# peft

PEFT is a library and framework for parameter-efficient fine-tuning and adapting large pretrained models by training only a small number of additional parameters.

This is one page of public article previews, not the complete archive. Follow Next page to continue. Summaries are not the original full articles.

## SPARK: Adaptive Parametric Knowledge Participation for Faithful Retrieval-Augmented Generation

DevFeed: [SPARK: Adaptive Parametric Knowledge Participation for Faithful Retrieval-Augmented Generation](<https://devfeed.tech/articles/spark-adaptive-parametric-knowledge-participation-for-faithful-retrieval-augmented-generation-145353.md>)

Original publisher: [Read original article](<https://arxiv.org/abs/2606.27786>)

Author: Ruochang Li, Pengcheng Huang, Zhenghao Liu, Yukun Yan, Huiyuan Xie, Yu Gu, Ge Yu, Maosong Sun

Published: 2026-11-20T05:00:00Z

Content type: article

Language: en

Sources: [arXiv: Artificial Intelligence](<https://devfeed.tech/sources/arxiv-artificial-intelligence.md>)

Topics: [Retrieval Augmented Generation (RAG)](<https://devfeed.tech/topics/retrieval-augmented-generation-rag.md>), [Large Language Model](<https://devfeed.tech/topics/llm.md>), [peft](<https://devfeed.tech/topics/peft.md>), [Enhancing AI model performance](<https://devfeed.tech/topics/enhancing-ai-model-performance.md>)

Tags: [adaptive](<https://devfeed.tech/tags/adaptive.md>), [cs-ai](<https://devfeed.tech/tags/cs-ai.md>), [cs-cl](<https://devfeed.tech/tags/cs-cl.md>), [faithfulness](<https://devfeed.tech/tags/faithfulness.md>), [generation](<https://devfeed.tech/tags/generation.md>), [knowledge](<https://devfeed.tech/tags/knowledge.md>), [large-language-models-llms](<https://devfeed.tech/tags/large-language-models-llms.md>), [retrieval](<https://devfeed.tech/tags/retrieval.md>), [retrieval-augmented-generation](<https://devfeed.tech/tags/retrieval-augmented-generation.md>), [spark](<https://devfeed.tech/tags/spark.md>)

### AI overview

SPARK addresses conflicts between retrieved evidence and an LLM's internal knowledge by learning a lightweight policy that dynamically adjusts the contribution of feed-forward-network-mediated parametric knowledge during generation. It keeps the backbone model frozen and optimizes the policy with a faithfulness-oriented reward, avoiding direct edits to the model's stored knowledge. The abstract reports experiments on six datasets, with 0.01% additional trainable parameters, and says SPARK achieved more faithful generation than competitive baselines while preserving the backbone's general capabilities.

### Source excerpt

arXiv:2606.27786v3 Announce Type: replace-cross Abstract: Retrieval-augmented generation (RAG) enhances the faithfulness of large language models (LLMs) by incorporating external knowledge into the generation process. However, conflicts between retrieved context and parametric knowledge can cause LLMs to rely on inconsistent internal knowledge, hindering their ability to faithfully leverage retrieved evidence. Existing approaches mitigate this issue by editing parametric knowledge to resolve such conflicts. However, locating and modifying knowledge encoded in model parameters is challenging and may inadvertently compromise the backbone LLM's capabilities. In this paper, we introduce SPARK, a novel framework that formulates parametric knowledge control as a latent participation policy learning problem, enabling LLMs to adaptively regulate the contribution of parametric knowledge during generation. Specifically, SPARK introduces a lightweight participation policy that dynamically modulates the contribution of FFN-mediated parametric knowledge while keeping the backbone LLM frozen. The participation policy is optimized with a faithfulness-oriented reward, enabling fine-grained control over parametric knowledge participation during generation. Extensive experiments on six datasets validate the effectiveness of SPARK, which introduces and optimizes only 0.01\% additional trainable parameters while keeping the backbone LLM fully frozen, thereby preserving its general capabilities and achieving more faithful generation than competitive baselines. All datasets and code are available at https://github.com/OpenBMB/SPARK.

## Parameter-Efficient Adaptation of Self-supervised Vision Backbones Survives Distribution Shift in Radar Micro-Doppler HAR

DevFeed: [Parameter-Efficient Adaptation of Self-supervised Vision Backbones Survives Distribution Shift in Radar Micro-Doppler HAR](<https://devfeed.tech/articles/parameter-efficient-adaptation-of-self-supervised-vision-backbones-survives-distribution-shift-in-radar-micro-doppler-har-145113.md>)

Original publisher: [Read original article](<https://arxiv.org/abs/2611.25874>)

Author: Keryan Chelouche (ETIS UMR 8051, CY Cergy Paris Universit\\'e, ENSEA, CNRS, Cergy, France, Fondation L\\'eonie Chaptal, Sarcelles, France), Petr Dobias (ETIS UMR 8051, CY Cergy Paris Universit\\'e, ENSEA

Published: 2026-11-20T05:00:00Z

Content type: article

Language: en

Sources: [arXiv: Machine Learning](<https://devfeed.tech/sources/arxiv-machine-learning.md>)

Topics: [peft](<https://devfeed.tech/topics/peft.md>), [few-shot](<https://devfeed.tech/topics/few-shot.md>), [AI Foundation Models](<https://devfeed.tech/topics/ai-foundation-models.md>)

Tags: [adapter](<https://devfeed.tech/tags/adapter.md>), [application](<https://devfeed.tech/tags/application.md>), [cs-lg](<https://devfeed.tech/tags/cs-lg.md>), [distribution](<https://devfeed.tech/tags/distribution.md>), [doppler](<https://devfeed.tech/tags/doppler.md>), [few-shot](<https://devfeed.tech/tags/few-shot.md>), [fewer](<https://devfeed.tech/tags/fewer.md>), [fine-tuning](<https://devfeed.tech/tags/fine-tuning.md>), [foundation](<https://devfeed.tech/tags/foundation.md>), [image](<https://devfeed.tech/tags/image.md>), [improvements](<https://devfeed.tech/tags/improvements.md>), [layer](<https://devfeed.tech/tags/layer.md>), [orientation](<https://devfeed.tech/tags/orientation.md>), [parameter](<https://devfeed.tech/tags/parameter.md>), [peft](<https://devfeed.tech/tags/peft.md>), [pretraining](<https://devfeed.tech/tags/pretraining.md>), [probing](<https://devfeed.tech/tags/probing.md>), [protocol](<https://devfeed.tech/tags/protocol.md>), [radar](<https://devfeed.tech/tags/radar.md>), [shift](<https://devfeed.tech/tags/shift.md>), [vision](<https://devfeed.tech/tags/vision.md>)

### AI overview

In the reported radar micro-Doppler HAR evaluations, parameter-efficient fine-tuning (PEFT) avoided two failures seen with alternatives: full fine-tuning hurt cross-dataset transfer, while frozen features under-adapted across orientations. DINOv3 adapted with PiSSA ranked first among PEFT methods in every tested setting. Its Small variant matched the prior Base-size PEFT state of the art in in-distribution accuracy with 4.5 times fewer total parameters and only 3% trainable; it also beat a size-matched baseline across evaluated settings, with transfer gains of 2.1, 6.6, and 8.9 percentage points as shift severity increased. Ablations credited gains to both pretraining and adapter design, with pretraining contributing most under compound shift.

### Source excerpt

arXiv:2611.25874v1 Announce Type: new Abstract: Micro-Doppler radar enables contactless, privacy-preserving human activity recognition (HAR), but deployment robustness remains underexplored. We compared full fine-tuning, frozen probing, and parameter-efficient fine-tuning (PEFT) using subject-disjoint in-distribution cross-validation and few-shot cross-age, cross-orientation, and cross-dataset transfer. Full fine-tuning suffered negative cross-dataset transfer, while frozen features under-adapted on cross-orientation. Only PEFT avoided both failures. We paired DINOv3 with all-layer principal-singular-direction adaptation (PiSSA), to our knowledge the first application of a self-supervised image foundation backbone to radar micro-Doppler HAR. DINOv3-PiSSA ranked first among PEFT methods in every evaluated setting. Under our protocol, its Small variant matched the in-distribution accuracy of the prior Base-size PEFT state of the art using 4.5x fewer total parameters, with only 3% trainable. It surpassed a size-matched variant of that baseline in every evaluated setting, with transfer gains increasing with shift severity: 2.1, 6.6, and 8.9 percentage points, respectively. Ablations attributed improvements to both pretraining and adapter recipe, with the pretraining contributing most under compound shift. Code and protocol are available at https://github.com/KeryanChelouche/udoppler-ssl.

