPT-2 (2019): 1.5 billion parameters - Language Models are Unsupervised Multitask Learners.

• GPT-3 (2020): 175 billion parameters - Language Models are Few-Shot Learners; NeurIPS 2020 best paper.

GPT Anatomy

• An autoregressive model that predicts the next token given tokens so far (either predicted or given as part of input).

Masked Self-Attention (Decoders)

• Masks future tokens by interfering in the self-attention calculation, blocking information from tokens to the right of the position being calculated.

• Prevents positions from 'peeking' at tokens to its right.

Decoder-only (block)

• As it processes each subword, it masks the “future” words and conditions on (i.e., attends to) the previous words.

Decoder Stack

• Diagram illustrating token embeddings and decoder layers.

Pre-training Decoder Stack

• Language models generate text by predicting the next token given a prompt.

• Formally, given a prompt (input) $x = (x<em>1, x</em>2, …, x<em>n)$, where $x</em>i$ is a token at the i-th index.

• A causal language model (autoregressive decoder) estimates the probability of the next token $x<em>{n+1}$ as $P(x</em>{n+1} \mid x<em>1, x</em>2, …, x_n; \theta)$, where $\theta$ is the parameter of the language model.

• Models are trained using causal language modeling loss:

  • $L = - \sum<em>{i=1}^{T} \log P(x</em>i \mid x<em>1, …, x</em>{i-1}; \theta)$, where T is the total number of tokens in a sequence.

• During inference (generation), given a prompt x, the model generates new tokens by sampling

• Model Inference based on the probability distribution of the vocabulary given the preceding tokens.

How do such models decode?

• Pick an output token based on its probability (sample).

Sampling Techniques

• Using various sampling techniques!

• For instance, top-k and top-p are most commonly used!

Top-k sampling

• Probabilities for all possible next tokens are generated and sorted in descending order.

• Only the top-k tokens are considered for the next step.

• A token is then randomly sampled from this reduced set based on their probabilities.

  • Lower k: more focused and deterministic output, as the model has fewer options.

  • Higher k: introduces more randomness and creativity, as a wider range of potential tokens is considered.

• In essence, “choosing the next word from the top-k suggestions”.

Top-p (nucleus) sampling

• Selects the smallest set of most probable tokens whose cumulative probability exceeds a threshold p.

• The cumulative probability is calculated as you go down the list.

• Once the cumulative probability reaches or exceeds p, the remaining lower-probability tokens are discarded.

• The next token is then randomly sampled from this nucleus of tokens.

  • Lower p: more focused and coherent output, as only the most likely tokens remain.

  • Higher p: allows for more diverse and surprising output, as a larger set of tokens becomes eligible for sampling.

• In essence, "choosing the next word from a set of suggestions that together make up a certain level of confidence (prob.)“.

Combining top-k and top-p sampling

• Apply top-k to narrow down the candidate set, then apply top-p to this reduced set.

The first GPT (or GPT-1)

• Illustrations of GPT-1 architecture for various tasks (Classification, Entailment, Similarity, Multiple Choice).

Emergent abilities of large language models: GPT (2018)

• GPT (117M parameters; Radford et al., 2018)

• Transformer decoder with 12 layers.

• Trained on BooksCorpus: over 7000 unique books (4.6GB text).

• Showed that language modeling at scale can be an effective pretraining technique for downstream tasks like natural language inference.

• Example: [START] The man is in the doorway [SEP] The person is near the door [EXTRACT]

GPT-2

• GPT-2 is quite identical to GPT-1, but:

  • Layer norm moved to the input of each sub-block.

  • Vocabulary extended to 50,257 tokens and context size increased from 512 to 1024.

  • Trained on 8 million docs from the web (Common Crawl), minus Wikipedia.

  • https://huggingface.co/gpt2

• Lab Assignment: Train your own GPT-2 from scratch on the “Tiny-Shakespeare” dataset, utilizing some pre-trained GPT-2 weights.

GPT-2 Variants

• GPT-2 Small, Medium, Large, Extra Large with varying model dimensionality and decoder layers.

Emergent abilities of large language models: GPT-2 (2019)

• GPT-2 (1.5B parameters; Radford et al., 2019)

• Same architecture as GPT, just bigger (117M -> 1.5B)

• Trained on much more data: 4GB -> 40GB of internet text data (WebText)

• Scrape links posted on Reddit w/ at least 3 upvotes (rough proxy of human quality)

Zero-shot Capabilities

• One key emergent ability in GPT-2 is zero-shot: the ability to do many tasks without any examples, and no gradient updates, by simply:

  • Specifying the right sequence prediction problem (e.g. question answering):
    Passage: Tom Brady… Q: Where was Tom Brady born? A: …

  • Comparing probabilities of sequences (e.g., Winograd Schema Challenge):
    The cat couldn’t fit into the hat because it was too big. Does it = the cat or the hat? ≡ Is P(…because the cat was too big) >= P(…because the hat was too big)? [Radford et al., 2019]

GPT-3

• More layers & parameters

• Bigger dataset

• Longer training

• Larger embedding/hidden dimension

• Larger context window

• GPT-3 (175B parameters; Brown et al., 2020)

• Another increase in size (1.5B -> 175B) and data (40GB -> over 600GB)

Bigger is better?

