Get Started
Home
Topics
Search
Library
Research questionHow can complete transformer layers be removed from LLMs without sacrificing quality or standard serving compatibility?Removing whole transformer layers can lower decoding cost, but quality losses vary across models and can become severe at aggressive compression levels. The challenge is to reduce model depth without requiring an altered serving architecture.
AI
Evaluation & Benchmarks
Inference Optimization
LLM Pretraining & Post-training
Research Paper
Technology
Latest papersRecent research connected to this question, newest first.XMerge: Cross-Axis Selection and Reconstructive Layer Merging for LLM Depth CompressionEvidence covers XMerge on seven Llama and Qwen backbones ranging from 0.5B to 8B parameters, three layer-reduction levels, CORE's 22-task aggregate, MMLU, zero-shot and in-context settings, perplexity, and a calibration probe on one backbone. The method is evaluated as post-training compression without task labels, end-to-end fine-tuning, architectural changes, or additional inference-time parameters; the reported construction cost amortizes after tens of thousands of requests. Results are limited to the tested models and evaluation settings.research paper · Sep 2, 2026
Related questions
How can extreme low-bit LLM quantization adapt its basis across layers, calibration data, and quantizers?Does reusing Transformer layers improve language-model quality when parameter, compute, and KV-cache budgets are matched?How can large language models cut training and inference costs without materially harming accuracy?How can lossy speculative decoding accelerate LLM inference without distorting token distributions or degrading generation quality?