Lillama: Large Language Models Compression via Low-Rank Feature Distillation

• URL: http://arxiv.org/abs/2412.16719v2
• Date: Sat, 28 Dec 2024 17:45:12 GMT
• Title: Lillama: Large Language Models Compression via Low-Rank Feature Distillation
• Authors: Yaya Sy, Christophe Cerisara, Irina Illina,
• Abstract summary: Lillama is a compression method that distills activations with low-rank weights. It compresses Mixtral-8x7B within minutes on a single A100 GPU, removing 10 billion parameters while retaining over 95% of its original performance. It generalizes well to non-transformer architectures, compressing Mamba-3B by 20% while maintaining 99% performance.
• Score: 8.090496457850852
• License:
• Abstract: Current LLM structured pruning methods typically involve two steps: (1) compression with calibration data and (2) costly continued pretraining on billions of tokens to recover lost performance. This second step is necessary as the first significantly impacts model accuracy. Prior research suggests pretrained Transformer weights aren't inherently low-rank, unlike their activations, which may explain this drop. Based on this observation, we propose Lillama, a compression method that locally distills activations with low-rank weights. Using SVD for initialization and a joint loss combining teacher and student activations, we accelerate convergence and reduce memory use with local gradient updates. Lillama compresses Mixtral-8x7B within minutes on a single A100 GPU, removing 10 billion parameters while retaining over 95% of its original performance. Phi-2 3B can be compressed by 40% with just 13 million calibration tokens, resulting in a small model that competes with recent models of similar size. The method generalizes well to non-transformer architectures, compressing Mamba-3B by 20% while maintaining 99% performance.

Related papers

• DeltaLLM: Compress LLMs with Low-Rank Deltas between Shared Weights [11.047879241587315]
  We introduce DeltaLLM, a new post-training compression technique to reduce the memory footprint of LLMs. For training, we adopt the progressing module replacement method and show that the lightweight training of the low-rank modules is sufficient to achieve performance on par with LLMs of comparable sizes trained from scratch. Our method also outperforms compression techniques JointDrop, LaCo, ShortGPT and SliceGPT with the same number of parameters removed.
  arXiv  Detail & Related papers  (2025-01-30T18:59:55Z)
• Direct Quantized Training of Language Models with Stochastic Rounding [12.028887152979046]
  This paper explores the potential of directly updating the quantized low-precision weight matrices without relying on the straight-through estimator during backpropagation. Experimental results on our LLaMA-structured models indicate that training with only low-precision weights is feasible even when they are constrained to ternary values. Our models can also perform inference using ternary weights, showcasing their flexibility in deployment.
  arXiv  Detail & Related papers  (2024-12-06T05:41:11Z)
• LoRC: Low-Rank Compression for LLMs KV Cache with a Progressive Compression Strategy [59.1298692559785]
  Key-Value ( KV) cache is crucial component in serving transformer-based autoregressive large language models (LLMs) Existing approaches to mitigate this issue include: (1) efficient attention variants integrated in upcycling stages; (2) KV cache compression at test time; and (3) KV cache compression at test time. We propose a low-rank approximation of KV weight matrices, allowing plug-in integration with existing transformer-based LLMs without model retraining. Our method is designed to function without model tuning in upcycling stages or task-specific profiling in test stages.
  arXiv  Detail & Related papers  (2024-10-04T03:10:53Z)
• Token Compensator: Altering Inference Cost of Vision Transformer without Re-Tuning [63.43972993473501]
  Token compression expedites the training and inference of Vision Transformers (ViTs) However, when applied to downstream tasks, compression degrees are mismatched between training and inference stages. We propose a model arithmetic framework to decouple the compression degrees between the two stages.
  arXiv  Detail & Related papers  (2024-08-13T10:36:43Z)
• From GaLore to WeLore: How Low-Rank Weights Non-uniformly Emerge from Low-Rank Gradients [86.40635601953446]
  We study the emergence of low-rank structures across different layers of Modern Large Language Models. We present Weight Low-Rank Projection (WeLore) that unifies weight compression and memory-efficient fine-tuning as ONE.
  arXiv  Detail & Related papers  (2024-07-15T21:05:20Z)
• Activations and Gradients Compression for Model-Parallel Training [85.99744701008802]
  We study how simultaneous compression of activations and gradients in model-parallel distributed training setup affects convergence. We find that gradients require milder compression rates than activations. Experiments also show that models trained with TopK perform well only when compression is also applied during inference.
  arXiv  Detail & Related papers  (2024-01-15T15:54:54Z)
• TinyCLIP: CLIP Distillation via Affinity Mimicking and Weight Inheritance [97.01406871579525]
  We propose a novel cross-modal distillation method, called TinyCLIP, for large-scale language-image pre-trained models. We show that TinyCLIP can reduce the size of the pre-trained CLIP ViT-B/32 by 50%, while maintaining comparable zero-shot performance. Our TinyCLIP ViT-8M/16, trained on YFCC-15M, achieves an impressive zero-shot top-1 accuracy of 41.1% on ImageNet.
  arXiv  Detail & Related papers  (2023-09-21T17:59:53Z)
• Blockwise Compression of Transformer-based Models without Retraining [6.118476907408718]
  We propose BCT, a framework of blockwise compression for transformers without retraining. Unlike layerwise compression methods, BCT achieves finer compression of the entire transformer by operating blockwise. BCT effectively compresses all components of the model, including but not limited to the embedding, matrix multiplication, GELU, Softmax, layer normalization, and intermediate results.
  arXiv  Detail & Related papers  (2023-04-04T02:55:40Z)
• CrAM: A Compression-Aware Minimizer [103.29159003723815]
  We propose a new compression-aware minimizer dubbed CrAM that modifies the optimization step in a principled way. CrAM produces dense models that can be more accurate than the standard SGD/Adam-based baselines, but which are stable under weight pruning. CrAM can produce sparse models which perform well for transfer learning, and it also works for semi-structured 2:4 pruning patterns supported by GPU hardware.
  arXiv  Detail & Related papers  (2022-07-28T16:13:28Z)
• Compression-aware Continual Learning using Singular Value Decomposition [2.4283778735260686]
  We propose a compression based continual task learning method that can dynamically grow a neural network. Inspired by the recent model compression techniques, we employ compression-aware training and perform low-rank weight approximations. Our method achieves compressed representations with minimal performance degradation without the need for costly fine-tuning.
  arXiv  Detail & Related papers  (2020-09-03T23:29:50Z)

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.