In recent years, model quantization for face recognition has gained prominence. Traditionally, compressing models involved vast datasets like the 5.8 million-image MS1M dataset as well as extensive training times, raising the question of whether such data enormity is essential. This paper addresses this by introducing an efficiency-driven approach, fine-tuning the model with just up to 14,000 images, 440 times smaller than MS1M. We demonstrate that effective quantization is achievable with a smaller dataset, presenting a new paradigm. Moreover, we incorporate an evaluation-based metric loss and achieve an outstanding 96.15% accuracy on the IJB-C dataset, establishing a new state-of-the-art compressed model training for face recognition. The subsequent analysis delves into potential applications, emphasizing the transformative power of this approach. This paper advances model quantization by highlighting the efficiency and optimal results with small data and training time.
Post-training quantization (PTQ) has emerged as a promising technique to reduce the cost of large language models (LLMs). Specifically, PTQ can effectively mitigate memory consumption and reduce computational overhead in LLMs. To meet the requirements of both high efficiency and performance across diverse scenarios, a comprehensive evaluation of quantized LLMs is essential to guide the selection of quantization methods. This paper presents a thorough evaluation of these factors by evaluating the effect of PTQ on Weight, Activation, and KV Cache on 11 model families, including OPT, LLaMA2, Falcon, Bloomz, Mistral, ChatGLM, Vicuna, LongChat, StableLM, Gemma, and Mamba, with parameters ranging from 125M to 180B. The evaluation encompasses five types of tasks: basic NLP, emergent ability, trustworthiness, dialogue, and long-context tasks. Moreover, we also evaluate the state-of-the-art (SOTA) quantization methods to demonstrate their applicability. Based on the extensive experiments, we systematically summarize the effect of quantization, provide recommendations to apply quantization techniques, and point out future directions. The code can be found in https://github.com/thu-nics/qllm-eval.
Key-Value (KV) Caching has become an essential technique for accelerating the inference speed and throughput of generative Large Language Models~(LLMs). However, the memory footprint of the KV cache poses a critical bottleneck in LLM deployment as the cache size grows with batch size and sequence length, often surpassing even the size of the model itself. Although recent methods were proposed to select and evict unimportant KV pairs from the cache to reduce memory consumption, the potential ramifications of eviction on the generative process are yet to be thoroughly examined. In this paper, we examine the detrimental impact of cache eviction and observe that unforeseen risks arise as the information contained in the KV pairs is exhaustively discarded, resulting in safety breaches, hallucinations, and context loss. Surprisingly, we find that preserving even a small amount of information contained in the evicted KV pairs via reduced precision quantization substantially recovers the incurred degradation. On the other hand, we observe that the important KV pairs must be kept at a relatively higher precision to safeguard the generation quality. Motivated by these observations, we propose \textit{Mixed-precision KV cache}~(MiKV), a reliable cache compression method that simultaneously preserves the context details by retaining the evicted KV pairs in low-precision and ensure generation quality by keeping the important KV pairs in high-precision. Experiments on diverse benchmarks and LLM backbones show that our proposed method offers a state-of-the-art trade-off between compression ratio and performance, compared to other baselines.
Large language models (LLMs) have demonstrated state-of-the-art performance across various tasks. However, the latency of inference and the large GPU memory consumption of LLMs restrict their deployment performance. Recently, there have been some efficient attempts to quantize LLMs, yet inference with large batch size or long sequence still has the issue of being compute-bound. Fine-grained quantization methods have showcased their proficiency in achieving low-bit quantization for LLMs, while requiring FP16 data type for linear layer computations, which is time-consuming when dealing with large batch size or long sequence. In this paper, we introduce a method called FlattenQuant, which significantly reduces the maximum value of the tensor by flattening the large channels in the tensor, to achieve low bit per-tensor quantization with minimal accuracy loss. Our experiments show that FlattenQuant can directly use 4 bits to achieve 48.29% of the linear layer calculation in LLMs, with the remaining layers using 8 bits. The 4-bit matrix multiplication introduced in the FlattenQuant method can effectively address the compute-bound caused by large matrix calculation. Our work achieves up to 2$\times$ speedup and 2.3$\times$ memory reduction for LLMs with negligible loss in accuracy.
