TinyR1-32B-Preview: Boosting Accuracy with Branch-Merge Distillation

• URL: http://arxiv.org/abs/2503.04872v2
• Date: Mon, 17 Mar 2025 10:36:30 GMT
• Title: TinyR1-32B-Preview: Boosting Accuracy with Branch-Merge Distillation
• Abstract summary: We introduce the Branch-Merge distillation approach, which enhances model compression through two phases.<n>We validate our distillation approach using DeepSeek-R1 as the teacher and DeepSeek-R1-Distill-Qwen-32B as the student.<n>The resulting merged model, TinyR1-32B-Preview, outperforms its counterpart DeepSeek-R1-Distill-Qwen-32B across multiple benchmarks.
• Score: 19.938309176933902
• License: http://creativecommons.org/licenses/by/4.0/
• Abstract: The challenge of reducing the size of Large Language Models (LLMs) while maintaining their performance has gained significant attention. However, existing methods, such as model distillation and transfer learning, often fail to achieve high accuracy. To address this limitation, we introduce the Branch-Merge distillation approach, which enhances model compression through two phases: (1) the Branch Phase, where knowledge from a large teacher model is \textit{selectively distilled} into specialized student models via domain-specific supervised fine-tuning (SFT); And (2) the Merge Phase, where these student models are merged to enable cross-domain knowledge transfer and improve generalization. We validate our distillation approach using DeepSeek-R1 as the teacher and DeepSeek-R1-Distill-Qwen-32B as the student. The resulting merged model, TinyR1-32B-Preview, outperforms its counterpart DeepSeek-R1-Distill-Qwen-32B across multiple benchmarks, including Mathematics (+5.5 points), Coding (+4.4 points) and Science (+2.9 points), while achieving near-equal performance to DeepSeek-R1 on AIME 2024. The Branch-Merge distillation approach provides a scalable solution for creating smaller, high-performing LLMs with reduced computational cost and time.

