Pangu Light: Weight Re-Initialization for Pruning and Accelerating LLMs

• URL: http://arxiv.org/abs/2505.20155v1
• Date: Mon, 26 May 2025 15:57:08 GMT
• Title: Pangu Light: Weight Re-Initialization for Pruning and Accelerating LLMs
• Authors: Hanting Chen, Jiarui Qin, Jialong Guo, Tao Yuan, Yichun Yin, Huiling Zhen, Yasheng Wang, Jinpeng Li, Xiaojun Meng, Meng Zhang, Rongju Ruan, Zheyuan Bai, Yehui Tang, Can Chen, Xinghao Chen, Fisher Yu, Ruiming Tang, Yunhe Wang,
• Abstract summary: Large Language Models (LLMs) deliver state-of-the-art capabilities across numerous tasks, but their immense size and inference costs pose significant computational challenges for practical deployment.<n>This paper argues that a critical, often overlooked, aspect in making such aggressive joint pruning viable is the strategic re-initialization and adjustment of remaining weights.<n>We introduce Pangu Light, a framework for LLM acceleration centered around structured pruning and novel weight re-initialization techniques.
• Score: 79.7618807098457
• License: http://arxiv.org/licenses/nonexclusive-distrib/1.0/
• Abstract: Large Language Models (LLMs) deliver state-of-the-art capabilities across numerous tasks, but their immense size and inference costs pose significant computational challenges for practical deployment. While structured pruning offers a promising avenue for model compression, existing methods often struggle with the detrimental effects of aggressive, simultaneous width and depth reductions, leading to substantial performance degradation. This paper argues that a critical, often overlooked, aspect in making such aggressive joint pruning viable is the strategic re-initialization and adjustment of remaining weights to improve the model post-pruning training accuracies. We introduce Pangu Light, a framework for LLM acceleration centered around structured pruning coupled with novel weight re-initialization techniques designed to address this ``missing piece''. Our framework systematically targets multiple axes, including model width, depth, attention heads, and RMSNorm, with its effectiveness rooted in novel re-initialization methods like Cross-Layer Attention Pruning (CLAP) and Stabilized LayerNorm Pruning (SLNP) that mitigate performance drops by providing the network a better training starting point. Further enhancing efficiency, Pangu Light incorporates specialized optimizations such as absorbing Post-RMSNorm computations and tailors its strategies to Ascend NPU characteristics. The Pangu Light models consistently exhibit a superior accuracy-efficiency trade-off, outperforming prominent baseline pruning methods like Nemotron and established LLMs like Qwen3 series. For instance, on Ascend NPUs, Pangu Light-32B's 81.6 average score and 2585 tokens/s throughput exceed Qwen3-32B's 80.9 average score and 2225 tokens/s.