Dithering is a technique that can improve human perception of low-resolution data by reducing quantization artifacts. In this work we formalize and analytically justify two metrics for quantization artifact prominence, using them to design a novel dithering method for distortion-controlled data compression. We present theoretical entropy calculations for this dither and experimentally validate its performance on a low-rate image compression task. The result is a drastic improvement in the perceptual quality of quantized images with a lower recompression entropy than any state-of-the-art dither technique, achieving 45 points lower PIQUE at the same rate or 40% lower rate at the same PIQUE. The proposed dither is an adaptable tool applicable for use in any lossy compression system, permitting precise control of rate-distortion characteristics for both compression and recompression.
Fluid data completion is a research problem with high potential benefit for both experimental and computational fluid dynamics. An effective fluid data completion method reduces the required number of sensors in a fluid dynamics experiment, and allows a coarser and more adaptive mesh for a Computational Fluid Dynamics (CFD) simulation. However, the ill-posed nature of the fluid data completion problem makes it prohibitively difficult to obtain a theoretical solution and presents high numerical uncertainty and instability for a data-driven approach (e.g., a neural network model). To address these challenges, we leverage recent advancements in computer vision, employing the vector quantization technique to map both complete and incomplete fluid data spaces onto discrete-valued lower-dimensional representations via a two-stage learning procedure. We demonstrated the effectiveness of our approach on Kolmogorov flow data (Reynolds number: 1000) occluded by masks of different size and arrangement. Experimental results show that our proposed model consistently outperforms benchmark models under different occlusion settings in terms of point-wise reconstruction accuracy as well as turbulent energy spectrum and vorticity distribution.
It is critical to deploy complicated neural network models on hardware with limited resources. This paper proposes a novel model quantization method, named the Low-Cost Proxy-Based Adaptive Mixed-Precision Model Quantization (LCPAQ), which contains three key modules. The hardware-aware module is designed by considering the hardware limitations, while an adaptive mixed-precision quantization module is developed to evaluate the quantization sensitivity by using the Hessian matrix and Pareto frontier techniques. Integer linear programming is used to fine-tune the quantization across different layers. Then the low-cost proxy neural architecture search module efficiently explores the ideal quantization hyperparameters. Experiments on the ImageNet demonstrate that the proposed LCPAQ achieves comparable or superior quantization accuracy to existing mixed-precision models. Notably, LCPAQ achieves 1/200 of the search time compared with existing methods, which provides a shortcut in practical quantization use for resource-limited devices.
The emerging conditional coding-based neural video codec (NVC) shows superiority over commonly-used residual coding-based codec and the latest NVC already claims to outperform the best traditional codec. However, there still exist critical problems blocking the practicality of NVC. In this paper, we propose a powerful conditional coding-based NVC that solves two critical problems via feature modulation. The first is how to support a wide quality range in a single model. Previous NVC with this capability only supports about 3.8 dB PSNR range on average. To tackle this limitation, we modulate the latent feature of the current frame via the learnable quantization scaler. During the training, we specially design the uniform quantization parameter sampling mechanism to improve the harmonization of encoding and quantization. This results in a better learning of the quantization scaler and helps our NVC support about 11.4 dB PSNR range. The second is how to make NVC still work under a long prediction chain. We expose that the previous SOTA NVC has an obvious quality degradation problem when using a large intra-period setting. To this end, we propose modulating the temporal feature with a periodically refreshing mechanism to boost the quality. %Besides solving the above two problems, we also design a single model that can support both RGB and YUV colorspaces. Notably, under single intra-frame setting, our codec can achieve 29.7\% bitrate saving over previous SOTA NVC with 16\% MACs reduction. Our codec serves as a notable landmark in the journey of NVC evolution. The codes are at https://github.com/microsoft/DCVC.