## Can you make models believe false facts in the prompt?

DevFeed: [Can you make models believe false facts in the prompt?](<https://devfeed.tech/articles/can-you-make-models-believe-false-facts-in-the-prompt-142727.md>)

Original publisher: [Read original article](<https://arxiv.org/abs/2611.23938>)

Author: Ananya Sutradhar, Fabien Roger

Published: 2026-11-19T05:00:00Z

Content type: article

Language: en

Sources: [arXiv: Artificial Intelligence](<https://devfeed.tech/sources/arxiv-artificial-intelligence.md>)

Topics: [peft](<https://devfeed.tech/topics/peft.md>), [ai alignment](<https://devfeed.tech/topics/ai-alignment.md>), [prompt](<https://devfeed.tech/topics/prompt.md>), [generalization in machine learning](<https://devfeed.tech/topics/generalization-in-machine-learning.md>)

Tags: [alignment](<https://devfeed.tech/tags/alignment.md>), [audits](<https://devfeed.tech/tags/audits.md>), [cs-ai](<https://devfeed.tech/tags/cs-ai.md>), [evaluations](<https://devfeed.tech/tags/evaluations.md>), [experiments](<https://devfeed.tech/tags/experiments.md>), [finetuning](<https://devfeed.tech/tags/finetuning.md>), [jacobian](<https://devfeed.tech/tags/jacobian.md>), [llms](<https://devfeed.tech/tags/llms.md>), [models](<https://devfeed.tech/tags/models.md>), [prompt](<https://devfeed.tech/tags/prompt.md>), [rl](<https://devfeed.tech/tags/rl.md>), [scale](<https://devfeed.tech/tags/scale.md>), [sdf](<https://devfeed.tech/tags/sdf.md>), [small-models](<https://devfeed.tech/tags/small-models.md>), [synthetic](<https://devfeed.tech/tags/synthetic.md>), [system](<https://devfeed.tech/tags/system.md>), [training](<https://devfeed.tech/tags/training.md>)

### AI overview

Prompt-conditioned Synthetic Document Finetuning trains one model on synthetic documents covering thousands of facts, conditioning each document on a fact placed in a <facts> prompt tag. The authors report that belief depth on unseen facts scales roughly log-linearly with the number of training facts, suggesting a potentially inexpensive way to scale false-belief evaluations. In preliminary alignment-audit experiments on the small model studied, results were negative because it often already acted as if it believed facts in ordinary honeypot system prompts. Adding RL to reduce verbalized compliance with prompt falsehoods generalized to held-out evaluations and nearly matched SDF on some evaluations. However, training against one belief-depth metric made that evaluation less trustworthy.

### Source excerpt

arXiv:2611.23938v1 Announce Type: new Abstract: Making LLMs believe false facts is useful for alignment evaluations and for building model organisms. Previous work explored doing so with Synthetic Document Finetuning (SDF). But SDF requires fine-tuning one new model for each set of injected false beliefs. To enable larger scale SDF-powered alignment evaluations, we attempt to produce a single model that believes any fact placed in a tag in the prompt. To do so, we explore doing prompt-conditioned SDF: fine-tuning a model on synthetic documents covering thousands of facts, with each document conditioned on a fact presented in the prompt. We evaluate how deeply the model believes the facts provided in the prompt using evaluations from prior work and a new Jacobian lens evaluation. We find that the belief depth on unseen facts scales roughly log-linearly with the number of training facts, making this a cheap, promising avenue for making models believe false facts, though preliminary experiments on applying it to alignment audits on the small model we studied showed negative results due to these small models often already acting as if they believed facts provided in regular honeypot system prompts. Using RL to reduce the rate at which models verbalize following a false fact in the prompt, on top of prompt-conditioned SDF, generalizes to held-out evals and almost matches SDF on some evaluations. But training against one of our metrics for belief depth makes it harder to trust those evaluations.

## Swiss-Knife: A Framework for Reconfigurable Externalised Multi-Objective Alignment at Decode Time

DevFeed: [Swiss-Knife: A Framework for Reconfigurable Externalised Multi-Objective Alignment at Decode Time](<https://devfeed.tech/articles/swiss-knife-a-framework-for-reconfigurable-externalised-multi-objective-alignment-at-decode-time-143311.md>)

Original publisher: [Read original article](<https://arxiv.org/abs/2609.22226>)

Author: Agnibh Karmakar, Mayur Parvatikar, Shreyash Dhoot, Amit Dhanda, Aman Chadha, Kapil Wanaskar, Vinija Jain, Amitava Das

Published: 2026-11-19T05:00:00Z

Content type: article

Language: en

Sources: [arXiv: Machine Learning](<https://devfeed.tech/sources/arxiv-machine-learning.md>)

Topics: [ai alignment](<https://devfeed.tech/topics/ai-alignment.md>), [dpo](<https://devfeed.tech/topics/dpo.md>), [peft](<https://devfeed.tech/topics/peft.md>), [lora](<https://devfeed.tech/topics/lora.md>), [post-training](<https://devfeed.tech/topics/post-training.md>)

Tags: [adversarial](<https://devfeed.tech/tags/adversarial.md>), [aggregation](<https://devfeed.tech/tags/aggregation.md>), [alignment](<https://devfeed.tech/tags/alignment.md>), [arm](<https://devfeed.tech/tags/arm.md>), [arxiv](<https://devfeed.tech/tags/arxiv.md>), [batch](<https://devfeed.tech/tags/batch.md>), [best](<https://devfeed.tech/tags/best.md>), [candidate](<https://devfeed.tech/tags/candidate.md>), [comparison](<https://devfeed.tech/tags/comparison.md>), [contamination](<https://devfeed.tech/tags/contamination.md>), [costs](<https://devfeed.tech/tags/costs.md>), [cs-cl](<https://devfeed.tech/tags/cs-cl.md>), [cs-lg](<https://devfeed.tech/tags/cs-lg.md>), [decode](<https://devfeed.tech/tags/decode.md>), [design](<https://devfeed.tech/tags/design.md>), [dpo](<https://devfeed.tech/tags/dpo.md>), [external](<https://devfeed.tech/tags/external.md>), [f](<https://devfeed.tech/tags/f.md>), [family](<https://devfeed.tech/tags/family.md>), [form](<https://devfeed.tech/tags/form.md>), [framework](<https://devfeed.tech/tags/framework.md>), [frontier](<https://devfeed.tech/tags/frontier.md>), [stat-ml](<https://devfeed.tech/tags/stat-ml.md>)

### AI overview

Swiss-Knife treats decode-time alignment as a reconfigurable runtime specification combining hot-swappable scoring models, batch normalization, pairwise aggregation and a selection rule. Its representation theorem describes admissible aggregators as a two-parameter family that includes probit, logistic and pointwise argmax rules. The authors report that pairwise aggregation is stable under adversarial reward contamination, whereas argmax is not; Candidate-Batch Normalization (CBN) makes objective weights invariant to reward-model rescaling. In their DPO-LoRA and uncertainty-aware tournament instantiation, the system achieved harmonic F1 0.797 versus 0.750 for the strongest of six methods, with the lowest refusal rate and highest helpfulness reported. Objective reconfiguration took 0.05 ms without gradient computation. Removing CBN reduced F1 by 0.217; the authors attribute the collapse to a low-variance blade retaining only 9% of its nominal influence. These are claims from the supplied abstract excerpt.

### Source excerpt

arXiv:2609.22226v2 Announce Type: replace-cross Abstract: Decode-time alignment methods steer a frozen language model by scoring candidate continuations with an external reward and selecting the maximiser. We argue that this shared design is a single degenerate point in a much larger space. We introduce Swiss-Knife, a framework for externalised multi-objective alignment in which the alignment specification is a first-class runtime object: hot-swappable scoring blades, a batch normaliser, a pairwise aggregation operator, and a selection rule. Six axioms characterise the admissible aggregation operators, and we prove a representation theorem: every operator satisfying them has the form $R_i = \sum_{j \neq i} g((\mu_i - \mu_j)/s(\sigma_i,\sigma_j))$, a two-parameter family containing probit and logistic comparison rules and pointwise argmax as named coordinates. Within it, pairwise aggregation is Lipschitz-stable under adversarial reward contamination while argmax is not, and Candidate-Batch Normalization (CBN) makes the weight simplex invariant to the rescalings under which reward models are only ever identified. Our reference instantiation pairs DPO-LoRA blades with an uncertainty-aware pairwise tournament. Sweeping the helpfulness/honesty/harmlessness simplex, it attains the best balanced frontier of six decode-time methods (harmonic $F_1$ 0.797 vs. 0.750 for the strongest baseline, $p < 10^{-14}$) with the lowest refusal rate and highest helpfulness of any arm, and reconfigures its objectives in 0.05 ms with no gradient computation. Ablating CBN costs 0.217 $F_1$, and the metadata confirms the predicted mechanism: the lowest-variance blade retains 9% of its nominal 33% influence, and the collapse follows the predicted ordering.

## Full Fine-Tuning (FFT) on SageMaker AI Serverless Model Customization

DevFeed: [Full Fine-Tuning (FFT) on SageMaker AI Serverless Model Customization](<https://devfeed.tech/articles/full-fine-tuning-fft-on-sagemaker-ai-serverless-model-customization-141875.md>)

Original publisher: [Read original article](<https://aws.amazon.com/blogs/machine-learning/full-fine-tuning-fft-on-sagemaker-ai-serverless-model-customization/>)

Author: Sandeep Raveesh-Babu

Published: 2026-11-19T00:15:58Z

Content type: article

Language: en

Sources: [Artificial Intelligence](<https://devfeed.tech/sources/artificial-intelligence.md>)

Topics: [Fine-tuning](<https://devfeed.tech/topics/fine-tuning.md>), [lora](<https://devfeed.tech/topics/lora.md>), [Amazon SageMaker](<https://devfeed.tech/topics/amazon-sagemaker.md>), [LLM evaluation / benchmarking](<https://devfeed.tech/topics/llm-evaluation-benchmarking.md>), [Serverless](<https://devfeed.tech/topics/serverless.md>), [peft](<https://devfeed.tech/topics/peft.md>), [Hyperparameter optimization](<https://devfeed.tech/topics/hyperparameter-optimization.md>), [post-training](<https://devfeed.tech/topics/post-training.md>), [Large Language Model](<https://devfeed.tech/topics/llm.md>)

