Reconstructing real-world 3D objects has numerous applications in computer vision, such as virtual reality, video games, and animations. Ideally, 3D reconstruction methods should generate high-fidelity results with 3D consistency in real-time. Traditional methods match pixels between images using photo-consistency constraints or learned features, while differentiable rendering methods like Neural Radiance Fields (NeRF) use differentiable volume rendering or surface-based representation to generate high-fidelity scenes. However, these methods require excessive runtime for rendering, making them impractical for daily applications. To address these challenges, we present $\textbf{EvaSurf}$, an $\textbf{E}$fficient $\textbf{V}$iew-$\textbf{A}$ware implicit textured $\textbf{Surf}$ace reconstruction method. In our method, we first employ an efficient surface-based model with a multi-view supervision module to ensure accurate mesh reconstruction. To enable high-fidelity rendering, we learn an implicit texture embedded with view-aware encoding to capture view-dependent information. Furthermore, with the explicit geometry and the implicit texture, we can employ a lightweight neural shader to reduce the expense of computation and further support real-time rendering on common mobile devices. Extensive experiments demonstrate that our method can reconstruct high-quality appearance and accurate mesh on both synthetic and real-world datasets. Moreover, our method can be trained in just 1-2 hours using a single GPU and run on mobile devices at over 40 FPS (Frames Per Second), with a final package required for rendering taking up only 40-50 MB.
Benchmarks are crucial in the development of machine learning algorithms, with available environments significantly influencing reinforcement learning (RL) research. Traditionally, RL environments run on the CPU, which limits their scalability with typical academic compute. However, recent advancements in JAX have enabled the wider use of hardware acceleration, enabling massively parallel RL training pipelines and environments. While this has been successfully applied to single-agent RL, it has not yet been widely adopted for multi-agent scenarios. In this paper, we present JaxMARL, the first open-source, Python-based library that combines GPU-enabled efficiency with support for a large number of commonly used MARL environments and popular baseline algorithms. Our experiments show that, in terms of wall clock time, our JAX-based training pipeline is around 14 times faster than existing approaches, and up to 12500x when multiple training runs are vectorized. This enables efficient and thorough evaluations, potentially alleviating the evaluation crisis in the field. We also introduce and benchmark SMAX, a JAX-based approximate reimplementation of the popular StarCraft Multi-Agent Challenge, which removes the need to run the StarCraft II game engine. This not only enables GPU acceleration, but also provides a more flexible MARL environment, unlocking the potential for self-play, meta-learning, and other future applications in MARL. The code is available at https://github.com/flairox/jaxmarl.
The illumination of improperly exposed photographs has been widely corrected using deep convolutional neural networks or Transformers. Despite with promising performance, these methods usually suffer from large parameter amounts and heavy computational FLOPs on high-resolution photographs. In this paper, we propose extremely light-weight (with only ~8K parameters) Multi-Scale Linear Transformation (MSLT) networks under the multi-layer perception architecture, which can process 4K-resolution sRGB images at 125 Frame-Per-Second (FPS) by a Titan RTX GPU. Specifically, the proposed MSLT networks first decompose an input image into high and low frequency layers by Laplacian pyramid techniques, and then sequentially correct different layers by pixel-adaptive linear transformation, which is implemented by efficient bilateral grid learning or 1x1 convolutions. Experiments on two benchmark datasets demonstrate the efficiency of our MSLTs against the state-of-the-arts on photo exposure correction. Extensive ablation studies validate the effectiveness of our contributions. The code is available at https://github.com/Zhou-Yijie/MSLTNet.
We propose \textbf{DMV3D}, a novel 3D generation approach that uses a transformer-based 3D large reconstruction model to denoise multi-view diffusion. Our reconstruction model incorporates a triplane NeRF representation and can denoise noisy multi-view images via NeRF reconstruction and rendering, achieving single-stage 3D generation in $\sim$30s on single A100 GPU. We train \textbf{DMV3D} on large-scale multi-view image datasets of highly diverse objects using only image reconstruction losses, without accessing 3D assets. We demonstrate state-of-the-art results for the single-image reconstruction problem where probabilistic modeling of unseen object parts is required for generating diverse reconstructions with sharp textures. We also show high-quality text-to-3D generation results outperforming previous 3D diffusion models. Our project website is at: https://justimyhxu.github.io/projects/dmv3d/ .
