👋 Welcome to Lil’Log

Special thanks to John Schulman for a lot of super valuable feedback and direct edits on this post. Test time compute (Graves et al. 2016, Ling, et al. 2017, Cobbe et al. 2021) and Chain-of-thought (CoT) (Wei et al. 2022, Nye et al. 2021), have led to significant improvements in model performance, while raising many research questions. This post aims to review recent developments in how to effectively use test-time compute (i.e. “thinking time”) and why it helps. ...

Reward hacking occurs when a reinforcement learning (RL) agent exploits flaws or ambiguities in the reward function to achieve high rewards, without genuinely learning or completing the intended task. Reward hacking exists because RL environments are often imperfect, and it is fundamentally challenging to accurately specify a reward function. With the rise of language models generalizing to a broad spectrum of tasks and RLHF becomes a de facto method for alignment training, reward hacking in RL training of language models has become a critical practical challenge. Instances where the model learns to modify unit tests to pass coding tasks, or where responses contain biases that mimic a user’s preference, are pretty concerning and are likely one of the major blockers for real-world deployment of more autonomous use cases of AI models. ...

Hallucination in large language models usually refers to the model generating unfaithful, fabricated, inconsistent, or nonsensical content. As a term, hallucination has been somewhat generalized to cases when the model makes mistakes. Here, I would like to narrow down the problem of hallucination to cases where the model output is fabricated and not grounded by either the provided context or world knowledge. There are two types of hallucination: In-context hallucination: The model output should be consistent with the source content in context. Extrinsic hallucination: The model output should be grounded by the pre-training dataset. However, given the size of the pre-training dataset, it is too expensive to retrieve and identify conflicts per generation. If we consider the pre-training data corpus as a proxy for world knowledge, we essentially try to ensure the model output is factual and verifiable by external world knowledge. Equally importantly, when the model does not know about a fact, it should say so. This post focuses on extrinsic hallucination. To avoid hallucination, LLMs need to be (1) factual and (2) acknowledge not knowing the answer when applicable. ...

Diffusion models have demonstrated strong results on image synthesis in past years. Now the research community has started working on a harder task—using it for video generation. The task itself is a superset of the image case, since an image is a video of 1 frame, and it is much more challenging because: It has extra requirements on temporal consistency across frames in time, which naturally demands more world knowledge to be encoded into the model. In comparison to text or images, it is more difficult to collect large amounts of high-quality, high-dimensional video data, let along text-video pairs. 🥑 Required Pre-read: Please make sure you have read the previous blog on “What are Diffusion Models?” for image generation before continue here. ...

[Special thank you to Ian Kivlichan for many useful pointers (E.g. the 100+ year old Nature paper “Vox populi”) and nice feedback. 🙏 ] High-quality data is the fuel for modern data deep learning model training. Most of the task-specific labeled data comes from human annotation, such as classification task or RLHF labeling (which can be constructed as classification format) for LLM alignment training. Lots of ML techniques in the post can help with data quality, but fundamentally human data collection involves attention to details and careful execution. The community knows the value of high quality data, but somehow we have this subtle impression that “Everyone wants to do the model work, not the data work” (Sambasivan et al. 2021). ...

The use of large language models in the real world has strongly accelerated by the launch of ChatGPT. We (including my team at OpenAI, shoutout to them) have invested a lot of effort to build default safe behavior into the model during the alignment process (e.g. via RLHF). However, adversarial attacks or jailbreak prompts could potentially trigger the model to output something undesired. A large body of ground work on adversarial attacks is on images, and differently it operates in the continuous, high-dimensional space. Attacks for discrete data like text have been considered to be a lot more challenging, due to lack of direct gradient signals. My past post on Controllable Text Generation is quite relevant to this topic, as attacking LLMs is essentially to control the model to output a certain type of (unsafe) content. ...

Building agents with LLM (large language model) as its core controller is a cool concept. Several proof-of-concepts demos, such as AutoGPT, GPT-Engineer and BabyAGI, serve as inspiring examples. The potentiality of LLM extends beyond generating well-written copies, stories, essays and programs; it can be framed as a powerful general problem solver. Agent System Overview In a LLM-powered autonomous agent system, LLM functions as the agent’s brain, complemented by several key components: Planning Subgoal and decomposition: The agent breaks down large tasks into smaller, manageable subgoals, enabling efficient handling of complex tasks. Reflection and refinement: The agent can do self-criticism and self-reflection over past actions, learn from mistakes and refine them for future steps, thereby improving the quality of final results. Memory Short-term memory: I would consider all the in-context learning (See Prompt Engineering) as utilizing short-term memory of the model to learn. Long-term memory: This provides the agent with the capability to retain and recall (infinite) information over extended periods, often by leveraging an external vector store and fast retrieval. Tool use The agent learns to call external APIs for extra information that is missing from the model weights (often hard to change after pre-training), including current information, code execution capability, access to proprietary information sources and more. Overview of a LLM-powered autonomous agent system. Component One: Planning A complicated task usually involves many steps. An agent needs to know what they are and plan ahead. ...

