A Simple Linear Patch Revives Layer-Pruned Large Language Models

Layer pruning has become a popular technique for compressing large language models (LLMs) due to its simplicity. However, existing layer pruning methods often suffer from significant performance drops. We identify that this degradation stems from the mismatch of activation magnitudes across layers and tokens at the pruning interface. To address this, we propose LinearPatch, a simple plug-and-play technique to revive the layer-pruned LLMs. The proposed method adopts Hadamard transformation to suppress massive outliers in particular tokens, and channel-wise scaling to align the activation magnitudes. These operations can be fused into a single matrix, which functions as a patch to bridge the pruning interface with negligible inference overhead. LinearPatch retains up to 94.15% performance of the original model when pruning 5 layers of LLaMA-3-8B on the question answering benchmark, surpassing existing state-of-the-art methods by 4%. In addition, the patch matrix can be further optimized with memory efficient offline knowledge distillation. With only 5K samples, the retained performance of LinearPatch can be further boosted to 95.16% within 30 minutes on a single computing card.

Faster and Better LLMs via Latency-Aware Test-Time Scaling

Test-Time Scaling (TTS) has proven effective in improving the performance of Large Language Models (LLMs) during inference. However, existing research has overlooked the efficiency of TTS from a latency-sensitive perspective. Through a latency-aware evaluation of representative TTS methods, we demonstrate that a compute-optimal TTS does not always result in the lowest latency in scenarios where latency is critical. To address this gap and achieve latency-optimal TTS, we propose two key approaches by optimizing the concurrency configurations: (1) branch-wise parallelism, which leverages multiple concurrent inference branches, and (2) sequence-wise parallelism, enabled by speculative decoding. By integrating these two approaches and allocating computational resources properly to each, our latency-optimal TTS enables a 32B model to reach 82.3% accuracy on MATH-500 within 1 minute and a smaller 3B model to achieve 72.4% within 10 seconds. Our work emphasizes the importance of latency-aware TTS and demonstrates its ability to deliver both speed and accuracy in latency-sensitive scenarios.

FreqKV: Frequency Domain Key-Value Compression for Efficient Context Window Extension

Extending the context window in large language models (LLMs) is essential for applications involving long-form content generation. However, the linear increase in key-value (KV) cache memory requirements and the quadratic complexity of self-attention with respect to sequence length present significant challenges during fine-tuning and inference. Existing methods suffer from performance degradation when extending to longer contexts. In this work, we introduce a novel context extension method that optimizes both fine-tuning and inference efficiency. Our method exploits a key observation: in the frequency domain, the energy distribution of the KV cache is primarily concentrated in low-frequency components. By filtering out the high-frequency components, the KV cache can be effectively compressed with minimal information loss. Building on this insight, we propose an efficient compression technique, FreqKV, that iteratively compresses the increasing KV cache to a fixed size in the frequency domain, applicable to both fine-tuning and inference. FreqKV introduces no additional parameters or architectural modifications. With minimal fine-tuning, LLMs can learn to leverage the limited cache that is compressed in the frequency domain and extend the context window efficiently. Experiments on various long context language modeling and understanding tasks demonstrate the efficiency and efficacy of the proposed method.

Quantization Hurts Reasoning? An Empirical Study on Quantized Reasoning Models

Recent advancements in reasoning language models have demonstrated remarkable performance in complex tasks, but their extended chain-of-thought reasoning process increases inference overhead. While quantization has been widely adopted to reduce the inference cost of large language models, its impact on reasoning models remains understudied. In this study, we conduct the first systematic study on quantized reasoning models, evaluating the open-sourced DeepSeek-R1-Distilled Qwen and LLaMA families ranging from 1.5B to 70B parameters, and QwQ-32B. Our investigation covers weight, KV cache, and activation quantization using state-of-the-art algorithms at varying bit-widths, with extensive evaluation across mathematical (AIME, MATH-500), scientific (GPQA), and programming (LiveCodeBench) reasoning benchmarks. Our findings reveal that while lossless quantization can be achieved with W8A8 or W4A16 quantization, lower bit-widths introduce significant accuracy risks. We further identify model size, model origin, and task difficulty as critical determinants of performance. Contrary to expectations, quantized models do not exhibit increased output lengths. In addition, strategically scaling the model sizes or reasoning steps can effectively enhance the performance. All quantized models and codes will be open-sourced in https://github.com/ruikangliu/Quantized-Reasoning-Models.

WeightedKV: Attention Scores Weighted Key-Value Cache Merging for Large Language Models

Large Language Models (LLMs) use key-value (KV) cache to reduce redundant computation in autoregressive generation. However, the KV cache size increases linearly during generation, leading to excessive memory usage, especially for long texts. Most KV cache compression methods evict the unimportant KV pairs to maintain a fixed cache size, which leads to the permanent loss of tokens during generation. However, singular value decomposition shows that \textit{values} do not exhibit a strong low-rank property as \textit{keys} do, suggesting that information is distributed more evenly across \textit{values}, in contrast to its more redundant distribution within \textit{keys}. Therefore, methods that evict both \textit{keys} and \textit{values} risk losing crucial information and compromise context integrity, ultimately degrading the output quality. To address this problem, we propose WeightedKV, a novel, training-free approach that discards the \textit{keys} of less important tokens, while merging their \textit{values} into neighboring tokens via a convex combination weighted by their average attention scores. In this way, the retained \textit{keys} serve as anchors that guide the generation process, while the merged \textit{values} provide a rich contextual backdrop. We assess our method on four widely used language modeling datasets, demonstrating superior performance compared to all baseline methods, particularly with a lower budget ratio.