In this work, we propose FastCoT, a model-agnostic framework based on parallel decoding without any further training of an auxiliary model or modification to the LLM itself. FastCoT uses a size-varying context window whose size changes with position to conduct parallel decoding and auto-regressive decoding simultaneously, thus fully utilizing GPU computation resources. In FastCoT, the parallel decoding part provides the LLM with a quick glance of the future composed of approximate tokens, which could lead to faster answers compared to regular autoregressive decoding used by causal transformers. We also provide an implementation of parallel decoding within LLM, which supports KV-cache generation and batch processing. Through extensive experiments, we demonstrate that FastCoT saves inference time by nearly 20% with only a negligible performance drop compared to the regular approach. Additionally, we show that the context window size exhibits considerable robustness for different tasks.
Recently, large language models (LLMs) have drawn extensive attention from academia and the public, due to the advent of the ChatGPT. While LLMs show their astonishing ability in text generation for various tasks, privacy concerns limit their usage in real-life businesses. More specifically, either the user's inputs (the user sends the query to the model-hosting server) or the model (the user downloads the complete model) itself will be revealed during the usage. Vertical federated learning (VFL) is a promising solution to this kind of problem. It protects both the user's input and the knowledge of the model by splitting the model into a bottom part and a top part, which is maintained by the user and the model provider, respectively. However, in this paper, we demonstrate that in LLMs, VFL fails to protect the user input since it is simple and cheap to reconstruct the input from the intermediate embeddings. Experiments show that even with a commercial GPU, the input sentence can be reconstructed in only one second. We also discuss several possible solutions to enhance the privacy of vertical federated LLMs.
Decision trees are essential yet NP-complete to train, prompting the widespread use of heuristic methods such as CART, which suffers from sub-optimal performance due to its greedy nature. Recently, breakthroughs in finding optimal decision trees have emerged; however, these methods still face significant computational costs and struggle with continuous features in large-scale datasets and deep trees. To address these limitations, we introduce a moving-horizon differential evolution algorithm for classification trees with continuous features (MH-DEOCT). Our approach consists of a discrete tree decoding method that eliminates duplicated searches between adjacent samples, a GPU-accelerated implementation that significantly reduces running time, and a moving-horizon strategy that iteratively trains shallow subtrees at each node to balance the vision and optimizer capability. Comprehensive studies on 68 UCI datasets demonstrate that our approach outperforms the heuristic method CART on training and testing accuracy by an average of 3.44% and 1.71%, respectively. Moreover, these numerical studies empirically demonstrate that MH-DEOCT achieves near-optimal performance (only 0.38% and 0.06% worse than the global optimal method on training and testing, respectively), while it offers remarkable scalability for deep trees (e.g., depth=8) and large-scale datasets (e.g., ten million samples).
Large language models (LLMs) have proven their remarkable versatility in handling a comprehensive range of language-centric applications. To expand LLMs' capabilities to a broader spectrum of modal inputs, multimodal large language models (MLLMs) have attracted growing interest. This work delves into enabling LLMs to tackle more vision-language-related tasks, particularly image captioning, visual question answering (VQA,) and visual grounding. To this end, we implemented a three-stage training scheme: starting with lightweight alignment pretraining, then moderate-weight multitask hybrid training, and finally, LLM fine-tuning to improve instruction following capability. Throughout the training process, the requirements on GPU memory gradually increase. To effectively manage the number of visual embeddings passed to the LLM while preserving their positional information, we introduce a straightforward visual adapter module dubbed pool-adapter. Our experiments demonstrate that preserving the positional information of visual embeddings through the pool-adapter is particularly beneficial for tasks like visual grounding. We name our proposed approach InfMLLM and have evaluated it extensively on various benchmark datasets. Our results demonstrate that InfMLLM achieves either state-of-the-art (SOTA) performance or performance comparable to recent MLLMs. The code and model will be made open-source at: \url{https://github.com/mightyzau/InfMLLM}.