Tags: [advanced-300](<https://devfeed.tech/tags/advanced-300.md>), [ai](<https://devfeed.tech/tags/ai.md>), [amazon](<https://devfeed.tech/tags/amazon.md>), [amazon-sagemaker](<https://devfeed.tech/tags/amazon-sagemaker.md>), [amazon-sagemaker-ai](<https://devfeed.tech/tags/amazon-sagemaker-ai.md>), [announcements](<https://devfeed.tech/tags/announcements.md>), [benchmarks](<https://devfeed.tech/tags/benchmarks.md>), [customization](<https://devfeed.tech/tags/customization.md>), [fft](<https://devfeed.tech/tags/fft.md>), [fine](<https://devfeed.tech/tags/fine.md>), [fine-tuning](<https://devfeed.tech/tags/fine-tuning.md>), [full](<https://devfeed.tech/tags/full.md>), [gap](<https://devfeed.tech/tags/gap.md>), [guidelines](<https://devfeed.tech/tags/guidelines.md>), [how-to](<https://devfeed.tech/tags/how-to.md>), [key](<https://devfeed.tech/tags/key.md>), [lora](<https://devfeed.tech/tags/lora.md>), [matrices](<https://devfeed.tech/tags/matrices.md>), [model](<https://devfeed.tech/tags/model.md>), [post](<https://devfeed.tech/tags/post.md>), [practical](<https://devfeed.tech/tags/practical.md>), [serverless](<https://devfeed.tech/tags/serverless.md>)

### AI overview

The excerpt reports that Full Fine-Tuning (FFT) can outperform LoRA on complex code-generation tasks with large datasets, while results depend heavily on hyperparameters and FFT can cause catastrophic forgetting. In experiments, Qwen3-4B was fine-tuned on 110,000 Magicoder coding pairs using SageMaker AI Serverless Model Customization. At 100K+ examples, FFT reached 90% LLM-as-Judge correctness; a baseline with 5% warmup, batch size 8, and learning rate 2e-5 best preserved MMLU knowledge. Three epochs yielded the reported correctness peak, while training beyond that degraded tested variants. Aggressive FFT learning rates of 5e-5 or higher sharply reduced MATH and MMLU scores. LoRA slightly exceeded FFT on one LLM-as-Judge comparison (81.5% versus 79.5%), and both methods fell below the base model on MATH exact match. The service handles compute allocation and checkpoints; switching between FFT and LoRA requires changing the training type. These findings reflect the reported coding experiments and do not establish FFT as universally superior.

### Source excerpt

Amazon SageMaker AI Serverless Model Customization now supports Full Fine-Tuning (FFT). This post explores when FFT outperforms LoRA across key benchmarks, examines the spectral properties of model weight matrices behind the quality gap, shares practical hyperparameter tuning guidelines, and shows how to get started.

## SpecSearch: Towards Hardware-Aware Efficient Speculative Decoding

DevFeed: [SpecSearch: Towards Hardware-Aware Efficient Speculative Decoding](<https://devfeed.tech/articles/specsearch-towards-hardware-aware-efficient-speculative-decoding-140005.md>)

Original publisher: [Read original article](<https://arxiv.org/abs/2611.22296>)

Author: Selin Yildirim, Ali Zeyrek, Deming Chen

Published: 2026-11-18T05:00:00Z

Content type: article

Language: en

Sources: [arXiv: Artificial Intelligence](<https://devfeed.tech/sources/arxiv-artificial-intelligence.md>), [arXiv: Distributed Computing](<https://devfeed.tech/sources/arxiv-distributed-computing.md>)

Topics: [LLM optimization](<https://devfeed.tech/topics/llm-optimization.md>), [Large Language Model](<https://devfeed.tech/topics/llm.md>), [Model Optimizer](<https://devfeed.tech/topics/model-optimizer.md>), [vllm](<https://devfeed.tech/topics/vllm.md>), [peft](<https://devfeed.tech/topics/peft.md>), [lora](<https://devfeed.tech/topics/lora.md>), [LLM evaluation / benchmarking](<https://devfeed.tech/topics/llm-evaluation-benchmarking.md>)

Tags: [cs-ai](<https://devfeed.tech/tags/cs-ai.md>), [cs-dc](<https://devfeed.tech/tags/cs-dc.md>), [decoding](<https://devfeed.tech/tags/decoding.md>), [draft](<https://devfeed.tech/tags/draft.md>), [gpu](<https://devfeed.tech/tags/gpu.md>), [h100](<https://devfeed.tech/tags/h100.md>), [hardware](<https://devfeed.tech/tags/hardware.md>), [jev](<https://devfeed.tech/tags/jev.md>), [language-model](<https://devfeed.tech/tags/language-model.md>), [large-language-model](<https://devfeed.tech/tags/large-language-model.md>), [llm-inference](<https://devfeed.tech/tags/llm-inference.md>), [lora](<https://devfeed.tech/tags/lora.md>), [memory](<https://devfeed.tech/tags/memory.md>), [nvidia](<https://devfeed.tech/tags/nvidia.md>), [nvidia-rtx](<https://devfeed.tech/tags/nvidia-rtx.md>), [optimization](<https://devfeed.tech/tags/optimization.md>), [precision](<https://devfeed.tech/tags/precision.md>), [sd](<https://devfeed.tech/tags/sd.md>), [search](<https://devfeed.tech/tags/search.md>), [search-engine](<https://devfeed.tech/tags/search-engine.md>), [speculative-decoding](<https://devfeed.tech/tags/speculative-decoding.md>), [throughput](<https://devfeed.tech/tags/throughput.md>)

### AI overview

SpecSearch searches jointly over decoding strategy, draft length, and precision for each workload-GPU pair. Its analytical cost model accounts for memory and compute limits to narrow feasible configurations, and calibration measurements guide throughput optimization. A Jev-based predictor, trained by adapting frozen Qwen3-4B with LoRA adapters and a linear correction layer, selects configurations for held-out GPUs without deployment-GPU calibration. Across RTX 5090, H100, L40, and MI300X, fully calibrated SpecSearch averaged 2.01x autoregressive and 1.36x speculative-decoding throughput, with both baselines served by vLLM. The predictor achieved 1.30x throughput over vLLM speculative decoding and retained 97.36% of fully calibrated SpecSearch throughput.

### Source excerpt

arXiv:2611.22296v1 Announce Type: new Abstract: Speculative decoding (SD) accelerates large language model (LLM) inference, but its optimal configuration depends jointly on workload characteristics, GPU hardware, and available memory. For example, on the RTX~5090, the state-of-the-art speculative drafter DFlash achieves $3.50\times$ the throughput of autoregressive (AR) decoding at low concurrency, but only $0.76\times$ at high concurrency with short outputs. We present SpecSearch, a hardware-aware search engine that jointly optimizes decoding strategy, draft length, and precision across workload--GPU pairs. Its analytical cost model captures memory and compute constraints to identify feasible configurations and narrow the search space for throughput optimization. Guided by this model, calibration measurements are collected from multiple GPUs. Using the measurements, a Jev-based predictor learns to select high-throughput decoding configurations on held-out GPUs without requiring deployment-GPU calibration. The predictor is parameter-efficiently trained on calibration data, while adapting a frozen Qwen3-4B using LoRA adapters and a linear correction layer. Across NVIDIA RTX~5090, H100, L40 and AMD MI300X GPUs, SpecSearch with full calibration achieves an average of $2.01\times$ the throughput of AR and $1.36\times$ the throughput of SD, both served by vLLM. The proposed Jev-based predictor closely matches this search performance with $1.30\times$ throughput improvement over SD with vLLM, while eliminating the need for held-out GPU calibration. Additionally, it retains $97.36\%$ of the throughput achieved by SpecSearch that utilizes full calibration.

## FaceRiff: Data-Efficient Unified Text-to-3D Face Asset Synthesis via Geometric Conditioning

DevFeed: [FaceRiff: Data-Efficient Unified Text-to-3D Face Asset Synthesis via Geometric Conditioning](<https://devfeed.tech/articles/faceriff-data-efficient-unified-text-to-3d-face-asset-synthesis-via-geometric-conditioning-139922.md>)

Original publisher: [Read original article](<https://arxiv.org/abs/2611.21867>)

Author: Elly Akhoundi, Mathieu Lamarre, Thibault Goyallon, Patrick Anderson, \\'Etienne Danvoye

Published: 2026-11-18T05:00:00Z

Content type: article

Language: en

Sources: [arXiv: Artificial Intelligence](<https://devfeed.tech/sources/arxiv-artificial-intelligence.md>)

Topics: [3D](<https://devfeed.tech/topics/3d.md>), [peft](<https://devfeed.tech/topics/peft.md>), [stable-diffusion](<https://devfeed.tech/topics/stable-diffusion.md>), [Image processing](<https://devfeed.tech/topics/image-processing.md>), [multimodal-ai](<https://devfeed.tech/topics/multimodal-ai.md>)

Tags: [3d](<https://devfeed.tech/tags/3d.md>), [alignment](<https://devfeed.tech/tags/alignment.md>), [asset](<https://devfeed.tech/tags/asset.md>), [cs-ai](<https://devfeed.tech/tags/cs-ai.md>), [cs-cv](<https://devfeed.tech/tags/cs-cv.md>), [cs-gr](<https://devfeed.tech/tags/cs-gr.md>), [data](<https://devfeed.tech/tags/data.md>), [dataset](<https://devfeed.tech/tags/dataset.md>), [datasets](<https://devfeed.tech/tags/datasets.md>), [diffusion](<https://devfeed.tech/tags/diffusion.md>), [diffusion-models](<https://devfeed.tech/tags/diffusion-models.md>), [diversity](<https://devfeed.tech/tags/diversity.md>), [fixed](<https://devfeed.tech/tags/fixed.md>), [framework](<https://devfeed.tech/tags/framework.md>), [generation](<https://devfeed.tech/tags/generation.md>), [gpu](<https://devfeed.tech/tags/gpu.md>), [image](<https://devfeed.tech/tags/image.md>), [independent](<https://devfeed.tech/tags/independent.md>), [lighting](<https://devfeed.tech/tags/lighting.md>), [lora](<https://devfeed.tech/tags/lora.md>), [maps](<https://devfeed.tech/tags/maps.md>), [synthesis](<https://devfeed.tech/tags/synthesis.md>), [text](<https://devfeed.tech/tags/text.md>)

### AI overview

FaceRiff generates text-conditioned 3D face assets--mesh, albedo, and displacement--in a shared UV space, using a pretrained SDXL model with task-specific LoRA adapters and UV normal maps for geometric conditioning. It was trained on 2,049 scans from about 350 identities on a single GPU, updating less than 3% of backbone parameters per task, and produces assets without per-instance optimization. The abstract reports greater per-prompt diversity than FaceG2E and DreamFace while maintaining prompt alignment and fixed-topology UV consistency; it also generates lighting-independent albedo for PBR relighting and pore-scale displacement. In a blind study of 45 industry professionals, perceived realism was substantially higher than DreamFace and approached that of real captures.

