Advanced Transformer Architectures

• Lead team: team-6
• Blog team: team-6

Efficiency

In this session, our readings cover:

Required Readings:

Advancing Transformer Architecture in Long-Context Large Language Models: A Comprehensive Survey

• https://arxiv.org/abs/2311.12351
• Transformer-based Large Language Models (LLMs) have been applied in diverse areas such as knowledge bases, human interfaces, and dynamic agents, and marking a stride towards achieving Artificial General Intelligence (AGI). However, current LLMs are predominantly pretrained on short text snippets, which compromises their effectiveness in processing the long-context prompts that are frequently encountered in practical scenarios. This article offers a comprehensive survey of the recent advancement in Transformer-based LLM architectures aimed at enhancing the long-context capabilities of LLMs throughout the entire model lifecycle, from pre-training through to inference. We first delineate and analyze the problems of handling long-context input and output with the current Transformer-based models. We then provide a taxonomy and the landscape of upgrades on Transformer architecture to solve these problems. Afterwards, we provide an investigation on wildly used evaluation necessities tailored for long-context LLMs, including datasets, metrics, and baseline models, as well as optimization toolkits such as libraries, frameworks, and compilers to boost the efficacy of LLMs across different stages in runtime. Finally, we discuss the challenges and potential avenues for future research. A curated repository of relevant literature, continuously updated, is available at this https URL.

FlashAttention: Fast and Memory-Efficient Exact Attention with IO-Awareness

• Tri Dao, Daniel Y. Fu, Stefano Ermon, Atri Rudra, Christopher Ré
• Paper: https://arxiv.org/abs/2205.14135

• Transformers are slow and memory-hungry on long sequences, since the time and memory complexity of self-attention are quadratic in sequence length. Approximate attention methods have attempted to address this problem by trading off model quality to reduce the compute complexity, but often do not achieve wall-clock speedup. We argue that a missing principle is making attention algorithms IO-aware – accounting for reads and writes between levels of GPU memory. We propose FlashAttention, an IO-aware exact attention algorithm that uses tiling to reduce the number of memory reads/writes between GPU high bandwidth memory (HBM) and GPU on-chip SRAM. We analyze the IO complexity of FlashAttention, showing that it requires fewer HBM accesses than standard attention, and is optimal for a range of SRAM sizes. We also extend FlashAttention to block-sparse attention, yielding an approximate attention algorithm that is faster than any existing approximate attention method. FlashAttention trains Transformers faster than existing baselines: 15% end-to-end wall-clock speedup on BERT-large (seq. length 512) compared to the MLPerf 1.1 training speed record, 3$\times$ speedup on GPT-2 (seq. length 1K), and 2.4$\times$ speedup on long-range arena (seq. length 1K-4K). FlashAttention and block-sparse FlashAttention enable longer context in Transformers, yielding higher quality models (0.7 better perplexity on GPT-2 and 6.4 points of lift on long-document classification) and entirely new capabilities: the first Transformers to achieve better-than-chance performance on the Path-X challenge (seq. length 16K, 61.4% accuracy) and Path-256 (seq. length 64K, 63.1% accuracy).

• Related: blogpost FlashAttention — Techniques for Efficient Inference of LLMs (III/IV)

JAMBA

• Introducing Jamba: AI21’s Groundbreaking SSM-Transformer Model Debuting the first production-grade Mamba-based model delivering best-in-class quality and performance.
• March 28, 2024
• https://www.ai21.com/blog/announcing-jamba
• We are thrilled to announce Jamba, the world’s first production-grade Mamba based model. By enhancing Mamba Structured State Space model (SSM) technology with elements of the traditional Transformer architecture, Jamba compensates for the inherent limitations of a pure SSM model. Offering a 256K context window, it is already demonstrating remarkable gains in throughput and efficiency—just the beginning of what can be possible with this innovative hybrid architecture. Notably, Jamba outperforms or matches other state-of-the-art models in its size class on a wide range of benchmarks.

More readings:

Mamba: Linear-Time Sequence Modeling with Selective State Spaces

• Albert Gu, Tri Dao
• https://arxiv.org/abs/2312.00752
• Foundation models, now powering most of the exciting applications in deep learning, are almost universally based on the Transformer architecture and its core attention module. Many subquadratic-time architectures such as linear attention, gated convolution and recurrent models, and structured state space models (SSMs) have been developed to address Transformers’ computational inefficiency on long sequences, but they have not performed as well as attention on important modalities such as language. We identify that a key weakness of such models is their inability to perform content-based reasoning, and make several improvements. First, simply letting the SSM parameters be functions of the input addresses their weakness with discrete modalities, allowing the model to selectively propagate or forget information along the sequence length dimension depending on the current token. Second, even though this change prevents the use of efficient convolutions, we design a hardware-aware parallel algorithm in recurrent mode. We integrate these selective SSMs into a simplified end-to-end neural network architecture without attention or even MLP blocks (Mamba). Mamba enjoys fast inference (5× higher throughput than Transformers) and linear scaling in sequence length, and its performance improves on real data up to million-length sequences. As a general sequence model backbone, Mamba achieves state-of-the-art performance across several modalities such as language, audio, and genomics. On language modeling, our Mamba-3B model outperforms Transformers of the same size and matches Transformers twice its size, both in pretraining and downstream evaluation.

Efficient Memory Management for Large Language Model Serving with PagedAttention

• Woosuk Kwon, Zhuohan Li, Siyuan Zhuang, Ying Sheng, Lianmin Zheng, Cody Hao Yu, Joseph E. Gonzalez, Hao Zhang, Ion Stoica
• High throughput serving of large language models (LLMs) requires batching sufficiently many requests at a time. However, existing systems struggle because the key-value cache (KV cache) memory for each request is huge and grows and shrinks dynamically. When managed inefficiently, this memory can be significantly wasted by fragmentation and redundant duplication, limiting the batch size. To address this problem, we propose PagedAttention, an attention algorithm inspired by the classical virtual memory and paging techniques in operating systems. On top of it, we build vLLM, an LLM serving system that achieves (1) near-zero waste in KV cache memory and (2) flexible sharing of KV cache within and across requests to further reduce memory usage. Our evaluations show that vLLM improves the throughput of popular LLMs by 2-4× with the same level of latency compared to the state-of-the-art systems, such as FasterTransformer and Orca. The improvement is more pronounced with longer sequences, larger models, and more complex decoding algorithms. vLLM’s source code is publicly available at this https URL

Attention Mechanisms in Computer Vision: A Survey

• Meng-Hao Guo, Tian-Xing Xu, Jiang-Jiang Liu, Zheng-Ning Liu, Peng-Tao Jiang, Tai-Jiang Mu, Song-Hai Zhang, Ralph R. Martin, Ming-Ming Cheng, Shi-Min Hu
• https://arxiv.org/abs/2111.07624
• Humans can naturally and effectively find salient regions in complex scenes. Motivated by this observation, attention mechanisms were introduced into computer vision with the aim of imitating this aspect of the human visual system. Such an attention mechanism can be regarded as a dynamic weight adjustment process based on features of the input image. Attention mechanisms have achieved great success in many visual tasks, including image classification, object detection, semantic segmentation, video understanding, image generation, 3D vision, multi-modal tasks and self-supervised learning. In this survey, we provide a comprehensive review of various attention mechanisms in computer vision and categorize them according to approach, such as channel attention, spatial attention, temporal attention and branch attention; a related repository this https URL is dedicated to collecting related work. We also suggest future directions for attention mechanism research.