The integration of visual inputs with large language models (LLMs) has led to remarkable advancements in multi-modal capabilities, giving rise to visual large language models (VLLMs). However, effectively harnessing VLLMs for intricate visual perception tasks remains a challenge. In this paper, we present a novel end-to-end framework named PerceptionGPT, which efficiently and effectively equips the VLLMs with visual perception abilities by leveraging the representation power of LLMs' token embedding. Our proposed method treats the token embedding of the LLM as the carrier of spatial information, then leverage lightweight visual task encoders and decoders to perform visual perception tasks (e.g., detection, segmentation). Our approach significantly alleviates the training difficulty suffered by previous approaches that formulate the visual outputs as discrete tokens, and enables achieving superior performance with fewer trainable parameters, less training data and shorted training time. Moreover, as only one token embedding is required to decode the visual outputs, the resulting sequence length during inference is significantly reduced. Consequently, our approach enables accurate and flexible representations, seamless integration of visual perception tasks, and efficient handling of a multiple of visual outputs. We validate the effectiveness and efficiency of our approach through extensive experiments. The results demonstrate significant improvements over previous methods with much fewer trainable parameters and GPU hours, which facilitates future research in enabling LLMs with visual perception abilities.
While Connectionist Temporal Classification (CTC) models deliver state-of-the-art accuracy in automated speech recognition (ASR) pipelines, their performance has been limited by CPU-based beam search decoding. We introduce a GPU-accelerated Weighted Finite State Transducer (WFST) beam search decoder compatible with current CTC models. It increases pipeline throughput and decreases latency, supports streaming inference, and also supports advanced features like utterance-specific word boosting via on-the-fly composition. We provide pre-built DLPack-based python bindings for ease of use with Python-based machine learning frameworks at https://github.com/nvidia-riva/riva-asrlib-decoder. We evaluated our decoder for offline and online scenarios, demonstrating that it is the fastest beam search decoder for CTC models. In the offline scenario it achieves up to 7 times more throughput than the current state-of-the-art CPU decoder and in the online streaming scenario, it achieves nearly 8 times lower latency, with same or better word error rate.
Latent Consistency Models (LCMs) have achieved impressive performance in accelerating text-to-image generative tasks, producing high-quality images with minimal inference steps. LCMs are distilled from pre-trained latent diffusion models (LDMs), requiring only ~32 A100 GPU training hours. This report further extends LCMs' potential in two aspects: First, by applying LoRA distillation to Stable-Diffusion models including SD-V1.5, SSD-1B, and SDXL, we have expanded LCM's scope to larger models with significantly less memory consumption, achieving superior image generation quality. Second, we identify the LoRA parameters obtained through LCM distillation as a universal Stable-Diffusion acceleration module, named LCM-LoRA. LCM-LoRA can be directly plugged into various Stable-Diffusion fine-tuned models or LoRAs without training, thus representing a universally applicable accelerator for diverse image generation tasks. Compared with previous numerical PF-ODE solvers such as DDIM, DPM-Solver, LCM-LoRA can be viewed as a plug-in neural PF-ODE solver that possesses strong generalization abilities. Project page: https://github.com/luosiallen/latent-consistency-model.
We present Prompt Cache, an approach for accelerating inference for large language models (LLM) by reusing attention states across different LLM prompts. Many input prompts have overlapping text segments, such as system messages, prompt templates, and documents provided for context. Our key insight is that by precomputing and storing the attention states of these frequently occurring text segments on the inference server, we can efficiently reuse them when these segments appear in user prompts. Prompt Cache employs a schema to explicitly define such reusable text segments, called prompt modules. The schema ensures positional accuracy during attention state reuse and provides users with an interface to access cached states in their prompt. Using a prototype implementation, we evaluate Prompt Cache across several LLMs. We show that Prompt Cache significantly reduce latency in time-to-first-token, especially for longer prompts such as document-based question answering and recommendations. The improvements range from 8x for GPU-based inference to 60x for CPU-based inference, all while maintaining output accuracy and without the need for model parameter modifications.
