LoRA-TSD: Tangent-Space Spectral Descent for LoRA via Muon-Style Updates

Low-rank adaptation (LoRA) is the standard way to fine-tune large models, yet when its two factors are trained independently, the update ignores the geometry of the low-rank weight change it induces. We introduce LoRA-TSD, an optimizer that treats every LoRA step as a tangent vector of the fixed-rank matrix manifold and takes the spectral-norm steepest-descent step of Muon inside that tangent space, mapping the result back to the factors through a retraction native to the LoRA parametrization. The step avoids expensive operations on full weight matrices, and its retraction is up to $2.8\times$ cheaper than the truncated-SVD retraction used by prior manifold methods. We prove that the Frobenius-norm version of our surrogate recovers LoRA-Pro, and we identify the tangent-projected gradient, the Riemannian gradient of the manifold, as the stationarity measure natural to LoRA training and computable from the factor gradients alone. Under this measure we give the first global convergence guarantees for both LoRA-Pro and LoRA-TSD, with rates that drive the factor-gradient norms to zero. Across six commonsense and natural-language-inference benchmarks with Llama-3.2-1B, Llama-3.1-8B and Qwen3-32B, LoRA-TSD outperforms every competing LoRA optimizer and stays robust to the adapter rank. Code is available at https://github.com/brain-lab-research/LoRA-TSD.

Breaking the Structural Identity: Personalized Federated LoRA Fine-tuning under Rank Heterogeneity

Large Language Models (LLMs) have achieved remarkable success across diverse domains, but their adaptation to privacy-sensitive, distributed datasets remains a challenge. While Federated Learning (FL) combined with Low-Rank Adaptation (LoRA) provides a resource-efficient paradigm for collaborative fine-tuning, practical deployments are hindered by the dual challenges of resource heterogeneity and data heterogeneity. Existing rank-heterogeneous methods primarily focus on bridging dimension mismatches for aggregation but typically provide a unified global model for all clients sharing the same rank, failing to capture client-specific features in non-IID scenarios. In this paper, we propose FedRoRA (Federated Rank-wise Personalized LoRA), a novel framework that enables fine-grained personalization within rank-heterogeneous federations. FedRoRA decouples adaptation into shared global directions and personalized rank-wise magnitudes governed by learnable diagonal scales. On the server side, it extracts a global subspace via singular value decomposition (SVD) and redistributes client-specific initializations through a personalized projection and top-$k$ selection mechanism. Extensive experiments on NLU and NLG benchmarks demonstrate that FedRoRA consistently outperforms state-of-the-art methods.

Fine-Tuning of Transformer models with Frames

Parameter-Efficient Fine-Tuning (PEFT) strategies such as Low-Rank Adaptation (LoRA) are effective solutions for fine-tuning large-scale pre-trained models; however, their memory requirements scale with the size of the model, $\mathcal{O}(dr)$, where $d$ is the model's hidden dimension and $r$ is the rank. Our proposal, FrameFT, models the parameter update $ΔW$ with a sparse coefficient matrix in a Fusion Frame basis. Fusion Frames can be generated algorithmically and shared across model layers, enabling very efficient updates. Only the sparse coefficients of the basis expansion are stored/optimized, reducing the memory footprint. The sparse structure of the coefficient matrix in FrameFT and the sparsity in the Fusion Frames give large compute benefits, and our analysis provides formal convergence results. We evaluate the idea across a suite of supervised fine-tuning benchmarks, focusing on language tasks, but also report application to vision models. Our experiments show that FrameFT achieves performance on par with/exceeding state-of-the-art PEFT techniques, but needs far fewer trainable parameters.

Low-Rank Ternary Adaptation for Fine-Tuning Transformers

Ternary transformers offer extreme memory and compute efficiency, but existing low-bit LoRA-based methods cannot directly fine-tune ternary weights. Current approaches either require dequantization, restoring low-bit base weights to higher precision to merge with adaptation weight, or update only quantization parameters, preventing a merged model that remains ternary. We propose ternary multiplicative adaptation, which represents discrete updates of ternary weights such as sign flips or zeroing through a low-rank Kronecker factorization into two small ternary matrices applied element-wise to ternary weights. This design is parameter-efficient and expressive, preserves the ternary domain, and supports direct merging without dequantization. Experiments on six models across language and vision, including ternarized LLaMA-3 1B and 3B and a ternary ViT-B/16, demonstrate that our method recovers much of the performance lost to quantization and outperforms strong low-bit and ternary baselines. Code is available at https://github.com/alexmanoo/ternary_adaptation.