Increasing the number of parameters in large language models (LLMs) usually improves performance in downstream tasks but raises compute and memory costs, making deployment difficult in resource-limited settings. Quantization techniques, which reduce the bits needed for model weights or activations with minimal performance loss, have become popular due to the rise of LLMs. However, most quantization studies use pre-trained LLMs, and the impact of quantization on instruction-tuned LLMs and the relationship between perplexity and benchmark performance of quantized LLMs are not well understood. Evaluation of quantized LLMs is often limited to language modeling and a few classification tasks, leaving their performance on other benchmarks unclear. To address these gaps, we propose a structured evaluation framework consisting of three critical dimensions: (1) knowledge \& capacity, (2) alignment, and (3) efficiency, and conduct extensive experiments across ten diverse benchmarks. Our experimental results indicate that LLMs with 4-bit quantization can retain performance comparable to their non-quantized counterparts, and perplexity can serve as a proxy metric for quantized LLMs on most benchmarks. Furthermore, quantized LLMs with larger parameter scales can outperform smaller LLMs. Despite the memory savings achieved through quantization, it can also slow down the inference speed of LLMs. Consequently, substantial engineering efforts and hardware support are imperative to achieve a balanced optimization of decoding speed and memory consumption in the context of quantized LLMs.
Language-conditioned robot behavior plays a vital role in executing complex tasks by associating human commands or instructions with perception and actions. The ability to compose long-horizon tasks based on unconstrained language instructions necessitates the acquisition of a diverse set of general-purpose skills. However, acquiring inherent primitive skills in a coupled and long-horizon environment without external rewards or human supervision presents significant challenges. In this paper, we evaluate the relationship between skills and language instructions from a mathematical perspective, employing two forms of mutual information within the framework of language-conditioned policy learning. To maximize the mutual information between language and skills in an unsupervised manner, we propose an end-to-end imitation learning approach known as Language Conditioned Skill Discovery (LCSD). Specifically, we utilize vector quantization to learn discrete latent skills and leverage skill sequences of trajectories to reconstruct high-level semantic instructions. Through extensive experiments on language-conditioned robotic navigation and manipulation tasks, encompassing BabyAI, LORel, and CALVIN, we demonstrate the superiority of our method over prior works. Our approach exhibits enhanced generalization capabilities towards unseen tasks, improved skill interpretability, and notably higher rates of task completion success.
Large Language Models (LLMs) are reshaping the research landscape in artificial intelligence, particularly as model parameters scale up significantly, unlocking remarkable capabilities across various domains. Nevertheless, the scalability of model parameters faces constraints due to limitations in GPU memory and computational speed. To address these constraints, various weight compression methods have emerged, such as Pruning and Quantization. Given the low-rank nature of weight matrices in language models, the reduction of weights through matrix decomposition undoubtedly holds significant potential and promise. In this paper, drawing upon the intrinsic structure of LLMs, we propose a novel approach termed Data-free Joint Rank-k Approximation for compressing the parameter matrices. Significantly, our method is characterized by without necessitating additional involvement of any corpus, while simultaneously preserving orthogonality in conjunction with pruning and quantization methods. We achieve a model pruning of 80% parameters while retaining 93.43% of the original performance without any calibration data. Additionally, we explore the fundamental properties of the weight matrix of LLMs undergone Rank-k Approximation and conduct comprehensive experiments to elucidate our hypothesis.
Representing the Neural Radiance Field (NeRF) with the explicit voxel grid (EVG) is a promising direction for improving NeRFs. However, the EVG representation is not efficient for storage and transmission because of the terrific memory cost. Current methods for compressing EVG mainly inherit the methods designed for neural network compression, such as pruning and quantization, which do not take full advantage of the spatial correlation of voxels. Inspired by prosperous digital image compression techniques, this paper proposes SPC-NeRF, a novel framework applying spatial predictive coding in EVG compression. The proposed framework can remove spatial redundancy efficiently for better compression performance.Moreover, we model the bitrate and design a novel form of the loss function, where we can jointly optimize compression ratio and distortion to achieve higher coding efficiency. Extensive experiments demonstrate that our method can achieve 32% bit saving compared to the state-of-the-art method VQRF on multiple representative test datasets, with comparable training time.