### Source excerpt

arXiv:2611.21867v1 Announce Type: cross Abstract: High-resolution facial datasets with consistent UV-mapped geometry, albedo, and physically based displacement are scarce, yet such assets are essential for photorealistic digital-human rendering. Existing 3D face generation methods are often slow, single-modality, or require large specialized datasets. Meanwhile, large image diffusion models encode strong facial priors but do not natively produce lighting-independent textures, pore-scale displacement, or UV-aligned geometry. We present FaceRiff, a unified, data- and parameter-efficient framework for generating complete 3D face assets (mesh, albedo, and displacement) directly from text in a shared UV space. FaceRiff adapts a pretrained SDXL backbone with task-specific LoRA adapters and uses UV normal maps as geometric conditioning: per-instance normals guide albedo and displacement synthesis, while a dataset-mean normal provides a canonical spatial anchor for mesh generation. Using only 2,049 scans from approximately 350 identities and training on a single GPU, FaceRiff updates less than 3% of the backbone parameters per task and produces render-ready 3D face assets without per-instance optimization. Compared with FaceG2E and DreamFace, FaceRiff produces greater per-prompt diversity while preserving prompt alignment and fixed-topology UV consistency, and additionally generates lighting-independent albedo suitable for PBR relighting and pore-scale displacement absent from prior text-driven baselines. A blind study with 45 industry professionals shows that FaceRiff achieves substantially higher perceived realism than DreamFace, approaching that of real captures. Overall, FaceRiff demonstrates that pretrained diffusion priors can be adapted to unified 3D face asset synthesis under limited data.

## Fine-Tuning Gemma 4 Edge Models for MATLAB Signal-Processing Code with SFT and QLoRA on a Laptop iGPU

DevFeed: [Fine-Tuning Gemma 4 Edge Models for MATLAB Signal-Processing Code with SFT and QLoRA on a Laptop iGPU](<https://devfeed.tech/articles/fine-tuning-gemma-4-edge-models-for-matlab-signal-processing-code-with-sft-and-qlora-on-a-laptop-igpu-139620.md>)

Original publisher: [Read original article](<https://arxiv.org/abs/2611.23205>)

Author: Mehmet Altan Toks\\"oz

Published: 2026-11-18T05:00:00Z

Content type: article

Language: en

Sources: [arXiv: Artificial Intelligence](<https://devfeed.tech/sources/arxiv-artificial-intelligence.md>), [arXiv: Machine Learning](<https://devfeed.tech/sources/arxiv-machine-learning.md>), [arXiv: Software Engineering](<https://devfeed.tech/sources/arxiv-software-engineering.md>)

Topics: [Fine-tuning](<https://devfeed.tech/topics/fine-tuning.md>), [gemma](<https://devfeed.tech/topics/gemma.md>), [lora](<https://devfeed.tech/topics/lora.md>), [peft](<https://devfeed.tech/topics/peft.md>), [MATLAB](<https://devfeed.tech/topics/matlab.md>)

Tags: [code](<https://devfeed.tech/tags/code.md>), [cs-ai](<https://devfeed.tech/tags/cs-ai.md>), [cs-lg](<https://devfeed.tech/tags/cs-lg.md>), [cs-se](<https://devfeed.tech/tags/cs-se.md>), [digital-signal-processing](<https://devfeed.tech/tags/digital-signal-processing.md>), [dsp](<https://devfeed.tech/tags/dsp.md>), [edge](<https://devfeed.tech/tags/edge.md>), [eess-sp](<https://devfeed.tech/tags/eess-sp.md>), [fine-tuning](<https://devfeed.tech/tags/fine-tuning.md>), [gemma](<https://devfeed.tech/tags/gemma.md>), [gemma-4](<https://devfeed.tech/tags/gemma-4.md>), [igpu](<https://devfeed.tech/tags/igpu.md>), [laptop](<https://devfeed.tech/tags/laptop.md>), [lora](<https://devfeed.tech/tags/lora.md>), [matlab](<https://devfeed.tech/tags/matlab.md>), [qlora](<https://devfeed.tech/tags/qlora.md>), [sft](<https://devfeed.tech/tags/sft.md>), [signal-processing](<https://devfeed.tech/tags/signal-processing.md>)

### AI overview

Laptop iGPU fine-tuning improved Gemma 4's performance on checked MATLAB signal-processing tasks, especially errors involving DSP conventions. Using 66 examples across 11 subtasks, the study applied SFT with LoRA or QLoRA and deployed int8 models through OpenVINO. On held-out variants, E2B pass rates rose from 18/66 to 37/66 with either LoRA method; E4B rose from 32/66 to 47/66 with QLoRA. Analysis of 159 failing programs found that adapters corrected many convention errors, including lag signs, units, and array shapes, but few errors in constructing signals and noise. Runtime adapter use reduced generation speed by 17-19%.

### Source excerpt

arXiv:2611.23205v1 Announce Type: cross Abstract: Small language models that run on a laptop are attractive for writing signal-processing code offline, but they make systematic errors on digital signal processing (DSP) conventions. We apply supervised fine-tuning (SFT) to the Gemma 4 edge models E2B and E4B with 66 checked examples covering 11 signal-processing MATLAB subtasks, using low-rank adaptation (LoRA) and quantized LoRA (QLoRA) on the integrated GPU (iGPU) of a laptop, and deploy the 8-bit integer (int8) models with the adapters through OpenVINO. On held-out variants scored by deterministic checkers, the pass rate rises from 18/66 to 37/66 (LoRA) and 37/66 (QLoRA) for E2B and from 32/66 to 47/66 for E4B (QLoRA). Reading all 159 failing programs shows that the adapters remove most errors of convention, such as the sign of a lag, units and array shapes, but few errors in building signals and noise. Applied at run time, the adapter costs 17-19% of the generation speed.

## Fairness-Aware Test-Time Prompt Tuning

DevFeed: [Fairness-Aware Test-Time Prompt Tuning](<https://devfeed.tech/articles/fairness-aware-test-time-prompt-tuning-141598.md>)

Original publisher: [Read original article](<https://arxiv.org/abs/2608.25707>)

Author: Yoann Launay, Parameswaran Kamalaruban, Tom Kempton, Stuart Burrell, David Sutton

Published: 2026-11-18T05:00:00Z

Content type: article

Language: en

Sources: [arXiv: Machine Learning](<https://devfeed.tech/sources/arxiv-machine-learning.md>)

Topics: [Enhancing AI model performance](<https://devfeed.tech/topics/enhancing-ai-model-performance.md>), [responsible-ai](<https://devfeed.tech/topics/responsible-ai.md>), [peft](<https://devfeed.tech/topics/peft.md>)

Tags: [clip](<https://devfeed.tech/tags/clip.md>), [cs-lg](<https://devfeed.tech/tags/cs-lg.md>), [fairness](<https://devfeed.tech/tags/fairness.md>), [machine-learning](<https://devfeed.tech/tags/machine-learning.md>), [prompt](<https://devfeed.tech/tags/prompt.md>), [test-time-adaptation](<https://devfeed.tech/tags/test-time-adaptation.md>), [vision-language-models](<https://devfeed.tech/tags/vision-language-models.md>)

### AI overview

FairTPT adapts CLIP at test time to address fairness under subpopulation shifts without retraining or demographic attributes. Through soft-prompt tuning, it jointly minimizes target marginal entropy and maximizes spurious marginal entropy. The authors report that standard episodic test-time adaptation generally worsens disparities, while excessive suppression of spurious attributes can cause catastrophic forgetting. Monitoring test-time changes in target loss within the linear regime can prevent this collapse while retaining fairness gains on reactive data and overall performance. The abstract reports that FairTPT outperforms the state-of-the-art episodic test-time debiasing methods.

### Source excerpt

arXiv:2608.25707v2 Announce Type: replace Abstract: Vision-language models have displayed remarkable capabilities in multi-modal understanding and are increasingly used in critical applications where economic and practical deployment constraints prohibit re-training or fine-tuning. However, these models can also exhibit systematic biases that disproportionately affect protected demographic groups and existing approaches to addressing these biases require extensive model retraining and access to demographic attributes. There is a clear need to develop test-time adaptation (TTA) approaches that improve the fairness characteristics of pretrained models under distributional shift. In this paper, we evaluate how episodic TTA affects fairness in CLIP classification under subpopulation shifts and develop FairTPT, a novel fairness-aware episodic TTA method that jointly minimizes target marginal entropy while maximizing spurious marginal entropy through soft-prompt tuning. We find that standard episodic TTA generally exacerbates disparities between majority and minority groups, that blinding a model to spurious attributes without degrading target performance is inherently challenging, and that excessive blinding can lead to catastrophic forgetting. This model collapse can be prevented by monitoring test-time changes in target loss within the linear regime, while still achieving fairness improvements on reactive data and preserving overall performance. FairTPT outperforms all state-of-the-art episodic test-time debiasing methods and establishes a foundation for robust TTA, which is essential for achieving fairness in practice.