SplitLite: Low-Rank Residual Compression for Split Learning

Federated fine-tuning of on-device large language models (LLMs) faces a significant computing burden. To overcome this limitation, split learning (SL) has emerged as a promising solution, which offloads the primary training workload to a powerful server. However, SL requires exchanging high-dimensional activations and gradients between clients and the server, resulting in prohibitive communication costs. To overcome this challenge, we propose SplitLite, a communication-efficient split federated LoRA fine-tuning method that exploits the low effective rank structure of consecutive-epoch activation and gradient residuals. Our key finding is that, when LoRA uses rank $r$ updates in parameter space, the activation and gradient residuals of the same data sample between adjacent epochs also exhibit effective rank-$2r$ and rank-$4r$ structures, respectively. By revealing this property, SplitLite transmits only quantized truncated singular value decomposition (SVD) residual factors, thereby significantly reducing both activation uplink and gradient downlink traffic. Extensive experiments on the GLUE benchmark across a series of advanced on-device LLMs demonstrate that our method reduces activation uplink communication costs by up to 93.5\% and total communication costs by up to 83.7\%, without performance degradation.

FedPA-LoRA: Product-Aligned Framework for Mitigating Aggregation and Initialization Errors in Heterogeneous Federated LoRA

Low-Rank Adaptation (LoRA) enables efficient federated fine-tuning of large language models, but its factorized parameterization creates a tension between accurate aggregation of local updates and continuity of locally optimized factors. Factor-wise aggregation incurs aggregation mismatch but better preserves factor continuity, whereas product-space reconstruction reduces this mismatch at the cost of greater factor-level initialization mismatch from newly reconstructed factors. We propose FedPA-LoRA, a product-aligned federated LoRA framework that jointly addresses these limitations and provably converges under both homogeneous and heterogeneous client ranks. Each client preserves its local factors across communication rounds and aligns its product toward a rank-specific global reference, maintaining local optimization continuity while promoting global consistency under data heterogeneity. The server aggregates heterogeneous-rank updates in the common product space and efficiently reconstructs a rank-constrained global adapter without forming the dense aggregate. This design supports client-specific computation and communication budgets. Experiments on natural language understanding and generation tasks show that FedPA-LoRA consistently outperforms representative baselines across varying levels of data heterogeneity and homogeneous- and heterogeneous-rank settings, with up to a $6.82$ percentage-point improvement in average GLUE accuracy under heterogeneous client ranks.

Approximate Muon with low-rank adapters

The Muon optimizer shows clear benefits versus alternatives when pretraining neural networks. However, it is used less frequently for parameter-efficient fine-tuning (PEFT). One potential reason is that the most common PEFT method, LoRA, does not naturally combine with Muon since it is not mathematically possible to orthogonalize the weight update given by a low-rank parameterization. In this paper, we address this issue by approximating the solution to a relaxed Muon objective in the low-rank setting via linearization and then least-squares. We provide an efficient implementation that uses matmul operations only, as opposed to more complex linear algebra decomposition routines. Our method, sMuon (small Muon), performs favourably across SFT and a ReLoRA pretraining experiment. While results are model- and eval-dependent, we find overall that using Muon for low-rank fine-tuning provides moderate performance improvements.

Router Sensitivity Under Lightweight Fine-Tuning Identifies Prunable Experts in Mixture-of-Experts Models