• BERT-Base model has 12 transformer blocks, 12 attention heads, 110M parameters

• BERT-Large model has 24 transformer blocks, 16 attention heads, 340M parameters

• GPT-2 is trained on 40GB of text data (8M webpages)! – 1.5B parameters

• GPT-3 is an even bigger version of GPT-2, but since then, none of the GPT-series models have been open-sourced – 175B parameters

• GPT-4 is estimated to have roughly 1.8 trillion parameters [estimated number, unconfirmed by OpenAI].

• GPT-4o Mini could be as small as 8 billion parameters.

GPT-4o onwards

• GPT-4 ‘o’-series of models can emulate the thinking process.

• Enabled by reinforcement learning (RL), and by adding more synthetic data to the RL pipeline (after pre-training; called post-training phase)

• Models can simulate an inner monologue, where they articulate their reasoning process step-by-step; helps in verifying the accuracy of their conclusion

Learning – Embedded within Pre-training Objective

• Examples:

  • University of Surrey is located in _, United Kingdom. [Trivia]

  • I put ___ fork down on the table. [syntax]

  • The woman walked across the street, checking for traffic over ___ shoulder. [coreference]

  • I went to the ocean to see the fish, turtles, seals, and _. [lexical semantics/topic]

  • Overall, the value I got from the two hours watching it was the sum total of the popcorn and the drink. The movie was ___. [sentiment]

  • Iroh went into the kitchen to make some tea. Standing next to Iroh, Zuko pondered his destiny. Zuko left the . [some reasoning – this is harder]

  • I was thinking about the sequence that goes 1, 1, 2, 3, 5, 8, 13, 21, ____ [some basic arithmetic; they may not learn the Fibonnaci sequence]

Prompting a language model

• Pretraining Knowledge

  • Language models learn various linguistic patterns during pretraining, such as syntax, coreference, and semantics.

• Zero/Few-Shot Capabilities

  • GPT-2 onwards LLMs can perform tasks with and without limited examples, by predicting sequences or comparing probabilities.

• Prompt Engineering

  • Crafting prompts that reformulate tasks to resemble those solved during pretraining.

Emergent few-shot learning

• Specify a task by simply prepending examples of the task before your example

• Also called in-context learning, to stress that no gradient updates are performed when learning a new task (there is a separate literature on few-shot learning with gradient updates).

• Example prompt for translation quality scoring.

Emergent few-shot learning: Observed improvement in performance as an increasing number of examples are provided

• Illustrated by leaderboard data.

Few-shot learning is an emergent property of model scale

• Cycle letters: pleap -> apple

• Random insertion: a.p!p/l!e -> apple

• Reversed words: elppa -> apple

New methods of "prompting" LMs

• Comparison of Zero/few-shot prompting vs. Traditional fine-tuning.

New methods of “prompting” LMs

• Lab Assignment: Get familiar with zero-shot, one-shot and few-shot scenarios by prompt engineering for Dialogue Summarization.

Limits of prompting for harder tasks?

• Some tasks seem too hard for even large LMs to learn through prompting alone, especially tasks involving richer, multi-step reasoning.

• Solution: change the prompt!

Chain-of-thought prompting

• Example illustrating the difference between standard prompting and chain-of-thought prompting for math word problems.

Chain-of-thought prompting, also emergent with model scale

• Graph comparing performance of standard prompting vs. chain-of-thought prompting for different model scales.

Chain-of-thought prompting: Is reasoning necessary?

• Do we even need examples of reasoning? Can we just ask the model to reason through things?

Zero-shot Chain-of-thought prompting

• Example:
  Q: A juggler can juggle 16 balls. Half of the balls are golf balls, and half of the golf balls are blue. How many blue golf balls are there?\ A: Let’s think step by step.
  There are 16 balls in total. Half of the balls are golf balls. That means there are 8 golf balls. Half of the golf balls are blue. That means there are 4 blue golf balls.

Zero-shot Chain-of-thought prompting: Results show greatly outperforms zero-shot

Zero-shot Chain-of-thought prompting: Different Trigger Prompts

• Varying accuracy based on the specific prompt used to trigger chain-of-thought reasoning.

Summary

• Variants to BERT are available.

• NLP tasks can be clubbed together, depending on the role/need in the pipeline.

• Autoregressive Decoders are Transformer variants with stacked decoders.

• Impressive task-level generation performance seen with emergent properties, in both zero-/few-shot scenarios, including CoT.

• Modeling for an NLP task may very well be a matter of creating/engineering a detailed prompt, but ensuring accurate outputs from LLMs still requires supervised training data.

• Next Lecture: LLM Prompting, Tuning & Task Evaluation in NLP.

Acknowledgements

• Part of slides used from J. Eisner, J. Hockenmaier and Vivek Srikumar.

• Images and GIF from Jay Alammar’s blogs.