## Hidden-State Similarity Predicts Re-Elicitation After Inoculation Prompting

DevFeed: [Hidden-State Similarity Predicts Re-Elicitation After Inoculation Prompting](<https://devfeed.tech/articles/hidden-state-similarity-predicts-re-elicitation-after-inoculation-prompting-140120.md>)

Original publisher: [Read original article](<https://arxiv.org/abs/2611.23096>)

Author: Rose N. Kwon, Samy Mammeri, Sabyasachi Sahoo, Christian Gagn\\'e, Thomas Jiralerspong

Published: 2026-11-18T05:00:00Z

Content type: article

Language: en

Sources: [arXiv: Artificial Intelligence](<https://devfeed.tech/sources/arxiv-artificial-intelligence.md>)

Topics: [post-training](<https://devfeed.tech/topics/post-training.md>), [Enhancing AI model performance](<https://devfeed.tech/topics/enhancing-ai-model-performance.md>), [peft](<https://devfeed.tech/topics/peft.md>)

Tags: [activation](<https://devfeed.tech/tags/activation.md>), [analysis](<https://devfeed.tech/tags/analysis.md>), [artificial-intelligence](<https://devfeed.tech/tags/artificial-intelligence.md>), [cs-ai](<https://devfeed.tech/tags/cs-ai.md>), [eval](<https://devfeed.tech/tags/eval.md>), [evaluation](<https://devfeed.tech/tags/evaluation.md>), [fine-tuning](<https://devfeed.tech/tags/fine-tuning.md>), [misalignment](<https://devfeed.tech/tags/misalignment.md>), [models](<https://devfeed.tech/tags/models.md>), [prompting](<https://devfeed.tech/tags/prompting.md>), [prompts](<https://devfeed.tech/tags/prompts.md>), [similarity](<https://devfeed.tech/tags/similarity.md>), [study](<https://devfeed.tech/tags/study.md>)

### AI overview

Inoculation prompting can suppress harmful behavior learned during narrow-task fine-tuning, but evaluation prompts may bring it back. The study finds that textual similarity to the inoculation prompt is an incomplete predictor of re-elicitation; prompts are more likely to trigger suppressed behavior when they produce activation states similar to those induced by the inoculation context. The authors suggest activation-space analysis may help identify when suppressed behaviors remain accessible under evaluation prompts.

### Source excerpt

arXiv:2611.23096v1 Announce Type: new Abstract: Fine-tuning on narrow harmful tasks can cause emergent misalignment, where models generalize harmful behavior beyond the training distribution. Inoculation prompting can reduce this effect by explicitly eliciting the undesired behavior during training, but recent work shows that the behavior can reappear when evaluation prompts contain cues from the training context. We study what makes such prompts effective triggers. We find that, surprisingly, textual similarity to the inoculation prompt is an incomplete predictor. Instead, prompts are more likely to re-elicit suppressed behavior when they induce activation states similar to those produced by the inoculation context. These findings advance our understanding of how inoculation prompting modulates conditional misalignment, and suggest that activation space analysis can help identify when suppressed behaviors remain accessible under eval-time prompts.

## Specificity-Guided Knowledge Diversion for Decomposable Generative Models

DevFeed: [Specificity-Guided Knowledge Diversion for Decomposable Generative Models](<https://devfeed.tech/articles/specificity-guided-knowledge-diversion-for-decomposable-generative-models-139159.md>)

Original publisher: [Read original article](<https://arxiv.org/abs/2611.21352>)

Author: Yucheng Xie, Fu Feng, Jing Wang, Xin Geng, Yong Rui

Published: 2026-11-17T05:00:00Z

Content type: article

Language: en

Sources: [arXiv: Machine Learning](<https://devfeed.tech/sources/arxiv-machine-learning.md>)

Topics: [AI Foundation Models](<https://devfeed.tech/topics/ai-foundation-models.md>), [peft](<https://devfeed.tech/topics/peft.md>), [Mixture of Experts (MoE)](<https://devfeed.tech/topics/mixture-of-experts-moe.md>)

Tags: [cs-cv](<https://devfeed.tech/tags/cs-cv.md>), [cs-lg](<https://devfeed.tech/tags/cs-lg.md>), [datasets](<https://devfeed.tech/tags/datasets.md>), [experiments](<https://devfeed.tech/tags/experiments.md>), [generative](<https://devfeed.tech/tags/generative.md>), [generative-models](<https://devfeed.tech/tags/generative-models.md>), [knowledge](<https://devfeed.tech/tags/knowledge.md>), [moe](<https://devfeed.tech/tags/moe.md>), [parameter](<https://devfeed.tech/tags/parameter.md>), [peft](<https://devfeed.tech/tags/peft.md>), [quality](<https://devfeed.tech/tags/quality.md>), [reuse](<https://devfeed.tech/tags/reuse.md>), [routing](<https://devfeed.tech/tags/routing.md>), [specialization](<https://devfeed.tech/tags/specialization.md>), [structure](<https://devfeed.tech/tags/structure.md>), [units](<https://devfeed.tech/tags/units.md>)

### AI overview

SGKDiv aims to make generative models easier to adapt selectively while retaining generation quality. It groups paired feed-forward channels into Knowledge Units without SVD-based reparameterization, then uses calibrated condition-wise routing that combines a shared usage prior with conditional adjustments. Complementary regularization shapes unit allocation, diversity, and magnitude, replacing fixed component assignments with learned specialization. The abstract reports better in-distribution generation than dense and MoE baselines and stronger downstream adaptation than representative PEFT methods across eight datasets, but provides no quantitative results or further methodological details.

### Source excerpt

arXiv:2611.21352v1 Announce Type: new Abstract: Generative model pre-training typically optimizes a model as a whole without explicitly organizing its parameters for selective reuse. Decomposable pre-training instead learns reusable components for downstream adaptation while retaining strong generative performance. However, existing knowledge diversion methods incur substantial parameter overhead through weight factorization and rely on fixed component assignments to induce specialization, potentially compromising in-distribution generation quality. We therefore propose Specificity-Guided Knowledge Diversion (SGKDiv), which learns how generative knowledge is shared and specialized for effective generation and selective downstream reuse. SGKDiv constructs Knowledge Units from paired feed-forward channels, preserving native computational structure without SVD-based reparameterization. Calibrated condition-wise routing combines a shared usage prior with conditional adjustments, while complementary regularization shapes their allocation, diversity, and magnitude. Together, these designs replace predefined component assignments with adaptively learned and explicitly regulated knowledge specialization. Experiments across diverse model and training settings demonstrate better in-distribution generation than dense and MoE baselines, alongside stronger downstream adaptation than representative PEFT methods across eight datasets.

## Quantum-inspired orthogonal adapters for certifiably robust medical image classification

DevFeed: [Quantum-inspired orthogonal adapters for certifiably robust medical image classification](<https://devfeed.tech/articles/quantum-inspired-orthogonal-adapters-for-certifiably-robust-medical-image-classification-138786.md>)

Original publisher: [Read original article](<https://arxiv.org/abs/2611.18775>)

Author: Olga Tsurkan, Laia Domingo, Victoria Orlova, Asel Sagingalieva, Iris Zalaudek, Raffaele Dante Caposiena Caro

Published: 2026-11-17T05:00:00Z

Content type: article

Language: en

Sources: [arXiv: Machine Learning](<https://devfeed.tech/sources/arxiv-machine-learning.md>)

Topics: [peft](<https://devfeed.tech/topics/peft.md>), [Medical imaging](<https://devfeed.tech/topics/medical-imaging.md>), [Adversarial attacks](<https://devfeed.tech/topics/adversarial-attacks.md>), [AI Foundation Models](<https://devfeed.tech/topics/ai-foundation-models.md>), [Image](<https://devfeed.tech/topics/image.md>), [ImageNet](<https://devfeed.tech/topics/imagenet.md>)

Tags: [adapters](<https://devfeed.tech/tags/adapters.md>), [adversarial](<https://devfeed.tech/tags/adversarial.md>), [attacks](<https://devfeed.tech/tags/attacks.md>), [butterfly](<https://devfeed.tech/tags/butterfly.md>), [certificates](<https://devfeed.tech/tags/certificates.md>), [circuit](<https://devfeed.tech/tags/circuit.md>), [classification](<https://devfeed.tech/tags/classification.md>), [classification-accuracy](<https://devfeed.tech/tags/classification-accuracy.md>), [classifiers](<https://devfeed.tech/tags/classifiers.md>), [connectivity](<https://devfeed.tech/tags/connectivity.md>), [cost](<https://devfeed.tech/tags/cost.md>), [covering](<https://devfeed.tech/tags/covering.md>), [cs-cv](<https://devfeed.tech/tags/cs-cv.md>), [cs-lg](<https://devfeed.tech/tags/cs-lg.md>), [design](<https://devfeed.tech/tags/design.md>), [diagnosis](<https://devfeed.tech/tags/diagnosis.md>), [experiments](<https://devfeed.tech/tags/experiments.md>), [image](<https://devfeed.tech/tags/image.md>), [image-classification](<https://devfeed.tech/tags/image-classification.md>), [medical](<https://devfeed.tech/tags/medical.md>), [quant-ph](<https://devfeed.tech/tags/quant-ph.md>), [quantum](<https://devfeed.tech/tags/quantum.md>), [robust](<https://devfeed.tech/tags/robust.md>)

### AI overview

Quantum-inspired orthogonal butterfly layers adapt certified pretrained image classifiers while preserving their Lipschitz bounds, using O(n log n) trainable parameters. In MNIST and MedMNIST experiments they improved certified robustness over conventional alternatives; on DermaMNIST transfer, they resisted tested attacks better than LoRA. For trichoscopic alopecia classification, adapting a certified ImageNet-pretrained network yielded about twice the certified accuracy of conventional fine-tuning and LoRA, with comparable classification accuracy and similar resistance to tested attacks on the same backbone and data split. Broader certificates, including for certain image-processing transformations, came at the cost of reduced accuracy. The abstract describes efficient classical implementation and a possible future quantum path under suitable connectivity assumptions.