Related papers

• Amber Pruner: Leveraging N:M Activation Sparsity for Efficient Prefill in Large Language Models [10.779802084898206]
  We introduce Amber Pruner, a training-free N:M activation sparsity method designed specifically for the prefill stage.<n>Experiments show that Amber Pruner can effectively sparsify and accelerate more than 55% of linear computations without requiring model retraining.<n>We propose Outstanding-sparse, a unified framework that integrates Amber Pruner with post-training W8A8 quantization.
  arXiv  Detail & Related papers  (2025-08-04T07:22:36Z)
• SPAP: Structured Pruning via Alternating Optimization and Penalty Methods [2.1388885579612804]
  Large language models (LLMs) are often constrained by their substantial computational and memory demands.<n>We propose SPAP (Structured Pruning via Alternating Optimization and Penalty Methods), a novel and efficient structured pruning framework for LLMs grounded in optimization theory.<n>Our work offers a practical, optimization-driven solution for pruning LLMs while preserving model performance.
  arXiv  Detail & Related papers  (2025-05-06T09:47:53Z)
• LESA: Learnable LLM Layer Scaling-Up [57.0510934286449]
  Training Large Language Models (LLMs) from scratch requires immense computational resources, making it prohibitively expensive.<n>Model scaling-up offers a promising solution by leveraging the parameters of smaller models to create larger ones.<n>We propose textbfLESA, a novel learnable method for depth scaling-up.
  arXiv  Detail & Related papers  (2025-02-19T14:58:48Z)
• RoSTE: An Efficient Quantization-Aware Supervised Fine-Tuning Approach for Large Language Models [53.571195477043496]
  We propose an algorithm named Rotated Straight-Through-Estimator (RoSTE)<n>RoSTE combines quantization-aware supervised fine-tuning (QA-SFT) with an adaptive rotation strategy to reduce activation outliers.<n>Our findings reveal that the prediction error is directly proportional to the quantization error of the converged weights, which can be effectively managed through an optimized rotation configuration.
  arXiv  Detail & Related papers  (2025-02-13T06:44:33Z)
• FASP: Fast and Accurate Structured Pruning of Large Language Models [24.185245582500876]
  We introduce FASP (Fast and Accurate Structured Pruning), a novel structured pruning framework for large language models (LLMs)<n>FASP employs a distinctive pruning structure that interlinks sequential layers, allowing for the removal of columns in one layer while simultaneously eliminating corresponding rows in the preceding layer without incurring additional performance loss.<n>We evaluate FASP on the OPT and LLaMA model families, demonstrating superior performance in terms of perplexity and accuracy on downstream tasks compared to state-of-the-art methods.
  arXiv  Detail & Related papers  (2025-01-16T09:38:39Z)
• PAT: Pruning-Aware Tuning for Large Language Models [19.622152991641045]
  Large language models excel in language tasks, especially with supervised fine-tuning after pre-training.<n>Traditional post-hoc pruning often leads to significant performance loss.<n>We propose the Pruning-Aware Tuning (PAT) paradigm to eliminate model redundancy.
  arXiv  Detail & Related papers  (2024-08-27T01:04:14Z)
• Bypass Back-propagation: Optimization-based Structural Pruning for Large Language Models via Policy Gradient [57.9629676017527]
  We propose an optimization-based structural pruning on Large-Language Models. We learn the pruning masks in a probabilistic space directly by optimizing the loss of the pruned model. Our method operates for 2.7 hours with around 35GB memory for the 13B models on a single A100 GPU.
  arXiv  Detail & Related papers  (2024-06-15T09:31:03Z)
• SPP: Sparsity-Preserved Parameter-Efficient Fine-Tuning for Large Language Models [53.638791265113625]
  Sparsity-Preserved efficient fine-tuning method for large language models. Code will be made available at https://github.com/Lucky-Lance/SPP.
  arXiv  Detail & Related papers  (2024-05-25T04:55:27Z)
• LoRAPrune: Structured Pruning Meets Low-Rank Parameter-Efficient Fine-Tuning [56.88751562302793]
  Low-rank adaption (LoRA) has emerged to fine-tune large language models (LLMs) LoRAPrune is a new framework that delivers an accurate structured pruned model in a highly memory-efficient manner. LoRAPrune achieves a reduction in perplexity by 4.81 on WikiText2 and 3.46 on PTB, while also decreasing memory usage by 52.6%.
  arXiv  Detail & Related papers  (2023-05-28T15:15:48Z)
• Iterative Soft Shrinkage Learning for Efficient Image Super-Resolution [91.3781512926942]
  Image super-resolution (SR) has witnessed extensive neural network designs from CNN to transformer architectures. This work investigates the potential of network pruning for super-resolution iteration to take advantage of off-the-shelf network designs and reduce the underlying computational overhead. We propose a novel Iterative Soft Shrinkage-Percentage (ISS-P) method by optimizing the sparse structure of a randomly network at each and tweaking unimportant weights with a small amount proportional to the magnitude scale on-the-fly.
  arXiv  Detail & Related papers  (2023-03-16T21:06:13Z)
• Extrapolation for Large-batch Training in Deep Learning [72.61259487233214]
  We show that a host of variations can be covered in a unified framework that we propose. We prove the convergence of this novel scheme and rigorously evaluate its empirical performance on ResNet, LSTM, and Transformer.
  arXiv  Detail & Related papers  (2020-06-10T08:22:41Z)

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.