FFN-SkipLLM: A Hidden Gem for Autoregressive Decoding with Adaptive Feed Forward Skipping (2024.emnlp-main)
Copied to clipboard
| Challenge: | Autoregressive Large Language Models (LLMs) are omnipresent but typically come with a substantial model size. |
| Approach: | They propose a novel fine-grained skip strategy for autoregressive large language models . they observe the saturation of computationally expensive feed-forward blocks of LLMs . |
| Outcome: | The proposed method can skip 25-30% of FFN blocks with marginal change in performance on knowledge-intensive generation tasks. |
Similar Papers
FiRST: Finetuning Router-Selective Transformers for Input-Adaptive Latency Reduction (2025.findings-emnlp)
Copied to clipboard
| Challenge: | Existing approaches to improve latency via skipping layers have limitations . fiRST is a model-agnostic framework that reduces inference latency while maintaining quality . |
| Approach: | They propose a model-agnostic framework that skips transformer layers during decoding . it is fully compatible with KV caching, enabling faster decoding while maintaining quality . |
| Outcome: | a new framework reduces inference latency by using layer-specific routers to skip transformer layers during decoding. |
Exploring the Hidden Capacity of LLMs for One-Step Text Generation (2025.emnlp-main)
Copied to clipboard
| Challenge: | Large language models (LLMs) can reconstruct surprisingly long texts via autoregressive generation from just one trained input embedding. |
| Approach: | They show that large language models can reconstruct surprisingly long texts via autoregressive generation from just one trained input embedding. |
| Outcome: | The proposed model can generate hundreds of accurate tokens in one token-parallel forward pass, when provided with only two learned embeddings. |
DiffSkip: Differential Layer Skipping in Large Language Models (2025.findings-acl)
Copied to clipboard
| Challenge: | Existing Large Language Models (LLMs) enforce uniform computation across all tokens. |
| Approach: | They propose to dynamically skip FFN blocks based on self-attention difference . they propose to use a lightweight router module to do the same . |
| Outcome: | The proposed model can skip FFN blocks even in token generation tasks. |
Speculative Decoding via Early-exiting for Faster LLM Inference with Thompson Sampling Control Mechanism (2024.findings-acl)
Copied to clipboard
| Challenge: | Existing approaches to generate draft tokens in large language models are expensive and resource-intensive. |
| Approach: | They propose an approach to generate draft tokens using a segment of the LLM and a self-distillation method to enhance the quality of draft token. |
| Outcome: | The proposed approach generates draft tokens using a segment of the LLM and a self-distillation method to improve quality and speed up generation. |
FractalLLM: Lossless Self-Speculative Decoding with Layer Embedded Self-Compression (2025.findings-emnlp)
Copied to clipboard
| Challenge: | Autoregressive decoding requires a full forward pass for each generated token, increasing inference latency. |
| Approach: | They propose a lossless self-speculative decoding method that embeds a compressed model within selected decoder layers of the original model. |
| Outcome: | The proposed method achieves substantial speed-ups (up to 2.47) over standard autoregressive decoding. |
LLM Braces: Straightening Out LLM Predictions with Relevant Sub-Updates (2025.acl-long)
Copied to clipboard
| Challenge: | Recent studies reveal that much of the knowledge in a Transformer-based Large Language Model (LLM) is encoded in its feed-forward (FFN) layers, where each FNN layer can be interpreted as the summation of sub-updates, each corresponding to a weighted column vector from the FFN’s value parameter matrix. |
| Approach: | They propose a method that computes relevance scores associated with value vectors in FFN layers and leverages these scores to dynamically adjust the contribution of sub-updates. |
| Outcome: | The proposed framework outperforms baseline approaches in fine-tuning and zero-shot settings while requiring significantly fewer tunable parameters. |
JumpCoder: Go Beyond Autoregressive Coder via Online Modification (2024.acl-long)
Copied to clipboard
| Challenge: | Existing code large language models lack reversibility and autoregressive sequential generation is incapable of correcting previous missing statements as humans do. |
| Approach: | They propose a model-agnostic framework that enables human-like online modification and non-sequential generation to augment code large language models. |
| Outcome: | The proposed framework enables human-like modification and non-sequential generation to augment code large language models. |
Position-Aware Depth Decay Decoding (D3): Boosting Large Language Model Inference Efficiency (2025.findings-acl)
Copied to clipboard
| Challenge: | Recent dynamic computation methods show that not all components are required for inference, enabling a training-free pipeline. |
| Approach: | They propose a token-position aware layer skipping framework to save 1.5x times operations efficiently while maintaining performance. |
| Outcome: | The proposed algorithm achieves 1.5x speedup on large language models with no retraining and with comparable performance on the GSM8K and BBH benchmarks. |
FBS: Modeling Native Parallel Reading inside a Transformer (2026.findings-acl)
Copied to clipboard
| Challenge: | Existing acceleration methods largely patch the autoregressive pipeline and miss core human-reading ingredients. |
| Approach: | They propose a trainable loop that injects a causal loop into Transformers via a 'parafoveal' approach. |
| Outcome: | The proposed model improves quality-efficiency trade-off without increasing parameters . ablations show the three modules are complementary . |
Discovering the Gems in Early Layers: Accelerating Long-Context LLMs with 1000x Input Token Reduction (2026.findings-acl)
Copied to clipboard
| Challenge: | Large Language Models (LLMs) have demonstrated remarkable capabilities in handling long context inputs, but this comes at the cost of increased computational resources and latency. |
| Approach: | They propose an algorithm that uses early LLM layers as filters to select and compress input tokens, reducing the context length for subsequent processing. |
| Outcome: | The proposed method outperforms existing techniques on the Needle in a Haystack task while demonstrating comparable performance on the LongBench challenge. |