Mixture-of-Experts (MoE) models decouple total parameters from per-token compute, but deployment still requires storing every expert. Recent theory shows that pruning experts with the smallest router-norm changes during fine-tuning can preserve accuracy, but assumes full fine-tuning. We test whether lightweight adaptation can recover this signal. We briefly fine-tune with a parameter-efficient adapter, rank experts by the induced $\ell_2$ router change, and prune the least-changed experts in one shot. On Mixtral-8$\times$7B-Instruct (44.83% MMLU-Pro), router-only LoRA trains 0.002% of parameters and outperforms all-module LoRA at matched rank with half the experts removed (27.54% vs. 24.42%); signal quality declines as adaptation spreads to attention and expert weights. Accuracy improves monotonically with LoRA rank, reaching 28.76%. IA3, which leaves router weights frozen, matches direct router adaptation, whereas unconstrained additive adapters degrade the signal. Router-guided MMLU-Pro accuracy decays quasi-linearly rather than collapsing, remains nearly 1.8 times that of magnitude-based or random pruning at maximal compression, and reduces memory by 49% and per-token latency by 37%. At 25% compression, retention is competitive with methods using full activation statistics. The criterion also transfers to Qwen1.5-MoE fine-tuned for mathematics, retaining 49.7% mean accuracy over eleven benchmarks with half the experts removed while random pruning falls to single digits. Router sensitivity under lightweight fine-tuning therefore makes provably motivated expert pruning practical at scale.

DeltaServe: Host-Agnostic Co-Serving of Inference and Fine-Tuning for LLMs

LLM serving systems are provisioned for peak load to meet strict latency targets, leaving substantial GPU compute idle whenever traffic falls below peak. We present DeltaServe, a host-agnostic co-serving design that converts this idle inference capacity into LoRA fine-tuning throughput while preserving inference service-level objectives (SLOs). DeltaServe integrates with existing inference engines through a compact hook interface that requires only multi-LoRA batching support. It exploits the shared execution structure of inference prefill and LoRA fine-tuning forward passes, and uses an SLO-aware scheduler to admit and execute fine-tuning only when sufficient inference headroom is available. The scheduler is driven by a CUDA-graph-aware latency model calibrated offline and refined online. We integrate DeltaServe with vLLM, SGLang, and S-LoRA. On a production trace from Company X, DeltaServe on vLLM delivers 2.9x higher fine-tuning throughput than LLMStation at 100% inference SLO compliance, versus 85% for LLMStation. It also achieves 39% higher fine-tuning throughput than a baseline running vLLM+torchtune, using no additional hardware and maintaining full SLO compliance.

How Small Can You Go? A Controlled Study of LoRA Rank, Target Modules, and Quantization Trade-offs for Text-to-SQL on a 60M-Parameter Model

Parameter-efficient fine-tuning (PEFT) and low-bit quantization are now standard tools for adapting language models under tight compute budgets, yet their interaction is most often studied on billion-parameter models where the design space is expensive to explore. We ask a complementary question: on a specific, fully reproducible 60M-parameter encoder-decoder model (T5-small) and a single-table text-to-SQL benchmark (WikiSQL), how much task accuracy does each efficiency knob actually cost? We run a controlled, single-variable study over (i) LoRA rank r in {2, 4, 8, 16, 32}, (ii) the set of adapted modules, and (iii) numerical precision. We report task accuracy alongside system-level metrics including trainable parameters, peak training memory, inference latency, and throughput, and frame adaptation as a constrained trade-off rather than an accuracy-only objective. Our results show that LoRA with r=16 recovers within 11.6 percentage points of full fine-tuning accuracy (59.6% vs. 71.2% exact-match) while training fewer than 1% of parameters and consuming 31% less peak GPU memory. Within this setting, rank beyond r=16 yields no measurable accuracy gain. QLoRA with INT8 and NF4 quantization achieves comparable accuracy (52.8% and 53.2%) at dramatically lower memory cost (0.60 GB each), demonstrating a compelling trade-off for memory-constrained deployments. All code, configurations, and logs are released for full reproducibility.

PoLoRA: A Preconditioned Orthogonalized LoRA Optimizer

Low-rank adaptation (LoRA) makes finetuning large language models cheaper by adding to each weight matrix a trainable low-rank update parameterized as the product of two matrices. These matrices are usually trained with Adam, which treats them as a single flat vector of parameters and ignores both the matrix and product structure of LoRA. Applying a matrix-aware optimizer such as Muon to each factor does not consistently improve over Adam, and neither do the product-aware Muon variants proposed in concurrent works. To realize consistent gains, we introduce PoLoRA, a Preconditioned Orthogonalized LoRA optimizer built from three ingredients: a product-aware spectral update direction, curvature preconditioning derived from controlling the per-sample loss change, and a magnitude rule that controls the sizes of both the factor and merged updates. We evaluate PoLoRA on instruction-tuning datasets for code and math across models from 1B to 8B parameters, and find that it reaches the final held-out loss achieved by tuned Adam in 1.2-1.7 times fewer steps, while adding at most 3% per-step overhead. Compared to Adam, PoLoRA is also less sensitive to the learning rate, and its optimal learning rate is stable across ranks.

