GRADE: Replacing Policy Gradients with Backpropagation for LLM Alignment

• URL: http://arxiv.org/abs/2601.11574v1
• Date: Tue, 30 Dec 2025 03:45:32 GMT
• Title: GRADE: Replacing Policy Gradients with Backpropagation for LLM Alignment
• Abstract summary: GRADE is a method that replaces high-variance policy gradient estimation with direct backpropagation.<n>GRADE-STE achieves a test reward of 0.763 +- 0.344 compared to PPO's 0.510 +- 0.313 and REINFORCE's 0.617 +- 0.378.
• Score: 0.0
• License: http://creativecommons.org/licenses/by-nc-sa/4.0/
• Abstract: Reinforcement learning from human feedback (RLHF) has become the dominant paradigm for aligning large language models with human preferences. However, policy gradient methods such as PPO suffer from high variance gradient estimates, requiring careful hyperparameter tuning and extensive computational resources. We introduce GRADE (Gumbel-softmax Relaxation for Alignment via Differentiable Estimation), a method that replaces high-variance policy gradient estimation with direct backpropagation through a differentiable relaxation of the discrete token sampling process. Using the Gumbel-Softmax reparameterization with straight-through estimation (GRADE-STE), we enable end-to-end gradient flow from reward signals through generated tokens to model parameters. On sentiment-controlled text generation using the IMDB dataset, GRADE-STE achieves a test reward of 0.763 +- 0.344 compared to PPO's 0.510 +- 0.313 and REINFORCE's 0.617 +- 0.378, representing a 50% relative improvement over PPO. Critically, GRADE-STE exhibits gradient variance over 14 times lower than REINFORCE and maintains stable training dynamics throughout optimization. Our rigorous evaluation with proper train/validation/test splits demonstrates that these improvements generalize to held-out data, with GRADE-STE showing the best generalization characteristics among all methods tested. GRADE offers a simpler, more stable, and more effective alternative to reinforcement learning for LLM alignment.