### Source excerpt

arXiv:2611.18775v1 Announce Type: cross Abstract: Medical image classifiers require not only high accuracy but also reliable predictions under small input perturbations. Certified robustness provides mathematical guarantees of prediction stability, but preserving these guarantees when adapting pretrained models to new tasks remains challenging. We propose a fine-tuning approach based on quantum-inspired orthogonal butterfly layers that preserves the Lipschitz bounds of pretrained certified networks. Inspired by quantum circuits of reconfigurable beam splitters, these layers preserve orthogonality by construction using only $\mathcal{O}(n\log n)$ trainable parameters. Their structured design enables efficient classical implementation and offers a path toward future quantum implementations with logarithmic circuit depth under suitable connectivity assumptions. Controlled experiments on MNIST and MedMNIST demonstrate improved certified robustness compared with conventional alternatives and, in the DermaMNIST transfer, greater resistance to the tested attacks than low-rank adaptation (LoRA). We then apply it to trichoscopic image classification for alopecia diagnosis and alopecia areata activity assessment. By adapting a certified ImageNet-pretrained network, butterfly fine-tuning achieves approximately twice the certified accuracy of conventional fine-tuning and LoRA, with comparable classification accuracy on the same backbone and data split and similar empirical resistance to the tested adversarial attacks. Larger robustness guarantees, including certificates covering certain image-processing transformations, can be obtained at the cost of reduced accuracy. These results highlight the potential of quantum-inspired butterfly layers for parameter-efficient adaptation of medical image classifiers while preserving the mathematical guarantees underlying certified robustness.

## GRAFT: Cross-Backbone LoRA Reuse for Continual Language and Multimodal Learning

DevFeed: [GRAFT: Cross-Backbone LoRA Reuse for Continual Language and Multimodal Learning](<https://devfeed.tech/articles/graft-cross-backbone-lora-reuse-for-continual-language-and-multimodal-learning-138757.md>)

Original publisher: [Read original article](<https://arxiv.org/abs/2611.18638>)

Author: Ha Pham Thien Dinh, Ming Liu, Jiesi Guo, Matteo Vergani, Kewen Liao

Published: 2026-11-17T05:00:00Z

Content type: article

Language: en

Sources: [arXiv: Machine Learning](<https://devfeed.tech/sources/arxiv-machine-learning.md>)

Topics: [lora](<https://devfeed.tech/topics/lora.md>), [peft](<https://devfeed.tech/topics/peft.md>), [Fine-tuning](<https://devfeed.tech/topics/fine-tuning.md>), [Large Language Model](<https://devfeed.tech/topics/llm.md>), [Model Development](<https://devfeed.tech/topics/model-development.md>), [qwen](<https://devfeed.tech/topics/qwen.md>), [Artificial Intelligence](<https://devfeed.tech/topics/ai.md>)

Tags: [adapter](<https://devfeed.tech/tags/adapter.md>), [adapters](<https://devfeed.tech/tags/adapters.md>), [bridge](<https://devfeed.tech/tags/bridge.md>), [computer-science](<https://devfeed.tech/tags/computer-science.md>), [continual-learning](<https://devfeed.tech/tags/continual-learning.md>), [cs-cl](<https://devfeed.tech/tags/cs-cl.md>), [cs-lg](<https://devfeed.tech/tags/cs-lg.md>), [feature](<https://devfeed.tech/tags/feature.md>), [fine-tuning](<https://devfeed.tech/tags/fine-tuning.md>), [language-models](<https://devfeed.tech/tags/language-models.md>), [large-language-models](<https://devfeed.tech/tags/large-language-models.md>), [library](<https://devfeed.tech/tags/library.md>), [linear](<https://devfeed.tech/tags/linear.md>), [lora](<https://devfeed.tech/tags/lora.md>), [machine-learning](<https://devfeed.tech/tags/machine-learning.md>), [multimodal](<https://devfeed.tech/tags/multimodal.md>), [multimodal-learning](<https://devfeed.tech/tags/multimodal-learning.md>), [multimodal-models](<https://devfeed.tech/tags/multimodal-models.md>), [reuse](<https://devfeed.tech/tags/reuse.md>)

### AI overview

GRAFT reuses LoRA adapters across different model backbones by retrieving a relevant adapter from a modality-aware library and mapping it to the target model with a learned linear bridge between feature spaces. The evaluation transfers Qwen2.5 adapters to Qwen2.5-VL on ten Super-NaturalInstructions text tasks and directly reuses Qwen2.5-VL adapters on five multimodal tasks. Compared with randomly initialized LoRA adapters, GRAFT improves performance on most evaluated tasks; the supplied excerpt does not specify the size of those gains.

### Source excerpt

arXiv:2611.18638v1 Announce Type: new Abstract: In continual learning, parameter-efficient fine-tuning enables large language models to acquire new capabilities through lightweight adapters such as LoRA, yet the reuse of pretrained adapters is typically restricted to models sharing the same backbone. This limits the transfer of learned capabilities from language models to multimodal models. We propose GRAFT (Generalist Reuse of Adapters via Feature-space Transfer), a framework for cross-backbone LoRA reuse. GRAFT retrieves the most relevant pretrained LoRA adapter from a modality-aware library and maps it into the target backbone through a learned linear bridge between corresponding source and target feature spaces. We evaluate GRAFT on ten Super-NaturalInstructions text tasks, transferring Qwen2.5 LoRAs to Qwen2.5-VL, and five multimodal tasks, where Qwen2.5-VL LoRAs are directly reused. Compared with randomly initialized LoRA adapters, GRAFT improves performance on most evaluated tasks. These results show that GRAFT provides a simple and effective mechanism for reusing learned capabilities across language and multimodal models in continual learning.

## UNITE: Consolidating Embodied VLM Specialists via Shared-Anchor One-Bit Factorization

DevFeed: [UNITE: Consolidating Embodied VLM Specialists via Shared-Anchor One-Bit Factorization](<https://devfeed.tech/articles/unite-consolidating-embodied-vlm-specialists-via-shared-anchor-one-bit-factorization-139093.md>)

Original publisher: [Read original article](<https://arxiv.org/abs/2611.20767>)

Author: Tao Tao, Anke Tang, Enneng Yang, Zhi Hou, Xiaogang Wang, Dacheng Tao

Published: 2026-11-17T05:00:00Z

Content type: article

Language: en

Sources: [arXiv: Machine Learning](<https://devfeed.tech/sources/arxiv-machine-learning.md>)

Topics: [Machine Perception](<https://devfeed.tech/topics/machine-perception.md>), [Mixture of Experts (MoE)](<https://devfeed.tech/topics/mixture-of-experts-moe.md>), [LLM evaluation / benchmarking](<https://devfeed.tech/topics/llm-evaluation-benchmarking.md>), [Fine-tuning](<https://devfeed.tech/topics/fine-tuning.md>), [peft](<https://devfeed.tech/topics/peft.md>)

Tags: [3d](<https://devfeed.tech/tags/3d.md>), [3d-scene-understanding](<https://devfeed.tech/tags/3d-scene-understanding.md>), [anchor](<https://devfeed.tech/tags/anchor.md>), [benchmarks](<https://devfeed.tech/tags/benchmarks.md>), [cs-cv](<https://devfeed.tech/tags/cs-cv.md>), [cs-lg](<https://devfeed.tech/tags/cs-lg.md>), [driving](<https://devfeed.tech/tags/driving.md>), [experts](<https://devfeed.tech/tags/experts.md>), [floating-point](<https://devfeed.tech/tags/floating-point.md>), [mixture-of-experts](<https://devfeed.tech/tags/mixture-of-experts.md>), [model](<https://devfeed.tech/tags/model.md>), [moe](<https://devfeed.tech/tags/moe.md>), [navigation](<https://devfeed.tech/tags/navigation.md>), [router](<https://devfeed.tech/tags/router.md>), [spatial](<https://devfeed.tech/tags/spatial.md>), [spatial-reasoning](<https://devfeed.tech/tags/spatial-reasoning.md>), [specialists](<https://devfeed.tech/tags/specialists.md>), [storage](<https://devfeed.tech/tags/storage.md>), [token](<https://devfeed.tech/tags/token.md>), [unite](<https://devfeed.tech/tags/unite.md>), [vlm](<https://devfeed.tech/tags/vlm.md>)

### AI overview

UNITE combines four embodied vision-language specialists in a shared-backbone mixture of experts, retaining each specialist's feed-forward block as a routable expert. It compresses the expert bank with a shared floating-point anchor and expert-specific row-scaled binary deviations, optimized from source weights without calibration data. Router adaptation freezes the backbone and experts and updates fewer than 0.6 million parameters. The resulting Qwen3-VL-8B model occupies 20.3 GB--1.16 times one dense specialist and 59.6% less than the uncompressed MoE. Across 22 benchmarks and R2R Val-Unseen, it matches six static merging methods in mean retention of best-specialist performance, leads them on Scan2Cap, ScanQA, and four navigation metrics, and exceeds every source specialist on five 3D scene-understanding benchmarks and R2R navigation. Factorization is nearly lossless versus the uncompressed MoE on 21 of 22 benchmarks; VSI-Bench is the exception, with frame-count experiments suggesting weaker use of cross-frame evidence.

