InfiGFusion: Graph-on-Logits Distillation via Efficient Gromov-Wasserstein for Model Fusion

• URL: http://arxiv.org/abs/2505.13893v1
• Date: Tue, 20 May 2025 03:55:35 GMT
• Title: InfiGFusion: Graph-on-Logits Distillation via Efficient Gromov-Wasserstein for Model Fusion
• Authors: Yuanyi Wang, Zhaoyi Yan, Yiming Zhang, Qi Zhou, Yanggan Gu, Fei Wu, Hongxia Yang,
• Abstract summary: InfiGFusion is a structure-aware fusion framework with a novel textitGraph-on-Logits Distillation (GLD) loss.<n>We show that GLD consistently improves fusion quality and stability.<n>It shows particular strength in complex reasoning tasks, with +35.6 improvement on Multistep Arithmetic and +37.06 on Causal Judgement over SFT.
• Score: 36.27704594180795
• License: http://creativecommons.org/licenses/by/4.0/
• Abstract: Recent advances in large language models (LLMs) have intensified efforts to fuse heterogeneous open-source models into a unified system that inherits their complementary strengths. Existing logit-based fusion methods maintain inference efficiency but treat vocabulary dimensions independently, overlooking semantic dependencies encoded by cross-dimension interactions. These dependencies reflect how token types interact under a model's internal reasoning and are essential for aligning models with diverse generation behaviors. To explicitly model these dependencies, we propose \textbf{InfiGFusion}, the first structure-aware fusion framework with a novel \textit{Graph-on-Logits Distillation} (GLD) loss. Specifically, we retain the top-$k$ logits per output and aggregate their outer products across sequence positions to form a global co-activation graph, where nodes represent vocabulary channels and edges quantify their joint activations. To ensure scalability and efficiency, we design a sorting-based closed-form approximation that reduces the original $O(n^4)$ cost of Gromov-Wasserstein distance to $O(n \log n)$, with provable approximation guarantees. Experiments across multiple fusion settings show that GLD consistently improves fusion quality and stability. InfiGFusion outperforms SOTA models and fusion baselines across 11 benchmarks spanning reasoning, coding, and mathematics. It shows particular strength in complex reasoning tasks, with +35.6 improvement on Multistep Arithmetic and +37.06 on Causal Judgement over SFT, demonstrating superior multi-step and relational inference.

Related papers

• Gated Recursive Fusion: A Stateful Approach to Scalable Multimodal Transformers [0.0]
  Gated Recurrent Fusion (GRF) is a novel architecture that captures the power of cross-modal attention within a linearly scalable, recurrent pipeline.<n>Our work presents a robust and efficient paradigm for powerful, scalable multimodal representation learning.
  arXiv  Detail & Related papers  (2025-07-01T09:33:38Z)
• $α$-Flow: A Unified Framework for Continuous-State Discrete Flow Matching Models [8.705749038874137]
  This work presents a unified framework for Continuous-State Discrete Flow Matching models.<n>We introduce $alpha$-Flow, a family of CS-DFM models that adheres to the canonical $alpha$-geometry of the statistical manifold.<n>We show that the flow matching loss for $alpha$-flow establishes a unified variational bound for the discrete negative log-likelihood.
  arXiv  Detail & Related papers  (2025-04-14T14:51:45Z)
• COrAL: Order-Agnostic Language Modeling for Efficient Iterative Refinement [80.18490952057125]
  Iterative refinement has emerged as an effective paradigm for enhancing the capabilities of large language models (LLMs) on complex tasks. We propose Context-Wise Order-Agnostic Language Modeling (COrAL) to overcome these challenges. Our approach models multiple token dependencies within manageable context windows, enabling the model to perform iterative refinement internally.
  arXiv  Detail & Related papers  (2024-10-12T23:56:19Z)
• Discrete Diffusion Modeling by Estimating the Ratios of the Data Distribution [67.9215891673174]
  We propose score entropy as a novel loss that naturally extends score matching to discrete spaces. We test our Score Entropy Discrete Diffusion models on standard language modeling tasks.
  arXiv  Detail & Related papers  (2023-10-25T17:59:12Z)
• Flag Aggregator: Scalable Distributed Training under Failures and Augmented Losses using Convex Optimization [14.732408788010313]
  ML applications increasingly rely on complex deep learning models and large datasets. To scale computation and data, these models are inevitably trained in a distributed manner in clusters of nodes, and their updates are aggregated before being applied to the model. With data augmentation added to these settings, there is a critical need for robust and efficient aggregation systems. We show that our approach significantly enhances the robustness of state-of-the-art Byzantine resilient aggregators.
  arXiv  Detail & Related papers  (2023-02-12T06:38:30Z)
• FeDXL: Provable Federated Learning for Deep X-Risk Optimization [105.17383135458897]
  We tackle a novel federated learning (FL) problem for optimizing a family of X-risks, to which no existing algorithms are applicable. The challenges for designing an FL algorithm for X-risks lie in the non-decomability of the objective over multiple machines and the interdependency between different machines.
  arXiv  Detail & Related papers  (2022-10-26T00:23:36Z)
• $\ exttt{FedBC}$: Calibrating Global and Local Models via Federated Learning Beyond Consensus [66.62731854746856]
  In federated learning (FL), the objective of collaboratively learning a global model through aggregation of model updates across devices tends to oppose the goal of personalization via local information. In this work, we calibrate this tradeoff in a quantitative manner through a multi-criterion-based optimization. We demonstrate that $texttFedBC$ balances the global and local model test accuracy metrics across a suite datasets.
  arXiv  Detail & Related papers  (2022-06-22T02:42:04Z)
• Efficient semidefinite bounds for multi-label discrete graphical models [6.226454551201676]
  One of the main queries on such models is to identify the SDPWCSP Function on Cost of a Posteri (MAP) Networks. We consider a traditional dualized constraint approach and a dedicated dedicated SDP/Monteiro style method based on row-by-row updates.
  arXiv  Detail & Related papers  (2021-11-24T13:38:34Z)
• AvgOut: A Simple Output-Probability Measure to Eliminate Dull Responses [97.50616524350123]
  We build dialogue models that are dynamically aware of what utterances or tokens are dull without any feature-engineering. The first model, MinAvgOut, directly maximizes the diversity score through the output distributions of each batch. The second model, Label Fine-Tuning (LFT), prepends to the source sequence a label continuously scaled by the diversity score to control the diversity level. The third model, RL, adopts Reinforcement Learning and treats the diversity score as a reward signal.
  arXiv  Detail & Related papers  (2020-01-15T18:32:06Z)

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.