The field of efficient Large Language Model (LLM) inference is rapidly evolving, presenting a unique blend of opportunities and challenges. Although the field has expanded and is vibrant, there hasn't been a concise framework that analyzes the various methods of LLM Inference to provide a clear understanding of this domain. Our survey stands out from traditional literature reviews by not only summarizing the current state of research but also by introducing a framework based on roofline model for systematic analysis of LLM inference techniques. This framework identifies the bottlenecks when deploying LLMs on hardware devices and provides a clear understanding of practical problems, such as why LLMs are memory-bound, how much memory and computation they need, and how to choose the right hardware. We systematically collate the latest advancements in efficient LLM inference, covering crucial areas such as model compression (e.g., Knowledge Distillation and Quantization), algorithm improvements (e.g., Early Exit and Mixture-of-Expert), and both hardware and system-level enhancements. Our survey stands out by analyzing these methods with roofline model, helping us understand their impact on memory access and computation. This distinctive approach not only showcases the current research landscape but also delivers valuable insights for practical implementation, positioning our work as an indispensable resource for researchers new to the field as well as for those seeking to deepen their understanding of efficient LLM deployment. The analyze tool, LLM-Viewer, is open-sourced.
Speech quality estimation has recently undergone a paradigm shift from human-hearing expert designs to machine-learning models. However, current models rely mainly on supervised learning, which is time-consuming and expensive for label collection. To solve this problem, we propose VQScore, a self-supervised metric for evaluating speech based on the quantization error of a vector-quantized-variational autoencoder (VQ-VAE). The training of VQ-VAE relies on clean speech; hence, large quantization errors can be expected when the speech is distorted. To further improve correlation with real quality scores, domain knowledge of speech processing is incorporated into the model design. We found that the vector quantization mechanism could also be used for self-supervised speech enhancement (SE) model training. To improve the robustness of the encoder for SE, a novel self-distillation mechanism combined with adversarial training is introduced. In summary, the proposed speech quality estimation method and enhancement models require only clean speech for training without any label requirements. Experimental results show that the proposed VQScore and enhancement model are competitive with supervised baselines. The code will be released after publication.
Model reparameterization is a widely accepted technique for improving inference speed without compromising performance. However, current Post-training Quantization (PTQ) methods often lead to significant accuracy degradation when applied to reparameterized models. This is primarily caused by channel-specific and sample-specific outliers, which appear only at specific samples and channels and impact on the selection of quantization parameters. To address this issue, we propose RepAPQ, a novel framework that preserves the accuracy of quantized reparameterization models. Different from previous frameworks using Mean Squared Error (MSE) as a measurement, we utilize Mean Absolute Error (MAE) to mitigate the influence of outliers on quantization parameters. Our framework comprises two main components: Quantization Protecting Reparameterization and Across-block Calibration. For effective calibration, Quantization Protecting Reparameterization combines multiple branches into a single convolution with an affine layer. During training, the affine layer accelerates convergence and amplifies the output of the convolution to better accommodate samples with outliers. Additionally, Across-block Calibration leverages the measurement of stage output as supervision to address the gradient problem introduced by MAE and enhance the interlayer correlation with quantization parameters. Comprehensive experiments demonstrate the effectiveness of RepAPQ across various models and tasks. Our framework outperforms previous methods by approximately 1\% for 8-bit PTQ and 2\% for 6-bit PTQ, showcasing its superior performance. The code is available at \url{https://github.com/ilur98/DLMC-QUANT}.
In this work we show that the size versus accuracy trade-off of neural network quantization can be significantly improved by increasing the quantization dimensionality. We propose the GPTVQ method, a new fast method for post-training vector quantization (VQ) that scales well to Large Language Models (LLMs). Our method interleaves quantization of one or more columns with updates to the remaining unquantized weights, using information from the Hessian of the per-layer output reconstruction MSE. Quantization codebooks are initialized using an efficient data-aware version of the EM algorithm. The codebooks are then updated, and further compressed by using integer quantization and SVD-based compression. GPTVQ establishes a new state-of-the art in the size vs accuracy trade-offs on a wide range of LLMs such as Llama-v2 and Mistral. Furthermore, our method is efficient: on a single H100 it takes between 3 and 11 hours to process a Llamav2-70B model, depending on quantization setting. Lastly, with on-device timings for VQ decompression on a mobile CPU we show that VQ leads to improved latency compared to using a 4-bit integer format.