### Source excerpt

arXiv:2611.20767v1 Announce Type: new Abstract: Embodied capabilities such as visual grounding, spatial reasoning, navigation, and driving understanding are often developed in separate vision-language specialists fine-tuned from a common pretrained model.Deploying the specialists separately multiplies storage, whereas merging them into one dense model forces conflicting task updates to share parameters and loses part of each specialist's expertise.We introduce UNITE, a framework that consolidates such specialists into a compact shared-backbone mixture of experts. UNITE keeps the feed-forward block of each specialist as a routable expert and represents the expert bank through a shared-anchor one-bit factorization: a shared floating-point anchor plus expert-specific, row-scaled binary deviations.The anchor is optimized jointly with the scales and signs through closed-form alternating updates, using only source weights and no calibration data.Router adaptation then adjusts token-level expert selection with the backbone and experts frozen, updating fewer than $0.6$M parameters ($<0.01\%$ of the $8.8$B active parameters). Built from four Qwen3-VL-8B specialists, the complete UNITE model occupies $20.3$ GB, $1.16\times$ the storage of one dense specialist and $59.6\%$ below the uncompressed MoE. Across $22$ benchmarks and R2R Val-Unseen, UNITE matches six static merging methods in mean retention of best-specialist performance, leads all of them on Scan2Cap, ScanQA, and all four navigation metrics, and surpasses every source specialist on all five 3D scene understanding benchmarks (up to $+19.1$ on Scan2Cap) and on R2R navigation. Relative to the uncompressed MoE, factorization is nearly lossless on $21$ of $22$ benchmarks; the exception is VSI-Bench, where frame-count experiments suggest impaired use of cross-frame evidence.UNITE thus consolidates complementary embodied capabilities near single-model storage.

## Tight PAC-Bayes Generalisation Guarantees for Large Language Model Judges

DevFeed: [Tight PAC-Bayes Generalisation Guarantees for Large Language Model Judges](<https://devfeed.tech/articles/tight-pac-bayes-generalisation-guarantees-for-large-language-model-judges-139098.md>)

Original publisher: [Read original article](<https://arxiv.org/abs/2611.20818>)

Author: Tom A. Lamb, Philip H. S. Torr, Tim G. J. Rudner

Published: 2026-11-17T05:00:00Z

Content type: article

Language: en

Sources: [arXiv: Machine Learning](<https://devfeed.tech/sources/arxiv-machine-learning.md>)

Topics: [Uncertainty quantification LLMs](<https://devfeed.tech/topics/uncertainty-quantification-llms.md>), [lora](<https://devfeed.tech/topics/lora.md>), [peft](<https://devfeed.tech/topics/peft.md>), [Fine-tuning](<https://devfeed.tech/topics/fine-tuning.md>), [LLM evaluation / benchmarking](<https://devfeed.tech/topics/llm-evaluation-benchmarking.md>), [Large Language Model](<https://devfeed.tech/topics/llm.md>)

Tags: [agents](<https://devfeed.tech/tags/agents.md>), [ai](<https://devfeed.tech/tags/ai.md>), [ai-agents](<https://devfeed.tech/tags/ai-agents.md>), [bayes](<https://devfeed.tech/tags/bayes.md>), [benchmarks](<https://devfeed.tech/tags/benchmarks.md>), [certificates](<https://devfeed.tech/tags/certificates.md>), [classification](<https://devfeed.tech/tags/classification.md>), [coding](<https://devfeed.tech/tags/coding.md>), [complexity](<https://devfeed.tech/tags/complexity.md>), [compression](<https://devfeed.tech/tags/compression.md>), [cs-lg](<https://devfeed.tech/tags/cs-lg.md>), [data](<https://devfeed.tech/tags/data.md>), [fine-tuning](<https://devfeed.tech/tags/fine-tuning.md>), [formal](<https://devfeed.tech/tags/formal.md>), [framework](<https://devfeed.tech/tags/framework.md>), [generator](<https://devfeed.tech/tags/generator.md>), [guarantees](<https://devfeed.tech/tags/guarantees.md>), [guardrail](<https://devfeed.tech/tags/guardrail.md>), [language-model](<https://devfeed.tech/tags/language-model.md>), [large-language-model](<https://devfeed.tech/tags/large-language-model.md>), [layers](<https://devfeed.tech/tags/layers.md>)

### AI overview

The paper applies PAC-Bayes bounds to certify safety judges on prompt-response pairs from a deployed generator. It reports that standard parameter-efficient fine-tuning can produce non-vacuous, sometimes tight certificates for classification and Brier risk without a held-out validation set. Its LoRA-GT method shares one quantisation scale across layers and gives the tightest certificates with minimal predictive-performance degradation. Across five guardrail models and four benchmarks, certificate tightness depends more on adaptation complexity than base-model scale. The authors also examine when quantisation raises risk enough to outweigh the tighter bounds from shorter descriptions.

### Source excerpt

arXiv:2611.20818v1 Announce Type: new Abstract: Large language model (LLM) judges are increasingly used to monitor the outputs of LLM-based AI agents, yet formal guarantees on the reliability of these judges remain largely absent, limiting confidence in their use for safety-critical oversight. We consider certifying such a judge and define its risk over safety classifications of prompt--response pairs drawn from a deployed generator. We apply a PAC-Bayes framework to this setting, refining existing compression-based methodology with the use of tighter bounds and more efficient coding. We show that standard parameter-efficient fine-tuning can yield non-vacuous and even tight certificates for safety judges, for both classification and Brier risk, without holding out data for validation. We introduce LoRA-GT, which shares a single quantisation scale across layers and achieves the tightest certificates with minimal degradation in predictive performance. Across five guardrail models and four benchmarks, we find that certificate tightness is governed more by adaptation complexity than by base model scale. Finally, we show how far quantisation can be pushed before the resulting increase in risk outweighs the bound tightening gained from shorter descriptions. Together, these results provide a practical route toward certifiable scalable oversight and offer insight into how the complexity of safety adaptations shapes generalisation.

## PMR: MoE Routing with Shared Prototypes and Congestion Pricing

DevFeed: [PMR: MoE Routing with Shared Prototypes and Congestion Pricing](<https://devfeed.tech/articles/pmr-moe-routing-with-shared-prototypes-and-congestion-pricing-137445.md>)

Original publisher: [Read original article](<https://arxiv.org/abs/2611.19413>)

Author: Wending Wang, Binqi Huang, Ju'an Li, Yu Guo, Lei Duan, Yuqi Ouyang, Dengchun Li

Published: 2026-11-17T05:00:00Z

Content type: article

Language: en

Sources: [arXiv: Artificial Intelligence](<https://devfeed.tech/sources/arxiv-artificial-intelligence.md>), [arXiv: Machine Learning](<https://devfeed.tech/sources/arxiv-machine-learning.md>)

Topics: [Mixture of Experts (MoE)](<https://devfeed.tech/topics/mixture-of-experts-moe.md>), [Machine learning](<https://devfeed.tech/topics/machine-learning.md>), [peft](<https://devfeed.tech/topics/peft.md>)

Tags: [arc](<https://devfeed.tech/tags/arc.md>), [balancing](<https://devfeed.tech/tags/balancing.md>), [benchmarks](<https://devfeed.tech/tags/benchmarks.md>), [conditional](<https://devfeed.tech/tags/conditional.md>), [context](<https://devfeed.tech/tags/context.md>), [cs-ai](<https://devfeed.tech/tags/cs-ai.md>), [cs-lg](<https://devfeed.tech/tags/cs-lg.md>), [dispatch](<https://devfeed.tech/tags/dispatch.md>), [distribution](<https://devfeed.tech/tags/distribution.md>), [experts](<https://devfeed.tech/tags/experts.md>), [fine](<https://devfeed.tech/tags/fine.md>), [fine-tuning](<https://devfeed.tech/tags/fine-tuning.md>), [linear](<https://devfeed.tech/tags/linear.md>), [load](<https://devfeed.tech/tags/load.md>), [many-to-many](<https://devfeed.tech/tags/many-to-many.md>), [matching](<https://devfeed.tech/tags/matching.md>), [mixture-of-experts](<https://devfeed.tech/tags/mixture-of-experts.md>), [models](<https://devfeed.tech/tags/models.md>), [moe](<https://devfeed.tech/tags/moe.md>), [pricing](<https://devfeed.tech/tags/pricing.md>), [prototypes](<https://devfeed.tech/tags/prototypes.md>), [routing](<https://devfeed.tech/tags/routing.md>)

### AI overview

Prototype-Mediated Routing (PMR) separates token-to-prototype matching from expert allocation: each token gets a prototype distribution, which mixes prototype-conditioned expert distributions. Short-term states adapt prototypes using preceding tokens, while congestion prices updated from dispatch feedback adjust expert distributions for loss-free load balancing. A two-pass blockwise approximation delivered over 20x the training throughput of eager tokenwise PMR with high routing agreement. In three pretraining configurations, PMR improved mean zero-shot accuracy across seven benchmarks over matched linear-routing baselines by 1.45-2.65 percentage points; ARC-Easy gains were 3.92-6.71 points. In parameter-efficient fine-tuning of 7B-8B backbones, gains over MixLoRA varied by backbone, reaching 1.76 points in mean accuracy across four tasks.