TreeKV: Smooth Key-Value Cache Compression with Tree Structures

Efficient key-value (KV) cache compression is critical for scaling transformer-based Large Language Models (LLMs) in long sequences and resource-limited settings. Existing methods evict tokens based on their positions or importance scores, but position-based strategies can miss crucial information outside predefined regions, while those relying on global importance scores resulting in strong regional biases, limiting the KV cache's overall context retention and potentially impairing the performance of LLMs on complex tasks. Our wavelet analysis reveals that as tokens approach the end of sequence, their contributions to generation gradually increase and tends to diverge more from neighboring tokens, indicating a smooth transition with increasing complexity and variability from distant to nearby context. Motivated by this observation, we propose TreeKV, an intuitive, training-free method that employs a tree structure for smooth cache compression. TreeKV maintains a fixed cache size, allowing LLMs to deliver high-quality output even in long text scenarios. Unlike most compression methods, TreeKV is applicable to both the generation and prefilling stages. It consistently surpasses all baseline models in language modeling tasks on PG19 and OpenWebText2, allowing LLMs trained with short context window to generalize to longer window with a 16x cache reduction. On the Longbench benchmark, TreeKV achieves the best performance with only 6\% of the budget at optimal efficiency.

FlatQuant: Flatness Matters for LLM Quantization

Recently, quantization has been widely used for the compression and acceleration of large language models~(LLMs). Due to the outliers in LLMs, it is crucial to flatten weights and activations to minimize quantization error with the equally spaced quantization points. Prior research explores various pre-quantization transformations to suppress outliers, such as per-channel scaling and Hadamard transformation. However, we observe that these transformed weights and activations can still remain steep and outspread. In this paper, we propose FlatQuant (Fast and Learnable Affine Transformation), a new post-training quantization approach to enhance flatness of weights and activations. Our approach identifies optimal affine transformations tailored to each linear layer, calibrated in hours via a lightweight objective. To reduce runtime overhead, we apply Kronecker decomposition to the transformation matrices, and fuse all operations in FlatQuant into a single kernel. Extensive experiments show that FlatQuant sets up a new state-of-the-art quantization benchmark. For instance, it achieves less than $\textbf{1}\%$ accuracy drop for W4A4 quantization on the LLaMA-3-70B model, surpassing SpinQuant by $\textbf{7.5}\%$. For inference latency, FlatQuant reduces the slowdown induced by pre-quantization transformation from 0.26x of QuaRot to merely $\textbf{0.07x}$, bringing up to $\textbf{2.3x}$ speedup for prefill and $\textbf{1.7x}$ speedup for decoding, respectively. Code is available at: \url{https://github.com/ruikangliu/FlatQuant}.

EMOVA: Empowering Language Models to See, Hear and Speak with Vivid Emotions

GPT-4o, an omni-modal model that enables vocal conversations with diverse emotions and tones, marks a milestone for omni-modal foundation models. However, empowering Large Language Models to perceive and generate images, texts, and speeches end-to-end with publicly available data remains challenging in the open-source community. Existing vision-language models rely on external tools for the speech processing, while speech-language models still suffer from limited or even without vision-understanding abilities. To address this gap, we propose EMOVA (EMotionally Omni-present Voice Assistant), to enable Large Language Models with end-to-end speech capabilities while maintaining the leading vision-language performance. With a semantic-acoustic disentangled speech tokenizer, we notice surprisingly that omni-modal alignment can further enhance vision-language and speech abilities compared with the corresponding bi-modal aligned counterparts. Moreover, a lightweight style module is proposed for flexible speech style controls (e.g., emotions and pitches). For the first time, EMOVA achieves state-of-the-art performance on both the vision-language and speech benchmarks, and meanwhile, supporting omni-modal spoken dialogue with vivid emotions.

S2D: Sorted Speculative Decoding For More Efficient Deployment of Nested Large Language Models

Deployment of autoregressive large language models (LLMs) is costly, and as these models increase in size, the associated costs will become even more considerable. Consequently, different methods have been proposed to accelerate the token generation process and reduce costs. Speculative decoding (SD) is among the most promising approaches to speed up the LLM decoding process by verifying multiple tokens in parallel and using an auxiliary smaller draft model to generate the possible tokens. In SD, usually, one draft model is used to serve a specific target model; however, in practice, LLMs are diverse, and we might need to deal with many target models or more than one target model simultaneously. In this scenario, it is not clear which draft model should be used for which target model, and searching among different draft models or training customized draft models can further increase deployment costs. In this paper, we first introduce a novel multi-target scenario for the deployment of draft models for faster inference. Then, we present a novel, more efficient sorted speculative decoding mechanism that outperforms regular baselines in multi-target settings. We evaluated our method on Spec-Bench in different settings, including base models such as Vicuna 7B, 13B, and LLama Chat 70B. Our results suggest that our draft models perform better than baselines for multiple target models at the same time.

Visually Guided Generative Text-Layout Pre-training for Document Intelligence

Prior study shows that pre-training techniques can boost the performance of visual document understanding (VDU), which typically requires models to gain abilities to perceive and reason both document texts and layouts (e.g., locations of texts and table-cells). To this end, we propose visually guided generative text-layout pre-training, named ViTLP. Given a document image, the model optimizes hierarchical language and layout modeling objectives to generate the interleaved text and layout sequence. In addition, to address the limitation of processing long documents by Transformers, we introduce a straightforward yet effective multi-segment generative pre-training scheme, facilitating ViTLP to process word-intensive documents of any length. ViTLP can function as a native OCR model to localize and recognize texts of document images. Besides, ViTLP can be effectively applied to various downstream VDU tasks. Extensive experiments show that ViTLP achieves competitive performance over existing baselines on benchmark VDU tasks, including information extraction, document classification, and document question answering.