Related papers

• SP3O: Reinforcement Learning from Segment Preferences without Reward Modeling [13.20526811256771]
  We introduce a novel reward-model-free, critic-free, and gradient-based PbRL algorithm compatible with segment preferences named Segment Pairwise Proximal Policy Optimization (SP3O)<n>We provide a theoretical basis for the algorithm and analyze the tradeoff in choosing the segment length.
  arXiv  Detail & Related papers  (2026-08-03T23:28:06Z)
• Stochastic Gradient Optimization with Model-Assisted Sampling [0.0]
  We propose a model-assisted sampling framework that interprets mini-batch gradients through survey sampling theory.<n>Our aim is to bridge machine learning optimization and survey sampling theory by combining their perspectives on sample-based estimation and variance reduction.
  arXiv  Detail & Related papers  (2026-06-25T15:39:19Z)
• Advantage Collapse in Group Relative Policy Optimization: Diagnosis and Mitigation [13.272542054938258]
  We introduce the Advantage Collapse Rate (ACR), the first metric quantifying the proportion of training batches with ineffective gradients.<n>We then propose Adaptive Virtual Sample Policy Optimization (AVSPO), a lightweight extension of GRPO that injects virtual reward samples.<n>AVSPO reduces advantage collapse by 58-63% relative to GRPO and yields consistent accuracy gains of 4-6 percentage points across all model scales.
  arXiv  Detail & Related papers  (2026-05-20T12:57:37Z)
• REAL: Regression-Aware Reinforcement Learning for LLM-as-a-Judge [83.2858110368572]
  We propose textbfREAL (underlineREgression-underlineAware Reinforcement underlineLThought), a principled RL framework designed to optimize regression rewards.<n>We show that REAL consistently outperforms both regression-aware SFT baselines and standard RL methods.
  arXiv  Detail & Related papers  (2026-03-17T21:19:08Z)
• Unbiased Dynamic Pruning for Efficient Group-Based Policy Optimization [60.87651283510059]
  Group Relative Policy Optimization (GRPO) effectively scales LLM reasoning but incurs prohibitive computational costs.<n>We propose Dynamic Pruning Policy Optimization (DPPO), a framework that enables dynamic pruning while preserving unbiased gradient estimation.<n>To mitigate the data sparsity induced by pruning, we introduce Dense Prompt Packing, a window-based greedy strategy.
  arXiv  Detail & Related papers  (2026-03-04T14:48:53Z)
• GradAlign: Gradient-Aligned Data Selection for LLM Reinforcement Learning [55.03441672267886]
  We propose GradAlign, a gradient-aligned data selection method for reinforcement learning.<n>We evaluate GradAlign across three data regimes: unreliable reward signals, distribution imbalance, and low-utility training corpus.
  arXiv  Detail & Related papers  (2026-02-25T01:54:50Z)
• Stein-Rule Shrinkage for Stochastic Gradient Estimation in High Dimensions [0.0]
  gradient methods are central to large-scale learning, but they treat mini-batch gradients as unbiased estimators, which classical decision theory shows are inadmissible in high dimensions.<n>We introduce a framework based on Stein-rule shrinkage and construct a gradient estimator that adaptively contracts mini-batch gradients toward a stable estimator derived from historical momentum.<n> Empirical evaluations on CIFAR10 and CIFAR100 show consistent improvements over Adam in the large-batch regime.
  arXiv  Detail & Related papers  (2026-02-02T08:01:13Z)
• Alternating Reinforcement Learning for Rubric-Based Reward Modeling in Non-Verifiable LLM Post-Training [29.56905427210088]
  gradient-ARM is a framework that jointly optimize a rubric generator and a judge using reinforcement learning from preference feedback.<n>We show that gradient-ARM achieves state-of-the-art performance among baselines on benchmarks and significantly improves downstream policy alignment in both offline and online reinforcement learning settings.
  arXiv  Detail & Related papers  (2026-02-02T00:50:53Z)
• Low-Rank Curvature for Zeroth-Order Optimization in LLM Fine-Tuning [8.349781300731225]
  We introduce LOREN, a curvature-aware zeroth-order (ZO) optimization method for fine-tuning large language models (LLMs)<n>Existing ZO methods, which estimate gradients via finite differences using random perturbations, often suffer from high variance and suboptimal search directions.<n>Our approach addresses these challenges by: (i) adaptively estimating an anisotropic perturbation distribution for gradient estimation, (ii) capturing curvature through a low-rank block diagonal preconditioner, and (iii) applying a REINFORCE leave-one-out (RLOO) gradient estimator to reduce variance.
  arXiv  Detail & Related papers  (2025-11-11T08:34:09Z)
• Margin Adaptive DPO: Leveraging Reward Model for Granular Control in Preference Optimization [0.0]
  Margin-Adaptive Direct Preference Optimization provides a stable, data-preserving, and instance-level solution.<n>We provide a comprehensive theoretical analysis, proving that MADPO has a well-behaved optimization landscape.<n>It achieves performance gains of up to +33.3% on High Quality data and +10.5% on Low Quality data over the next-best method.
  arXiv  Detail & Related papers  (2025-10-06T20:09:37Z)
• From Noisy Traces to Stable Gradients: Bias-Variance Optimized Preference Optimization for Aligning Large Reasoning Models [90.45197506653341]
  Large reasoning models generate intermediate reasoning traces before producing final answers.<n> aligning LRMs with human preferences, a crucial prerequisite for model deployment, remains underexplored.<n>A common workaround optimized a single sampled trajectory, which introduces substantial gradient variance from trace sampling.
  arXiv  Detail & Related papers  (2025-10-06T17:58:01Z)
• FedHL: Federated Learning for Heterogeneous Low-Rank Adaptation via Unbiased Aggregation [6.5370850242187855]
  Federated Learning (FL) facilitates the fine-tuning of Foundation Models (FMs) using distributed data sources.<n>Low-Rank Adaptation (LoRA) gaining popularity due to its low communication costs and strong performance.<n>Existing methods lack formal convergence guarantees due to parameter truncation and biased gradient updates.
  arXiv  Detail & Related papers  (2025-05-24T04:12:12Z)
• ALoRE: Efficient Visual Adaptation via Aggregating Low Rank Experts [71.91042186338163]
  ALoRE is a novel PETL method that reuses the hypercomplex parameterized space constructed by Kronecker product to Aggregate Low Rank Experts.<n>Thanks to the artful design, ALoRE maintains negligible extra parameters and can be effortlessly merged into the frozen backbone.
  arXiv  Detail & Related papers  (2024-12-11T12:31:30Z)
• Fine-Tuning Language Models with Reward Learning on Policy [68.70065254564642]
  Reinforcement learning from human feedback (RLHF) has emerged as an effective approach to aligning large language models (LLMs) to human preferences. Despite its popularity, (fixed) reward models may suffer from inaccurate off-distribution. We propose reward learning on policy (RLP), an unsupervised framework that refines a reward model using policy samples to keep it on-distribution.
  arXiv  Detail & Related papers  (2024-03-28T10:02:10Z)
• Non-asymptotic Analysis of Biased Adaptive Stochastic Approximation [3.328448170090945]
  Gradient Descent (SGD) with adaptive steps is widely used to train deep neural networks and generative models.<n>This paper provides a comprehensive analysis of the effect of bias on gradient functions.
  arXiv  Detail & Related papers  (2024-02-05T10:17:36Z)
• Model-Based Reparameterization Policy Gradient Methods: Theory and Practical Algorithms [88.74308282658133]
  Reization (RP) Policy Gradient Methods (PGMs) have been widely adopted for continuous control tasks in robotics and computer graphics. Recent studies have revealed that, when applied to long-term reinforcement learning problems, model-based RP PGMs may experience chaotic and non-smooth optimization landscapes. We propose a spectral normalization method to mitigate the exploding variance issue caused by long model unrolls.
  arXiv  Detail & Related papers  (2023-10-30T18:43:21Z)
• Adaptive Gradient Method with Resilience and Momentum [120.83046824742455]
  We propose an Adaptive Gradient Method with Resilience and Momentum (AdaRem) AdaRem adjusts the parameter-wise learning rate according to whether the direction of one parameter changes in the past is aligned with the direction of the current gradient. Our method outperforms previous adaptive learning rate-based algorithms in terms of the training speed and the test error.
  arXiv  Detail & Related papers  (2020-10-21T14:49:00Z)
• Extrapolation for Large-batch Training in Deep Learning [72.61259487233214]
  We show that a host of variations can be covered in a unified framework that we propose. We prove the convergence of this novel scheme and rigorously evaluate its empirical performance on ResNet, LSTM, and Transformer.
  arXiv  Detail & Related papers  (2020-06-10T08:22:41Z)

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.