LightTransfer: Your Long-Context LLM is Secretly a Hybrid Model with Effortless Adaptation

• URL: http://arxiv.org/abs/2410.13846v2
• Date: Tue, 04 Feb 2025 13:45:37 GMT
• Title: LightTransfer: Your Long-Context LLM is Secretly a Hybrid Model with Effortless Adaptation
• Authors: Xuan Zhang, Fengzhuo Zhang, Cunxiao Du, Chao Du, Tianyu Pang, Wei Gao, Min Lin,
• Abstract summary: Scaling language models to handle longer contexts introduces substantial memory challenges.<n>We propose LightTransfer, a method that transforms models such as LLaMA into hybrid variants.<n>Our approach identifies lazy layers -- those focusing on recent or initial tokens -- and replaces their full attention with streaming attention.
• Score: 37.21518386315535
• License: http://arxiv.org/licenses/nonexclusive-distrib/1.0/
• Abstract: Scaling language models to handle longer contexts introduces substantial memory challenges due to the growing cost of key-value (KV) caches. Motivated by the efficiency gains of hybrid models and the broad availability of pretrained large transformer backbones, we explore transitioning transformer models into hybrid architectures for a more efficient generation. In this work, we propose LightTransfer, a lightweight method that transforms models such as LLaMA into hybrid variants. Our approach identifies lazy layers -- those focusing on recent or initial tokens -- and replaces their full attention with streaming attention. This transformation can be performed without any training for long-context understanding tasks or with minimal fine-tuning for o1-like long reasoning generation tasks that require stronger reasoning capabilities. Experiments across diverse benchmarks and models (e.g., LLaMA, Mistral, QwQ-STILL) demonstrate that, even when half of the layers are identified as lazy, LightTransfer achieves up to 2.17$\times$ throughput improvement with minimal performance loss ($<1.5\%$ on LongBench) and achieves 53.3\% on math benchmark AIME24 of advanced o1-like long reasoning model QwQ-STILL.