Click-through rate (CTR) prediction is widely used in academia and industry. Most CTR tasks fall into a feature embedding \& feature interaction paradigm, where the accuracy of CTR prediction is mainly improved by designing practical feature interaction structures. However, recent studies have argued that the fixed feature embedding learned only through the embedding layer limits the performance of existing CTR models. Some works apply extra modules on top of the embedding layer to dynamically refine feature representations in different instances, making it effective and easy to integrate with existing CTR methods. Despite the promising results, there is a lack of a systematic review and summarization of this new promising direction on the CTR task. To fill this gap, we comprehensively summarize and define a new module, namely \textbf{feature refinement} (FR) module, that can be applied between feature embedding and interaction layers. We extract 14 FR modules from previous works, including instances where the FR module was proposed but not clearly defined or explained. We fully assess the effectiveness and compatibility of existing FR modules through comprehensive and extensive experiments with over 200 augmented models and over 4,000 runs for more than 15,000 GPU hours. The results offer insightful guidelines for researchers, and all benchmarking code and experimental results are open-sourced. In addition, we present a new architecture of assigning independent FR modules to separate sub-networks for parallel CTR models, as opposed to the conventional method of inserting a shared FR module on top of the embedding layer. Our approach is also supported by comprehensive experiments demonstrating its effectiveness.
The relentless advancement of artificial intelligence (AI) and machine learning (ML) applications necessitates the development of specialized hardware accelerators capable of handling the increasing complexity and computational demands. Traditional computing architectures, based on the von Neumann model, are being outstripped by the requirements of contemporary AI/ML algorithms, leading to a surge in the creation of accelerators like the Graphcore Intelligence Processing Unit (IPU), Sambanova Reconfigurable Dataflow Unit (RDU), and enhanced GPU platforms. These hardware accelerators are characterized by their innovative data-flow architectures and other design optimizations that promise to deliver superior performance and energy efficiency for AI/ML tasks. This research provides a preliminary evaluation and comparison of these commercial AI/ML accelerators, delving into their hardware and software design features to discern their strengths and unique capabilities. By conducting a series of benchmark evaluations on common DNN operators and other AI/ML workloads, we aim to illuminate the advantages of data-flow architectures over conventional processor designs and offer insights into the performance trade-offs of each platform. The findings from our study will serve as a valuable reference for the design and performance expectations of research prototypes, thereby facilitating the development of next-generation hardware accelerators tailored for the ever-evolving landscape of AI/ML applications. Through this analysis, we aspire to contribute to the broader understanding of current accelerator technologies and to provide guidance for future innovations in the field.
We present LongQLoRA, an efficient and effective method to extend context length of large language models with less training resources. LongQLoRA combines the advantages of Position Interpolation, QLoRA and Shift Short Attention of LongLoRA. With a single 32GB V100 GPU, LongQLoRA can extend the context length of LLaMA2 7B and 13B from 4096 to 8192 and even to 12k within 1000 finetuning steps. LongQLoRA achieves competitive perplexity performance on PG19 and Proof-pile datasets, our model outperforms LongLoRA and is very close to MPT-7B-8K within the evaluation context length of 8192. We collect and build 39k long instruction data to extend context length of Vicuna-13B from 4096 to 8192 and achieve good performance both in long and short context generation task. We also do some ablation experiments to study the effect of LoRA rank, finetuning steps and attention patterns in inference.The model weights, training data and code are avaliable at https://github.com/yangjianxin1/LongQLoRA.
Building on the cost-efficient pretraining advancements brought about by Crammed BERT, we enhance its performance and interpretability further by introducing a novel pretrained model Dependency Agreement Crammed BERT (DACBERT) and its two-stage pretraining framework - Dependency Agreement Pretraining. This framework, grounded by linguistic theories, seamlessly weaves syntax and semantic information into the pretraining process. The first stage employs four dedicated submodels to capture representative dependency agreements at the chunk level, effectively converting these agreements into embeddings. The second stage uses these refined embeddings, in tandem with conventional BERT embeddings, to guide the pretraining of the rest of the model. Evaluated on the GLUE benchmark, our DACBERT demonstrates notable improvement across various tasks, surpassing Crammed BERT by 3.13% in the RTE task and by 2.26% in the MRPC task. Furthermore, our method boosts the average GLUE score by 0.83%, underscoring its significant potential. The pretraining process can be efficiently executed on a single GPU within a 24-hour cycle, necessitating no supplementary computational resources or extending the pretraining duration compared with the Crammed BERT. Extensive studies further illuminate our approach's instrumental role in bolstering the interpretability of pretrained language models for natural language understanding tasks.