We propose a straightforward approach called Distillation Contrastive Decoding (DCD) to enhance the reasoning capabilities of Large Language Models (LLMs) during inference. In contrast to previous approaches that relied on smaller amateur models or analysis of hidden state differences, DCD employs Contrastive Chain-of-thought Prompting and advanced distillation techniques, including Dropout and Quantization. This approach effectively addresses the limitations of Contrastive Decoding (CD), which typically requires both an expert and an amateur model, thus increasing computational resource demands. By integrating contrastive prompts with distillation, DCD obviates the need for an amateur model and reduces memory usage. Our evaluations demonstrate that DCD significantly enhances LLM performance across a range of reasoning benchmarks, surpassing both CD and existing methods in the GSM8K and StrategyQA datasets.
Large Language Models (LLMs) have greatly advanced the natural language processing paradigm. However, the high computational load and huge model sizes pose a grand challenge for deployment on edge devices. To this end, we propose APTQ (Attention-aware Post-Training Mixed-Precision Quantization) for LLMs, which considers not only the second-order information of each layer's weights, but also, for the first time, the nonlinear effect of attention outputs on the entire model. We leverage the Hessian trace as a sensitivity metric for mixed-precision quantization, ensuring an informed precision reduction that retains model performance. Experiments show APTQ surpasses previous quantization methods, achieving an average of 4 bit width a 5.22 perplexity nearly equivalent to full precision in the C4 dataset. In addition, APTQ attains state-of-the-art zero-shot accuracy of 68.24\% and 70.48\% at an average bitwidth of 3.8 in LLaMa-7B and LLaMa-13B, respectively, demonstrating its effectiveness to produce high-quality quantized LLMs.
Just as the arrow of time structures physics, the arrow of inference organizes cognition, directing the flow of information in perception, action, and memory. The Context-Content Uncertainty Principle (CCUP) formalizes this asymmetry, between high-entropy context and low-entropy content, and frames inference as a cycle that aligns the two through selective, bidirectional interaction. Cycle formation resolves the Information Bottleneck (IB) in Optimal Transport (OT) by coordinating bottom-up contextual disambiguation with top-down content reconstruction, a process neurobiologically mirrored in the cyclical interplay between dorsal (context) and ventral (content) streams. Local inference cycles extend into memory chains that simulate goals, support counterfactual reasoning, and scaffold internal model refinement across time. By operating on delta-seeded goal manifolds, each level of the hierarchy circumvents the curse of dimensionality through structured diffusion guided by priors and context. This mechanism generalizes across timescales, from perception-action loops to the sleep-wake cycle-and scales socially through language, which externalizes inference by transmitting latent content across minds. Thus, CCUP provides a unifying framework for understanding cognition as cycle-consistent inference, anchoring both individual thought and collective intelligence.
In human activity recognition (HAR), the availability of substantial ground truth is necessary for training efficient models. However, acquiring ground pressure data through physical sensors itself can be cost-prohibitive, time-consuming. To address this critical need, we introduce Text-to-Pressure (T2P), a framework designed to generate extensive ground pressure sequences from textual descriptions of human activities using deep learning techniques. We show that the combination of vector quantization of sensor data along with simple text conditioned auto regressive strategy allows us to obtain high-quality generated pressure sequences from textual descriptions with the help of discrete latent correlation between text and pressure maps. We achieved comparable performance on the consistency between text and generated motion with an R squared value of 0.722, Masked R squared value of 0.892, and FID score of 1.83. Additionally, we trained a HAR model with the the synthesized data and evaluated it on pressure dynamics collected by a real pressure sensor which is on par with a model trained on only real data. Combining both real and synthesized training data increases the overall macro F1 score by 5.9 percent.