LoRA-Muon: Spectral Steepest Descent on the Low-Rank Manifold

Low-Rank Adaptation (LoRA) significantly reduces compute and memory costs for finetuning Deep Learning models but is often harder to tune than dense training: when using factor-wise optimizers such as AdamW, it is sensitive to initialization choices, its optimal learning rates transfer poorly across ranks, and it often fails to beat dense baselines. We derive LoRA-Muon by applying the Muon optimizer's spectral steepest-descent rule to the low-rank setting. Along with our split weight-decay rule, our main claim is that LoRA-Muon is a good low-rank proxy for full-rank Muon and Shampoo-family optimizers. Its optimal learning rates transfer across rank, width, depth, and factor-rescaling. In our compute-matched TinyShakespeare study, a rank-$2$ proxy recovers the dense best tested learning rate, and a rank-$32$ LoRA-Muon run attains lower mean validation loss than the dense baseline in the seed-averaged sweep. We further show that the Spectron optimizer depends on arbitrary factor scaling, so it would likely be a poor fit when finetuning starts from badly imbalanced factors, and that LoRA-RITE's simplified QR-coordinate core implements the same spectral update. LoRA-Muon computes that update without QR-decomposition and avoids storing second moments, making it more accelerator-friendly and memory-efficient.

TriSP: Tri-Signal Structured Pruning for Large Language Models

Large language models (LLMs) achieve strong performance across diverse tasks but their deployment is constrained by the memory and compute cost of their parameters. Structured pruning addresses this by removing entire structures such as attention heads and Multi-Layer Perceptron (MLP) neurons to produce smaller dense models that run efficiently on standard hardware. However, existing methods rely on either gradient-based importance estimation, which is memory-prohibitive, or activation-based statistical proxies, which do not directly measure the effect of removal on the loss. Furthermore, the interaction between the importance criterion and the post-pruning recovery strategy has not been systematically studied. We propose TriSP (Tri-Signal Structured Pruning), an importance metric that combines weight magnitude scaled by activation norm with first-order gradient sensitivity via a geometric mean, producing a channel-level score that captures both structural and loss-sensitivity signals. Combined with adaptive per-layer budget allocation and low-rank adaptation (LoRA) recovery, TriSP achieves the lowest perplexity and highest zero-shot accuracy across all tested configurations, reaching 6.80 WikiText-2 perplexity at 20% pruning on LLaMA-7B. Inference throughput improves by 82% at 50% pruning, while still maintaining competitive performance.

Amortizing Federated Adaptation: Hypernetwork Driven LoRA for Personalized Foundation Models

Federated fine-tuning of foundation models using Low-Rank Adaptation (LoRA) offers a communication efficient solution for distributed learning. However, existing federated LoRA methods suffer from two fundamental limitations: (1) structural aggregation bias, where independently averaging low rank factors fails to approximate the true combined update, and (2) client side initialization lag, as clients repeatedly reinitialize LoRA parameters across communication rounds, slowing convergence. We propose HyperLoRA, a unified framework that addresses both issues through amortized federated adaptation through hypernetwork-driven LoRA generation and product space aggregation. Instead of iterative per-client optimization, HyperLoRA employs a learned generator that maps client distribution signatures to LoRA initializations, effectively amortizing per client adaptation. On the server side, we introduce a learned aggregation module that directly synthesizes updates in the low-rank product space, eliminating the inconsistencies of factor-wise averaging. A lightweight residual correction module further improves stability under heterogenous (non-IID) client distributions.By replacing iterative optimization and heuristic averaging with learned operators, HyperLoRA jointly enables efficient personalization, unbiased aggregation, and faster convergence. Experiments on federated vision and vision-language benchmarks show that HyperLoRA achieves improved convergence speed, greater robustness to distribution shift, and stronger personalization performance compared to prior federated LoRA methods.