Large language models (LLMs) have shown impressive success in various applications. However, these models are often not well aligned with human intents, which calls for additional treatments on them; that is, the alignment problem. To make LLMs better follow user instructions, existing alignment methods primarily focus on further training them. However, the extra training of LLMs is usually expensive in terms of GPU computing; even worse, some LLMs are not accessible for user-demanded training, such as GPTs. In this work, we take a different perspective -- Black-Box Prompt Optimization (BPO) -- to perform alignments. The idea is to optimize user prompts to suit LLMs' input understanding, so as to best realize users' intents without updating LLMs' parameters. BPO leverages human preferences to optimize prompts, thus making it superior to LLM (e.g., ChatGPT) as a prompt engineer. Moreover, BPO is model-agnostic, and the empirical results demonstrate that the BPO-aligned ChatGPT yields a 22% increase in the win rate against its original version and 10% for GPT-4. Notably, the BPO-aligned LLMs can outperform the same models aligned by PPO and DPO, and it also brings additional performance gains when combining BPO with PPO or DPO. Code and datasets are released at https://github.com/thu-coai/BPO.
This paper identifies and addresses the problems with naively combining (reinforcement) learning-based controllers and state estimators for robotic in-hand manipulation. Specifically, we tackle the challenging task of purely tactile, goal-conditioned, dextrous in-hand reorientation with the hand pointing downwards. Due to the limited sensing available, many control strategies that are feasible in simulation when having full knowledge of the object's state do not allow for accurate state estimation. Hence, separately training the controller and the estimator and combining the two at test time leads to poor performance. We solve this problem by coupling the control policy to the state estimator already during training in simulation. This approach leads to more robust state estimation and overall higher performance on the task while maintaining an interpretability advantage over end-to-end policy learning. With our GPU-accelerated implementation, learning from scratch takes a median training time of only 6.5 hours on a single, low-cost GPU. In simulation experiments with the DLR-Hand II and for four significantly different object shapes, we provide an in-depth analysis of the performance of our approach. We demonstrate the successful sim2real transfer by rotating the four objects to all 24 orientations in the $\pi/2$ discretization of SO(3), which has never been achieved for such a diverse set of shapes. Finally, our method can reorient a cube consecutively to nine goals (median), which was beyond the reach of previous methods in this challenging setting.
Large Language Models (LLMs) have seen great advance in both academia and industry, and their popularity results in numerous open-source frameworks and techniques in accelerating LLM pre-training, fine-tuning, and inference. Training and deploying LLMs are expensive as it requires considerable computing resources and memory, hence many efficient approaches have been developed for improving system pipelines as well as operators. However, the runtime performance can vary significantly across hardware and software stacks, which makes it difficult to choose the best configuration. In this work, we aim to benchmark the performance from both macro and micro perspectives. First, we benchmark the end-to-end performance of pre-training, fine-tuning, and serving LLMs in different sizes , i.e., 7, 13, and 70 billion parameters (7B, 13B, and 70B) on three 8-GPU platforms with and without individual optimization techniques, including ZeRO, quantization, recomputation, FlashAttention. Then, we dive deeper to provide a detailed runtime analysis of the sub-modules, including computing and communication operators in LLMs. For end users, our benchmark and findings help better understand different optimization techniques, training and inference frameworks, together with hardware platforms in choosing configurations for deploying LLMs. For researchers, our in-depth module-wise analyses discover potential opportunities for future work to further optimize the runtime performance of LLMs.
Weight-sharing is ubiquitous in deep learning. Motivated by this, we propose a "weight-sharing regularization" penalty on the weights $w \in \mathbb{R}^d$ of a neural network, defined as $\mathcal{R}(w) = \frac{1}{d - 1}\sum_{i > j}^d |w_i - w_j|$. We study the proximal mapping of $\mathcal{R}$ and provide an intuitive interpretation of it in terms of a physical system of interacting particles. We also parallelize existing algorithms for $\operatorname{prox}_\mathcal{R}$ (to run on GPU) and find that one of them is fast in practice but slow ($O(d)$) for worst-case inputs. Using the physical interpretation, we design a novel parallel algorithm which runs in $O(\log^3 d)$ when sufficient processors are available, thus guaranteeing fast training. Our experiments reveal that weight-sharing regularization enables fully connected networks to learn convolution-like filters even when pixels have been shuffled while convolutional neural networks fail in this setting. Our code is available on github.