An official website of the United States government
Aligning large language models (LLMs) with human objectives is crucial for real-world applications. However, fine-tuning LLMs for alignment often suffers from unstable training and requires substantial computing resources. Test-time alignment techniques, such as prompting and guided decoding, do not modify the underlying model, and their performance remains dependent on the original model's capabilities. To address these challenges, we propose aligning LLMs through representation editing. The core of our method is to view a pre-trained autoregressive LLM as a discrete-time stochastic dynamical system. To achieve alignment for specific objectives, we introduce external control signals into the state space of this language dynamical system. We train a value function directly on the hidden states according to the Bellman equation, enabling gradient-based optimization to obtain the optimal control signals at test time. Our experiments demonstrate that our method outperforms existing test-time alignment techniques while requiring significantly fewer resources compared to fine-tuning methods. Our code is available at https://github.com/Lingkai-Kong/RE-Control.
more »
« less
Towards Green AI in Fine-Tuning Large Language Models via Adaptive Backpropagation
Fine-tuning is essential to adapting pre-trained large language models to downstream applications. With the increasing popularity of LLM-enabled applications, fine-tuning has been performed intensively worldwide, incurring a tremendous amount of computing costs that correspond to big carbon footprint and environmental impact. Mitigating such environmental impact directly correlates to reducing the fine-tuning FLOPs. Existing fine-tuning schemes focus on either saving memory or reducing the overhead of computing weight updates, but cannot achieve sufficient FLOPs reduction due to their ignorance of the training cost in backpropagation. To address this limitation, in this paper we present GreenTrainer, a new technique that minimizes the FLOPs of LLM fine-tuning via adaptive backpropagation, which adaptively selects the most appropriate set of LLM tensors for fine-tuning based on their importance and backpropagation cost in training. Experiment results show that GreenTrainer can save up to 64% training FLOPs compared to full fine-tuning, without any noticeable accuracy loss. Compared to the existing schemes such as Prefix Tuning and LoRA, GreenTrainer can achieve up to 4% improvement of model accuracy, with on-par FLOPs reduction.
more »
« less
- PAR ID:
- 10615691
- Publisher / Repository:
- in Proceedings of the 12th International Conference on Learning Representations (ICLR), 2024.
- Date Published:
- Format(s):
- Medium: X
- Sponsoring Org:
- National Science Foundation
More Like this
-
-
Transfer learning on edge is challenging due to on-device limited resources. Existing work addresses this issue by training a subset of parameters or adding model patches. Developed with inference in mind, Inverted Residual Blocks (IRBs) split a convolutional layer into depthwise and pointwise convolutions, leading to more stacking layers, e.g., convolution, normalization, and activation layers. Though they are efficient for inference, IRBs require that additional activation maps are stored in memory for training weights for convolution layers and scales for normalization layers. As a result, their high memory cost prohibits training IRBs on resource-limited edge devices, and making them unsuitable in the context of transfer learning. To address this issue, we present MobileTL, a memory and computationally efficient on-device transfer learning method for models built with IRBs. MobileTL trains the shifts for internal normalization layers to avoid storing activation maps for the backward pass. Also, MobileTL approximates the backward computation of the activation layer (e.g., Hard-Swish and ReLU6) as a signed function which enables storing a binary mask instead of activation maps for the backward pass. MobileTL fine-tunes a few top blocks (close to output) rather than propagating the gradient through the whole network to reduce the computation cost. Our method reduces memory usage by 46% and 53% for MobileNetV2 and V3 IRBs, respectively. For MobileNetV3, we observe a 36% reduction in floating-point operations (FLOPs) when fine-tuning 5 blocks, while only incurring a 0.6% accuracy reduction on CIFAR10. Extensive experiments on multiple datasets demonstrate that our method is Pareto-optimal (best accuracy under given hardware constraints) compared to prior work in transfer learning for edge devices.more » « less
-
null (Ed.)This paper aims at reducing computation for Retinanet, an mAP-30-tier network, to facilitate its practical deployment on edge devices for providing IoT-based object detection services. We first validate RetinaNet has the best FLOP-mAP trade-off among all mAP-30-tier network. Then, we propose a light-weight RetinaNet structure with effective computation- accuracy trade-off by only reducing FLOPs in computationally intensive layers. Compared with the most common way of trading off computation with accuracy-input image scaling, the proposed solution shows a consistently better FLOPs-mAP trade-off curve. Light-weight RetinaNet achieves a 0.3% mAP improvement at 1.8x FLOPs reduction point over the original RetinaNet, and gains 1.8x more energy-efficiency on an Intel Arria 10 FPGA accelerator in the context of edge computing. The proposed method potentially can help a wide range of the object detection applications to move closer to a preferred corner for a better runtime and accuracy, while enjoys more energy-efficient inference at the edge.more » « less
-
Current PEFT methods for LLMs can achieve either high quality, efficient training, or scalable serving, but not all three simultaneously. To address this limitation, we investigate sparse fine-tuning and observe a remarkable improvement in generalization ability. Utilizing this key insight, we propose a family of \underline{S}tructured \underline{S}parse \underline{F}ine-\underline{T}uning (\textbf{\model}) methods for LLMs, which \textit{concurrently achieve state-of-the-art fine-tuning performance, training efficiency, and inference scalability}. \model \mbox{accomplishes this by ``selecting sparsely and computing densely". It selects a few} heads and channels in the MHA and FFN modules for each Transformer block, respectively. Next, it co-permutes weight matrices on both sides of the coupled structures in LLMs to connect the selected components in each layer into a dense submatrix. Finally, \model performs in-place gradient updates on all submatrices. Through theoretical analysis and empirical results, our method prevents overfitting and forgetting, delivers SOTA performance on both commonsense and arithmetic reasoning with 4.6$$\%$$ and 1.3$$\%$$ average improvements compared to LoRA, and surpasses full FT by 11.5$$\%$$ when generalizing to various domains after instruction tuning. Using our partial backpropagation algorithm, \model saves training memory up to 3$$\times$$ and improves latency by 1.5-2.7$$\times$$ compared to full FT, while delivering an average 10\% improvement over LoRA on both metrics. We further demonstrate that the weight updates in \model can be decoupled into adapters, enabling effective fusion, fast switch, and efficient parallelism for serving multiple fine-tuned models.more » « less
-
Brain-inspired Hyperdimensional (HD) computing models cognition by exploiting properties of high dimensional statistics– high-dimensional vectors, instead of working with numeric values used in contemporary processors. A fundamental weakness of existing HD computing algorithms is that they require to use floating point models in order to provide acceptable accuracy on realistic classification problems. However, working with floating point values significantly increases the HD computation cost. To address this issue, we proposed QuantHD, a novel framework for quantization of HD computing model during training. QuantHD enables HD computing to work with a low-cost quantized model (binary or ternary model) while providing a similar accuracy as the floating point model. We accordingly propose an FPGA implementation which accelerates HD computing in both training and inference phases. We evaluate QuantHD accuracy and efficiency on various real-world applications, and observe that QuantHD can achieve on average 17.2% accuracy improvement as compared to the existing binarized HD computing algorithms which provide a similar computation cost. In terms of efficiency, QuantHD FPGA implementation can achieve on average 42.3× and 4.7× (34.1× and 4.1×) energy efficiency improvement and speedup during inference (training) as compared to the state-of-the-art HD computing algorithms.more » « less
- Free Publicly Accessible Full Text
-
Accepted Manuscript1.0
- Conference Paper:
- The DOI is not currently available.