Related papers

• Towards Compact Unified Multimodal Tracking: Synergizing Knowledge Distillation with Structural Pruning [80.73561460934809]
  multimodal object tracking has achieved remarkable robustness by leveraging complementary sensor data.<n>We identify the prediction head as a critical but often overlooked efficiency bottleneck.<n>By strategically streamlining the decoder architecture, we unlock the potential for real-time inference.
  arXiv  Detail & Related papers  (2026-08-02T20:41:33Z)
• Turning the TIDE: Cross-Architecture Distillation for Diffusion Large Language Models [25.171094765557996]
  Distilling 8B dense and 16B MoE teachers into a 0.6B student via two heterogeneous pipelines outperforms the baseline by an average of 1.53 points across eight benchmarks.<n>HumanEval scores reach 48.78 compared to 32.3 for the AR baseline.
  arXiv  Detail & Related papers  (2026-04-29T17:59:01Z)
• Probing to Refine: Reinforcement Distillation of LLMs via Explanatory Inversion [84.20493238687187]
  We introduce a novel framework that moves beyond simple mimicry to instill a deeper conceptual understanding.<n>underlinetextitFirst, to address pattern memorization, Explanatory Inversion (EI) generates targeted explanatory probes''<n>underlinetextitSecond, to improve generalization, Explanatory GRPO (texttEXGRPO) uses a reinforcement learning algorithm with a novel Dialogue Structure Utility Bonus.
  arXiv  Detail & Related papers  (2026-02-26T23:01:46Z)
• Large Scale Diffusion Distillation via Score-Regularized Continuous-Time Consistency [60.74505433956616]
  continuous-time consistency model (sCM) is theoretically principled and empirically powerful for accelerating academic-scale diffusion.<n>We first develop a parallelism-compatible FlashAttention-2 JVP kernel, enabling sCM training on models with over 10 billion parameters and high-dimensional video tasks.<n>We propose the score-regularized continuous-time consistency model (rCM), which incorporates score distillation as a long-skip regularizer.
  arXiv  Detail & Related papers  (2025-10-09T16:45:30Z)
• AHDMIL: Asymmetric Hierarchical Distillation Multi-Instance Learning for Fast and Accurate Whole-Slide Image Classification [51.525891360380285]
  AHDMIL is an Asymmetric Hierarchical Distillation Multi-Instance Learning framework.<n>It eliminates irrelevant patches through a two-step training process.<n>It consistently outperforms previous state-of-the-art methods in both classification performance and inference speed.
  arXiv  Detail & Related papers  (2025-08-07T07:47:16Z)
• CAST: Contrastive Adaptation and Distillation for Semi-Supervised Instance Segmentation [14.605642634292655]
  We introduce CAST, a semi-supervised knowledge distillation (SSKD) framework that compresses pre-trained vision foundation models (VFM) into compact experts using limited labeled and abundant unlabeled data.<n>Cast unfolds in three stages: (1) domain adaptation of the VFM(s) via self-training with contrastive calibration, (2) knowledge transfer through a unified multi-objective loss, and (3) student refinement to mitigate residual pseudo-label bias.
  arXiv  Detail & Related papers  (2025-05-28T02:45:42Z)
• Gated Attention for Large Language Models: Non-linearity, Sparsity, and Attention-Sink-Free [81.65559031466452]
  We conduct experiments to investigate gating-augmented softmax attention variants.<n>We find that a simple modification-applying a head-specific sigmoid gate after the Scaled Dot-Product Attention (SDPA)-consistently improves performance.
  arXiv  Detail & Related papers  (2025-05-10T17:15:49Z)
• Marco-o1 v2: Towards Widening The Distillation Bottleneck for Reasoning Models [39.22557129190619]
  Distillation--post-training on LRMs-generated data--is a straightforward yet effective method to enhance the reasoning abilities of smaller models.<n>To alleviate this bottleneck, we propose constructing tree-based CoT data from scratch via Monte Carlo Tree Search.<n>We then exploit a set of CoT-aware approaches, including Thoughts Length Balance, Fine-grained DPO, and Joint Post-training Objective, to enhance SFT and RL on the constructed data.
  arXiv  Detail & Related papers  (2025-03-03T12:17:36Z)
• Multi-Level Decoupled Relational Distillation for Heterogeneous Architectures [6.231548250160585]
  Multi-Level Decoupled Knowledge Distillation (MLDR-KD) improves student model performance with gains of up to 4.86% on CodeAR-100 and 2.78% on Tiny-ImageNet datasets respectively.
  arXiv  Detail & Related papers  (2025-02-10T06:41:20Z)
• Gap Preserving Distillation by Building Bidirectional Mappings with A Dynamic Teacher [43.678380057638016]
  Gap Preserving Distillation (GPD) method trains an additional dynamic teacher model from scratch along with training the student to bridge this gap. In experiments, GPD significantly outperforms existing distillation methods on top of both CNNs and transformers architectures. GPD also generalizes well to the scenarios without a pre-trained teacher, including training from scratch and fine-tuning, yielding a large improvement of 1.80% and 0.89% on ResNet18.
  arXiv  Detail & Related papers  (2024-10-05T12:29:51Z)
• EmbedDistill: A Geometric Knowledge Distillation for Information Retrieval [83.79667141681418]
  Large neural models (such as Transformers) achieve state-of-the-art performance for information retrieval (IR) We propose a novel distillation approach that leverages the relative geometry among queries and documents learned by the large teacher model. We show that our approach successfully distills from both dual-encoder (DE) and cross-encoder (CE) teacher models to 1/10th size asymmetric students that can retain 95-97% of the teacher performance.
  arXiv  Detail & Related papers  (2023-01-27T22:04:37Z)
• TinyMIM: An Empirical Study of Distilling MIM Pre-trained Models [31.16595289223858]
  Masked image modeling (MIM) performs strongly in pre-training large vision Transformers (ViTs) However, small models that are critical for real-world applications cannot or only marginally benefit from this pre-training approach. We explore distillation techniques to transfer the success of large MIM-based pre-trained models to smaller ones.
  arXiv  Detail & Related papers  (2023-01-03T18:59:54Z)
• Point-to-Voxel Knowledge Distillation for LiDAR Semantic Segmentation [74.67594286008317]
  This article addresses the problem of distilling knowledge from a large teacher model to a slim student network for LiDAR semantic segmentation. We propose the Point-to-Voxel Knowledge Distillation (PVD), which transfers the hidden knowledge from both point level and voxel level.
  arXiv  Detail & Related papers  (2022-06-05T05:28:32Z)
• ERNIE-Search: Bridging Cross-Encoder with Dual-Encoder via Self On-the-fly Distillation for Dense Passage Retrieval [54.54667085792404]
  We propose a novel distillation method that significantly advances cross-architecture distillation for dual-encoders. Our method 1) introduces a self on-the-fly distillation method that can effectively distill late interaction (i.e., ColBERT) to vanilla dual-encoder, and 2) incorporates a cascade distillation process to further improve the performance with a cross-encoder teacher.
  arXiv  Detail & Related papers  (2022-05-18T18:05:13Z)
• New Perspective on Progressive GANs Distillation for One-class Novelty Detection [21.90786581579228]
  Generative Adversarial Network based on thecoder-Decoder-Encoder scheme (EDE-GAN) achieves state-of-the-art performance. New technology, Progressive Knowledge Distillation with GANs (P-KDGAN) connects two standard GANs through the designed distillation loss. Two-step progressive learning continuously augments the performance of student GANs with improved results over single-step approach.
  arXiv  Detail & Related papers  (2021-09-15T13:45:30Z)
• Towards Accurate Knowledge Transfer via Target-awareness Representation Disentanglement [56.40587594647692]
  We propose a novel transfer learning algorithm, introducing the idea of Target-awareness REpresentation Disentanglement (TRED) TRED disentangles the relevant knowledge with respect to the target task from the original source model and used as a regularizer during fine-tuning the target model. Experiments on various real world datasets show that our method stably improves the standard fine-tuning by more than 2% in average.
  arXiv  Detail & Related papers  (2020-10-16T17:45:08Z)
• MetaDistiller: Network Self-Boosting via Meta-Learned Top-Down Distillation [153.56211546576978]
  In this work, we propose that better soft targets with higher compatibil-ity can be generated by using a label generator. We can employ the meta-learning technique to optimize this label generator. The experiments are conducted on two standard classificationbenchmarks, namely CIFAR-100 and ILSVRC2012.
  arXiv  Detail & Related papers  (2020-08-27T13:04:27Z)
• Distilling Object Detectors with Task Adaptive Regularization [97.52935611385179]
  Current state-of-the-art object detectors are at the expense of high computational costs and are hard to deploy to low-end devices. Knowledge distillation, which aims at training a smaller student network by transferring knowledge from a larger teacher model, is one of the promising solutions for model miniaturization.
  arXiv  Detail & Related papers  (2020-06-23T15:58:22Z)

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.