Prompt Engineering, also known as In-Context Prompting, refers to methods for how to communicate with LLM to steer its behavior for desired outcomes without updating the model weights. It is an empirical science and the effect of prompt engineering methods can vary a lot among models, thus requiring heavy experimentation and heuristics. This post only focuses on prompt engineering for autoregressive language models, so nothing with Cloze tests, image generation or multimodality models. At its core, the goal of prompt engineering is about alignment and model steerability. Check my previous post on controllable text generation. ...

Many new Transformer architecture improvements have been proposed since my last post on “The Transformer Family” about three years ago. Here I did a big refactoring and enrichment of that 2020 post — restructure the hierarchy of sections and improve many sections with more recent papers. Version 2.0 is a superset of the old version, about twice the length. Notations Symbol Meaning $d$ The model size / hidden state dimension / positional encoding size. $h$ The number of heads in multi-head attention layer. $L$ The segment length of input sequence. $N$ The total number of attention layers in the model; not considering MoE. $\mathbf{X} \in \mathbb{R}^{L \times d}$ The input sequence where each element has been mapped into an embedding vector of shape $d$, same as the model size. $\mathbf{W}^k \in \mathbb{R}^{d \times d_k}$ The key weight matrix. $\mathbf{W}^q \in \mathbb{R}^{d \times d_k}$ The query weight matrix. $\mathbf{W}^v \in \mathbb{R}^{d \times d_v}$ The value weight matrix. Often we have $d_k = d_v = d$. $\mathbf{W}^k_i, \mathbf{W}^q_i \in \mathbb{R}^{d \times d_k/h}; \mathbf{W}^v_i \in \mathbb{R}^{d \times d_v/h}$ The weight matrices per head. $\mathbf{W}^o \in \mathbb{R}^{d_v \times d}$ The output weight matrix. $\mathbf{Q} = \mathbf{X}\mathbf{W}^q \in \mathbb{R}^{L \times d_k}$ The query embedding inputs. $\mathbf{K} = \mathbf{X}\mathbf{W}^k \in \mathbb{R}^{L \times d_k}$ The key embedding inputs. $\mathbf{V} = \mathbf{X}\mathbf{W}^v \in \mathbb{R}^{L \times d_v}$ The value embedding inputs. $\mathbf{q}_i, \mathbf{k}_i \in \mathbb{R}^{d_k}, \mathbf{v}_i \in \mathbb{R}^{d_v}$ Row vectors in query, key, value matrices, $\mathbf{Q}$, $\mathbf{K}$ and $\mathbf{V}$. $S_i$ A collection of key positions for the $i$-th query $\mathbf{q}_i$ to attend to. $\mathbf{A} \in \mathbb{R}^{L \times L}$ The self-attention matrix between a input sequence of lenght $L$ and itself. $\mathbf{A} = \text{softmax}(\mathbf{Q}\mathbf{K}^\top / \sqrt{d_k})$. $a_{ij} \in \mathbf{A}$ The scalar attention score between query $\mathbf{q}_i$ and key $\mathbf{k}_j$. $\mathbf{P} \in \mathbb{R}^{L \times d}$ position encoding matrix, where the $i$-th row $\mathbf{p}_i$ is the positional encoding for input $\mathbf{x}_i$. Transformer Basics The Transformer (which will be referred to as “vanilla Transformer” to distinguish it from other enhanced versions; Vaswani, et al., 2017) model has an encoder-decoder architecture, as commonly used in many NMT models. Later simplified Transformer was shown to achieve great performance in language modeling tasks, like in encoder-only BERT or decoder-only GPT. ...

[Updated on 2023-01-24: add a small section on Distillation.] Large transformer models are mainstream nowadays, creating SoTA results for a variety of tasks. They are powerful but very expensive to train and use. The extremely high inference cost, in both time and memory, is a big bottleneck for adopting a powerful transformer for solving real-world tasks at scale. Why is it hard to run inference for large transformer models? Besides the increasing size of SoTA models, there are two main factors contributing to the inference challenge (Pope et al. 2022): ...