More Expressive Feedforward Layers: Part I. Token-Adaptive Mixing of Activations

• URL: http://arxiv.org/abs/2605.26647v1
• Date: Tue, 26 May 2026 07:30:53 GMT
• Title: More Expressive Feedforward Layers: Part I. Token-Adaptive Mixing of Activations
• Abstract summary: We propose a token-adaptive FFN design that mixes a dictionary of activation functions using lightweight input-dependent gates.<n>MoA consistently achieves lower terminal loss and exhibits more favorable scaling behavior than well-tuned baselines.<n>These results suggest that token-adaptive activation mixing is a simple and effective mechanism for improving FFN expressivity in large language models.
• Score: 16.610720814187754
• License: http://arxiv.org/licenses/nonexclusive-distrib/1.0/
• Abstract: Feedforward network (FFN) layers account for a large fraction of parameters and nonlinear expressivity in Transformer-based large language models (LLMs). Despite the evolution from ReLU and GELU to gated variants such as SwiGLU, most FFN designs still use a single fixed activation function, applying the same nonlinear transformation to all tokens. In this work, we propose Mixture of Activations (MoA), a token-adaptive FFN design that mixes a dictionary of activation functions using lightweight input-dependent gates while sharing the same linear projections. As an input-independent counterpart, we also introduce learnable activations (LA), which form linear combinations of activation functions for both ReLU-type and SwiGLU-type FFNs. Theoretically, we establish strict finite-width expressive separations among fixed-activation FFNs, LA, and MoA: LA strictly contains fixed-activation FFNs, while MoA strictly contains LA, with the additional expressivity arising from input-dependent nonlinear hybridization. Empirically, we evaluate MoA through extensive pre-training experiments on dense and MoE language models ranging from 0.12B to 2B parameters under different token budgets, optimizers, and learning rate schedules. MoA consistently achieves lower terminal loss and exhibits more favorable scaling behavior than well-tuned baselines, with minimal parameter and computational overhead. These results suggest that token-adaptive activation mixing is a simple and effective mechanism for improving FFN expressivity in LLMs.

