Research map / Architectures and efficiency / Quantization
PocketLLM: Ultimate Compression of Large Language Models via Meta Networks
Ye Tian, Chengcheng Wang, Jing Han, Yehui Tang, Kai Han
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| add Reshaped Layer Normalization (RLN) (ablation) | PocketLLM · Llama 2-7B reconstruction | vq_loss: 8.51 -> 2.36 | Improved |
| add RLN (without codebook init) (ablation) | PocketLLM · Llama 2-7B reconstruction | mse: 4.5e-5 -> 9.2e-6 | Improved |
| add codebook initialization (ablation) | PocketLLM · Llama 2-7B reconstruction | vq_loss: 1.23 -> 0.06 | Improved |
| add codebook initialization (ablation) | PocketLLM · Llama 2-7B reconstruction | mse_top100: 0.89 -> 0.36 | Improved |
| use 3-layer MLP instead of 1-layer (ablation) | PocketLLM · Llama 2-7B reconstruction | vq_loss: 4.67 -> 0.06 | Improved |
| use 3-layer MLP instead of 5-layer (ablation) | PocketLLM · Llama 2-7B reconstruction | vq_loss: 0.20 -> 0.06 | Improved |
| increase codebook size from 256 to 32768 (ablation) | PocketLLM · Llama 2-7B reconstruction | vq_loss: 5.54 -> 0.06 | Improved |
| LoRA fine-tuning after compression | PocketLLM · Llama 2-7B ~10x | Avg_acc: 59.53 -> 64.07 | Improved |
| LoRA fine-tuning after compression | PocketLLM · Llama 2-7B ~16x | Avg_acc: 52.45 -> 60.61 | Improved |
| LoRA fine-tuning after compression | PocketLLM · Llama 2-7B ~20x | Avg_acc: 37.91 -> 55.45 | Improved |
| LoRA fine-tuning after compression | PocketLLM · Llama 2-7B ~8x | WikiText-2 perplexity: 5.74 -> 5.27 | Improved |
| compress all attention layers (q,k,v,o) vs none | PocketLLM · Llama 2-7B ~8x | MMLU (5-shot): 45.34 -> 44.14 | Hurt |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 1 of 1 on most shared results.
- QuIP# — QuIP: 2-Bit Quantization of Large Language Models With Guarantees (NeurIPS 2023) · better on 2 of 2 shared results
Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.
Closest papers by meaning
- Compressing LLMs: The Truth is Rarely Pure and Never Simple (ICLR 2024 · similarity 0.91)
- SoLA: Leveraging Soft Activation Sparsity and Low-Rank Decomposition for Large Language Model Compression (AAAI 2025 · similarity 0.91)
- LLMCBench: Benchmarking Large Language Model Compression for Efficient Deployment (NeurIPS 2024 · similarity 0.91)
- Compressing Large Language Models using Low Rank and Low Precision Decomposition (NeurIPS 2024 · similarity 0.90)
- Compact Language Models via Pruning and Knowledge Distillation (NeurIPS 2024 · similarity 0.90)
- BiLLM: Pushing the Limit of Post-Training Quantization for LLMs (ICML 2024 · similarity 0.90)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.