Related papers

• Sliding Window Attention Training for Efficient Large Language Models [55.56483740523027]
  We introduce SWAT, which enables efficient long-context handling via Sliding Window Attention Training. This paper first attributes the inefficiency of Transformers to the attention sink phenomenon resulting from the high variance of softmax operation. Experiments demonstrate that SWAT achieves SOTA performance compared with state-of-the-art linear recurrent architectures on eight benchmarks.
  arXiv  Detail & Related papers  (2025-02-26T05:31:44Z)
• p-MoD: Building Mixture-of-Depths MLLMs via Progressive Ratio Decay [18.958138693220704]
  We propose to build efficient multimodal large language models (MLLMs) by leveraging the Mixture-of-Depths (MoD) mechanism. We adapt the MoD module with two novel designs: tanh-gated weight normalization (TanhNorm) and symmetric token reweighting (STRing) Our model, p-MoD, matches or even surpasses the performance of the baseline models, with only 55.6% TFLOPs and 53.8% KV cache storage during inference, and 77.7% GPU hours during training.
  arXiv  Detail & Related papers  (2024-12-05T18:58:03Z)
• Reassessing Layer Pruning in LLMs: New Insights and Methods [24.394438652261982]
  We show that a simple approach, i.e., pruning the final 25% of layers followed by fine-tuning the textttlm_head and the remaining last three layer, yields remarkably strong performance. We release the optimal model weights on Hface, and the code is available on GitHub.
  arXiv  Detail & Related papers  (2024-11-23T13:31:16Z)
• ShadowKV: KV Cache in Shadows for High-Throughput Long-Context LLM Inference [25.638980944695728]
  ShadowKV is an efficient long-context large language models (LLMs) inference system. It stores the low-rank key cache and offloads the value cache to reduce the memory footprint for larger batch sizes and longer sequences. It can support up to 6$times$ larger batch sizes and boost throughput by up to 3.04$times$ on an A100 GPU.
  arXiv  Detail & Related papers  (2024-10-28T19:08:12Z)
• KVSharer: Efficient Inference via Layer-Wise Dissimilar KV Cache Sharing [58.29726147780976]
  We propose a plug-and-play method called textit KVSharer, which shares the KV cache between layers to achieve layer-wise compression. Experiments show that textit KVSharer can reduce KV cache computation by 30%, thereby lowering memory consumption. We verify that textit KVSharer is compatible with existing intra-layer KV cache compression methods, and combining both can further save memory.
  arXiv  Detail & Related papers  (2024-10-24T08:06:41Z)
• Transformer Layer Injection: A Novel Approach for Efficient Upscaling of Large Language Models [0.0]
  Transformer Layer Injection (TLI) is a novel method for efficiently upscaling large language models (LLMs) Our approach improves upon the conventional Depth Up-Scaling (DUS) technique by injecting new layers into every set of K layers.
  arXiv  Detail & Related papers  (2024-10-15T14:41:44Z)
• The Mamba in the Llama: Distilling and Accelerating Hybrid Models [76.64055251296548]
  We show how to distill large Transformers into linear RNNs by reusing the linear projection weights from attention layers with academic GPU resources. The resulting hybrid model achieves performance comparable to the original Transformer in chat benchmarks. We also introduce a hardware-aware speculative decoding algorithm that accelerates the inference speed of Mamba and hybrid models.
  arXiv  Detail & Related papers  (2024-08-27T17:56:11Z)
• ThinK: Thinner Key Cache by Query-Driven Pruning [63.13363917871414]
  Large Language Models (LLMs) have revolutionized the field of natural language processing, achieving unprecedented performance across a variety of applications. This paper focuses on the long-context scenario, addressing the inefficiencies in KV cache memory consumption during inference. We propose ThinK, a novel query-dependent KV cache pruning method designed to minimize attention weight loss while selectively pruning the least significant channels.
  arXiv  Detail & Related papers  (2024-07-30T17:59:08Z)
• Efficient Inference of Vision Instruction-Following Models with Elastic Cache [76.44955111634545]
  We introduce Elastic Cache, a novel strategy for efficient deployment of instruction-following large vision-language models. We propose an importance-driven cache merging strategy to prune redundancy caches. For instruction encoding, we utilize the frequency to evaluate the importance of caches. Results on a range of LVLMs demonstrate that Elastic Cache not only boosts efficiency but also notably outperforms existing pruning methods in language generation.
  arXiv  Detail & Related papers  (2024-07-25T15:29:05Z)
• SHERL: Synthesizing High Accuracy and Efficient Memory for Resource-Limited Transfer Learning [63.93193829913252]
  We propose an innovative METL strategy called SHERL for resource-limited scenarios. In the early route, intermediate outputs are consolidated via an anti-redundancy operation. In the late route, utilizing minimal late pre-trained layers could alleviate the peak demand on memory overhead.
  arXiv  Detail & Related papers  (2024-07-10T10:22:35Z)
• PQCache: Product Quantization-based KVCache for Long Context LLM Inference [27.523568511043273]
  Key-Value Cache (KVCache) is a crucial component in Large Language Models (LLMs) Current methods selectively determine suitable keys and values for self-attention in LLMs to address the issue. We propose PQCache, which employs Product Quantization (PQ) to manage KVCache, maintaining model quality while ensuring low serving latency.
  arXiv  Detail & Related papers  (2024-07-01T13:05:42Z)
• MiniCache: KV Cache Compression in Depth Dimension for Large Language Models [48.03117580340151]
  Key-Value ( KV) cache stores key-value states of previously generated tokens. The size of the KV cache grows linearly with sequence length, posing challenges for applications requiring long context input and extensive sequence generation. We present a simple yet effective approach, called MiniCache, to compress the KV cache across layers from a novel depth perspective.
  arXiv  Detail & Related papers  (2024-05-23T09:43:52Z)
• Get More with LESS: Synthesizing Recurrence with KV Cache Compression for Efficient LLM Inference [78.65321721142624]
  We focus on a memory bottleneck imposed by the key-value ( KV) cache. Existing KV cache methods approach this problem by pruning or evicting large swaths of relatively less important KV pairs. We propose LESS, a simple integration of a constant sized cache with eviction-based cache methods.
  arXiv  Detail & Related papers  (2024-02-14T18:54:56Z)
• KIVI: A Tuning-Free Asymmetric 2bit Quantization for KV Cache [67.9776980972508]
  We develop a tuning-free 2bit KV cache quantization algorithm named KIVI. KIVI can enable Llama, Falcon, and Mistral models to maintain almost the same quality while using $mathbf2.6times$ less peak memory.
  arXiv  Detail & Related papers  (2024-02-05T06:06:47Z)
• TransNormerLLM: A Faster and Better Large Language Model with Improved TransNormer [34.790081960470964]
  We present TransNormerLLM, the first linear attention-based Large Language Model (LLM) We make advanced modifications that include positional embedding, linear attention acceleration, gating mechanisms, tensor normalization, and inference acceleration and stabilization. We validate our model design through a series of ablations and train models with sizes of 385M, 1B, and 7B on our self-collected corpus.
  arXiv  Detail & Related papers  (2023-07-27T16:45:33Z)
• Fourier Transformer: Fast Long Range Modeling by Removing Sequence Redundancy with FFT Operator [24.690247474891958]
  Fourier Transformer is able to significantly reduce computational costs while retain the ability to inherit from various large pretrained models. Our model achieves state-of-the-art performances among all transformer-based models on the long-range modeling benchmark LRA. For generative seq-to-seq tasks including CNN/DailyMail and ELI5, by inheriting the BART weights our model outperforms the standard BART.
  arXiv  Detail & Related papers  (2023-05-24T12:33:06Z)
• Transformer-based Models for Long-Form Document Matching: Challenges and Empirical Analysis [12.269318291685753]
  We show that simple neural models outperform the more complex BERT-based models. Simple models are also more robust to variations in document length and text perturbations.
  arXiv  Detail & Related papers  (2023-02-07T21:51:05Z)
• Adapted Multimodal BERT with Layer-wise Fusion for Sentiment Analysis [84.12658971655253]
  We propose Adapted Multimodal BERT, a BERT-based architecture for multimodal tasks. adapter adjusts the pretrained language model for the task at hand, while the fusion layers perform task-specific, layer-wise fusion of audio-visual information with textual BERT representations. In our ablations we see that this approach leads to efficient models, that can outperform their fine-tuned counterparts and are robust to input noise.
  arXiv  Detail & Related papers  (2022-12-01T17:31:42Z)
• Long-Short Transformer: Efficient Transformers for Language and Vision [97.2850205384295]
  Long-Short Transformer (Transformer-LS) is an efficient self-attention mechanism for modeling long sequences with linear complexity for both language and vision tasks. It aggregates a novel long-range attention with dynamic projection to model distant correlations and a short-term attention to capture fine-grained local correlations. Our method outperforms the state-of-the-art models on multiple tasks in language and vision domains, including the Long Range Arena benchmark, autoregressive language modeling, and ImageNet classification.
  arXiv  Detail & Related papers  (2021-07-05T18:00:14Z)

This list is automatically generated from the titles and abstracts of the papers in this site.

This site does not guarantee the quality of this site (including all information) and is not responsible for any consequences.