Related papers

• Empowering Compact LLMs with Fusion of Layer-wise Exits for Recommendation [59.216721553953]
  Large language model-based recommender systems (LLM-RSs) have demonstrated remarkable capabilities, but are computationally unsustainable for many real-world applications.<n>We propose a discriminative framework that enhances compact LLMs while retaining scalable full-corpus ranking.<n>Experiments on three real-world datasets with Qwen 3 1.7B and Llama 3.2 3B show that FLEXRec achieves state-of-the-art accuracy while remaining highly efficient.
  arXiv  Detail & Related papers  (2026-08-18T03:14:28Z)
• Adaptive Depth Sparse Framework: Similarity-Driven Resource Allocation for Pre-Trained LLMs [18.457250148137494]
  We present an Adaptive Depth Sparse Framework (AdaDSF) that converts off-the-shelf pre-trained large language models into depth-sparse models without retraining.<n>On GPT-NeoX and Qwen2.5 over language modeling and commonsense reasoning, AdaDSF substantially reduces inference FLOPs while preserving performance close to dense counterparts.
  arXiv  Detail & Related papers  (2026-07-23T13:13:04Z)
• LimiX-2M: Mitigating Low-Rank Collapse and Attention Bottlenecks in Tabular Foundation Models [56.999481798138625]
  LimiX-2M is a 2M- parameter model that outperforms larger TabPFN-v2 and TabICL baselines on widely used benchmarks.<n>We present a unified tokenize-and-route framework for strong Tabular foundation models (TFMs)<n>Results highlight value-aware tokenization and readout-aligned routing as key levers for improving the accuracy--efficiency trade-off in TFMs.
  arXiv  Detail & Related papers  (2026-06-03T06:07:33Z)
• NerVE: Nonlinear Eigenspectrum Dynamics in LLM Feed-Forward Networks [2.8232103900765693]
  We introduce NerVE, a unified espectral framework for understanding how feedforward networks (FFNs) regulate information flow in high-dimensional latent space.<n>NerVE addresses this gap through lightweight, memory-efficient tracking of eigen dynamics via four complementary geometries.
  arXiv  Detail & Related papers  (2026-03-06T22:50:43Z)
• FreeAct: Freeing Activations for LLM Quantization [89.97086263978058]
  Quantization is pivotal for mitigating the significant memory and computational overhead of Large Language Models.<n>FreeAct is a novel quantization framework that relaxes the static one-to-one constraint to accommodate dynamic activation disparities.<n>Experiments across dLLMs and MLLMs demonstrate that FreeAct significantly outperforms baselines, up to 5.3% performance improvement.
  arXiv  Detail & Related papers  (2026-03-02T12:02:17Z)
• LLaDA-o: An Effective and Length-Adaptive Omni Diffusion Model [77.66516875262963]
  We present textbfLLaDA-o, an effective and length-adaptive omni diffusion model for multimodal understanding and generation.<n>Building on MoD, we introduce a data-centric length adaptation strategy that enables flexible-length decoding in multimodal settings.<n>Experiments show that LLaDA-o achieves state-of-the-art performance among omni-diffusion models on multimodal understanding and generation benchmarks.
  arXiv  Detail & Related papers  (2026-03-01T12:05:06Z)
• WUSH: Near-Optimal Adaptive Transforms for LLM Quantization [52.77441224845925]
  Quantization to low bitwidth is a standard approach for deploying large language models.<n>A few extreme weights and activations stretch the dynamic range and reduce the effective resolution of the quantizer.<n>We derive, for the first time, closed-form optimal linear blockwise transforms for joint weight-activation quantization.
  arXiv  Detail & Related papers  (2025-11-30T16:17:34Z)
• R-Sparse: Rank-Aware Activation Sparsity for Efficient LLM Inference [77.47238561728459]
  R-Sparse is a training-free activation sparsity approach capable of achieving high sparsity levels in advanced LLMs.<n> Experiments on Llama-2/3 and Mistral models across ten diverse tasks demonstrate that R-Sparse achieves comparable performance at 50% model-level sparsity.
  arXiv  Detail & Related papers  (2025-04-28T03:30:32Z)
• DiffoRA: Enabling Parameter-Efficient Fine-Tuning via Differential Module Selection [32.369133126167085]
  Low-Rank Adaptation (LoRA) has gained popularity for its streamlined design by incorporating low-rank matrices into existing pre-trained models.<n>We propose DiffoRA, which enables adaptive adoption of the low-rank decomposition matrices.
  arXiv  Detail & Related papers  (2025-02-13T02:41:34Z)
• FFN-SkipLLM: A Hidden Gem for Autoregressive Decoding with Adaptive Feed Forward Skipping [49.66872823080736]
  Autoregressive Large Language Models (e.g., LLaMa, GPTs) are omnipresent achieving remarkable success in language understanding and generation. To mitigate overload incurred during generation, several early-exit and layer-dropping strategies have been proposed. We propose FFN-SkipLLM, which is an input-adaptive feed-forward skipping strategy.
  arXiv  Detail & Related papers  (2024-04-05T02:35:43Z)
• CRaSh: Clustering, Removing, and Sharing Enhance Fine-tuning without Full Large Language Model [22.870512676002463]
  This paper focuses on Offsite-Tuning (OFT), a representative technique that transfers transformer blocks between centralized LLMs and downstream emulators. Inspired by these observations, we propose CRaSh, involving Clustering, Removing, and Sharing, a training-free strategy to derive improved emulators from LLMs. Our findings demonstrate a linear connectivity among these optima falling over the same basin, thereby highlighting the effectiveness of CRaSh and OFT.
  arXiv  Detail & Related papers  (2023-10-24T03:08:58Z)
• Towards A Unified View of Sparse Feed-Forward Network in Pretraining Large Language Model [58.9100867327305]
  Large and sparse feed-forward layers (S-FFN) have proven effective in scaling up Transformers model size for textitpretraining large language models. We analyzed two major design choices of S-FFN: the memory block (a.k.a. expert) size and the memory block selection method. We found a simpler selection method -- textbftextttAvg-K that selects blocks through their mean aggregated hidden states, achieving lower perplexity in language model pretraining.
  arXiv  Detail & Related papers  (2023-05-23T12:28:37Z)

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.