### Source excerpt

arXiv:2611.19413v1 Announce Type: new Abstract: Conventional linear routers in sparse mixture-of-experts models select experts directly from token-expert matching scores, coupling input matching with expert allocation. We propose Prototype-Mediated Routing (PMR), a two-stage router with shared prototypes. PMR first computes each token's distribution over prototypes, then uses it to mix prototype-conditioned expert distributions according to the law of total probability. This formulation models input matching and expert allocation separately, explicitly represents many-to-many associations between prototypes and experts, and provides separate interfaces for context adaptation and load regulation. Within this framework, short-term states summarize preceding tokens to adapt the prototypes, while congestion prices updated from dispatch feedback achieve loss-free load balancing by adjusting the conditional expert distributions. We use a two-pass blockwise approximation to accelerate short-term state processing, achieving over 20 times the training throughput of the eager tokenwise PMR implementation while maintaining high agreement in routing decisions. Across three pretraining configurations with 0.43B, 1.18B, and 0.48B total parameters, PMR improves mean zero-shot accuracy on seven benchmarks over linear routing baselines with matched expert architectures by 1.54, 2.65, and 1.45 percentage points (pp), respectively. The corresponding gains on ARC-Easy are 6.71, 5.80, and 3.92 pp. In parameter-efficient fine-tuning of 7B-8B backbones, PMR's improvements over MixLoRA vary across backbones, with gains of up to 1.76 pp in mean accuracy across four tasks.

## Efficiently Aligning Draft Models via Parameter- and Data-Efficient Adaptation

DevFeed: [Efficiently Aligning Draft Models via Parameter- and Data-Efficient Adaptation](<https://devfeed.tech/articles/efficiently-aligning-draft-models-via-parameter-and-data-efficient-adaptation-134966.md>)

Original publisher: [Read original article](<https://arxiv.org/abs/2603.09527>)

Author: Luxi Lin, Zhihang Lin, Zhanpeng Zeng, Yuhao Chen, Qingyu Zhang, Jixiang Luo, Xuelong Li, Rongrong Ji

Published: 2026-11-16T05:00:00Z

Content type: article

Language: en

Sources: [arXiv: Artificial Intelligence](<https://devfeed.tech/sources/arxiv-artificial-intelligence.md>), [arXiv: Machine Learning](<https://devfeed.tech/sources/arxiv-machine-learning.md>)

Topics: [peft](<https://devfeed.tech/topics/peft.md>), [Model Optimizer](<https://devfeed.tech/topics/model-optimizer.md>), [AI Foundation Models](<https://devfeed.tech/topics/ai-foundation-models.md>)

Tags: [adapting](<https://devfeed.tech/tags/adapting.md>), [architecture](<https://devfeed.tech/tags/architecture.md>), [component](<https://devfeed.tech/tags/component.md>), [components](<https://devfeed.tech/tags/components.md>), [cs-ai](<https://devfeed.tech/tags/cs-ai.md>), [cs-lg](<https://devfeed.tech/tags/cs-lg.md>), [data](<https://devfeed.tech/tags/data.md>), [distributions](<https://devfeed.tech/tags/distributions.md>), [draft](<https://devfeed.tech/tags/draft.md>), [eda](<https://devfeed.tech/tags/eda.md>), [framework](<https://devfeed.tech/tags/framework.md>), [inference](<https://devfeed.tech/tags/inference.md>), [llm](<https://devfeed.tech/tags/llm.md>), [llm-inference](<https://devfeed.tech/tags/llm-inference.md>), [models](<https://devfeed.tech/tags/models.md>), [parameter](<https://devfeed.tech/tags/parameter.md>), [practical](<https://devfeed.tech/tags/practical.md>), [private](<https://devfeed.tech/tags/private.md>), [solution](<https://devfeed.tech/tags/solution.md>), [speculative-decoding](<https://devfeed.tech/tags/speculative-decoding.md>), [target](<https://devfeed.tech/tags/target.md>), [updating](<https://devfeed.tech/tags/updating.md>)

### AI overview

Efficient Draft Adaptation (EDA) aims to restore speculative decoding performance when a target language model is fine-tuned for a specific domain, without retraining a draft model from scratch. It separates shared and target-specific output modeling, updating only a lightweight private component; regenerates training data with the fine-tuned target to better align adaptation with decoding; and selects high-value samples. The abstract reports higher average acceptance lengths and significantly lower training costs than full retraining, but provides no quantitative results or experimental details.

### Source excerpt

arXiv:2603.09527v2 Announce Type: replace-cross Abstract: Speculative decoding accelerates LLM inference but suffers from performance degradation when target models are fine-tuned for specific domains. A naive solution is to retrain draft models for every target model, which is costly and inefficient. To address this, we introduce a practical parameter- and data-efficient framework named Efficient Draft Adaptation, abbreviated as EDA, for efficiently adapting draft models. EDA introduces three innovations: (1) a decoupled architecture that utilizes shared and private components to model the shared and target-specific output distributions separately, enabling parameter-efficient adaptation by updating only the lightweight private component; (2) a data regeneration strategy that utilizes the fine-tuned target model to regenerate training data, thereby improving the alignment between training and speculative decoding, leading to higher average acceptance length; (3) a sample selection mechanism that prioritizes high-value data for efficient adaptation. Our experiments show that EDA effectively restores speculative performance on fine-tuned models, achieving superior average acceptance lengths with significantly reduced training costs compared to full retraining. Code is available at https://github.com/Lyn-Lucy/Efficient-Draft-Adaptation.

## Astraios: Parameter-Efficient Instruction Tuning Code Large Language Models

DevFeed: [Astraios: Parameter-Efficient Instruction Tuning Code Large Language Models](<https://devfeed.tech/articles/astraios-parameter-efficient-instruction-tuning-code-large-language-models-134941.md>)

Original publisher: [Read original article](<https://arxiv.org/abs/2401.00788>)

Author: Terry Yue Zhuo, Armel Zebaze, Nitchakarn Suppattarachai, Leandro von Werra, Harm de Vries, Qian Liu, Niklas Muennighoff

Published: 2026-11-16T05:00:00Z

Content type: article

Language: en

Sources: [arXiv: Artificial Intelligence](<https://devfeed.tech/sources/arxiv-artificial-intelligence.md>), [arXiv: Software Engineering](<https://devfeed.tech/sources/arxiv-software-engineering.md>)

Topics: [peft](<https://devfeed.tech/topics/peft.md>), [Fine-tuning](<https://devfeed.tech/topics/fine-tuning.md>), [lora](<https://devfeed.tech/topics/lora.md>), [LLM evaluation / benchmarking](<https://devfeed.tech/topics/llm-evaluation-benchmarking.md>)

Tags: [code](<https://devfeed.tech/tags/code.md>), [code-generation](<https://devfeed.tech/tags/code-generation.md>), [code-security](<https://devfeed.tech/tags/code-security.md>), [cs-ai](<https://devfeed.tech/tags/cs-ai.md>), [cs-cl](<https://devfeed.tech/tags/cs-cl.md>), [cs-se](<https://devfeed.tech/tags/cs-se.md>), [datasets](<https://devfeed.tech/tags/datasets.md>), [entropy](<https://devfeed.tech/tags/entropy.md>), [fft](<https://devfeed.tech/tags/fft.md>), [fine-tuning](<https://devfeed.tech/tags/fine-tuning.md>), [language-models](<https://devfeed.tech/tags/language-models.md>), [large-language-models](<https://devfeed.tech/tags/large-language-models.md>), [lora](<https://devfeed.tech/tags/lora.md>), [model-robustness](<https://devfeed.tech/tags/model-robustness.md>), [peft](<https://devfeed.tech/tags/peft.md>), [performance](<https://devfeed.tech/tags/performance.md>), [relationships](<https://devfeed.tech/tags/relationships.md>), [robustness](<https://devfeed.tech/tags/robustness.md>), [scale](<https://devfeed.tech/tags/scale.md>), [security](<https://devfeed.tech/tags/security.md>), [tasks](<https://devfeed.tech/tags/tasks.md>), [updated](<https://devfeed.tech/tags/updated.md>), [validation](<https://devfeed.tech/tags/validation.md>)

### AI overview

Astraios evaluates seven tuning methods across 28 instruction-tuned OctoCoder models, four sizes up to 16 billion parameters, and five tasks spanning eight code comprehension and generation datasets. Full-parameter fine-tuning generally achieves the best downstream performance, while PEFT methods vary in effectiveness by model scale; LoRA usually provides the strongest cost-performance trade-off. The abstract also reports reduced robustness and code security in larger models, and finds that tuning effectiveness in small models generalizes to larger ones. Instruction-tuning validation loss may serve as a reliable indicator of downstream performance.

### Source excerpt

arXiv:2401.00788v2 Announce Type: replace-cross Abstract: The high cost of full-parameter fine-tuning (FFT) of Large Language Models (LLMs) has led to a series of parameter-efficient fine-tuning (PEFT) methods. However, it remains unclear which methods provide the best cost-performance trade-off at different model scales. We introduce Astraios, a suite of 28 instruction-tuned OctoCoder models using 7 tuning methods and 4 model sizes up to 16 billion parameters. Through investigations across 5 tasks and 8 different datasets encompassing both code comprehension and code generation tasks, we find that FFT generally leads to the best downstream performance across all scales, and PEFT methods differ significantly in their efficacy based on the model scale. LoRA usually offers the most favorable trade-off between cost and performance. Further investigation into the effects of these methods on both model robustness and code security reveals that larger models tend to demonstrate reduced robustness and less security. At last, we explore the relationships among updated parameters, cross-entropy loss, and task performance. We find that the tuning effectiveness observed in small models generalizes well to larger models, and the validation loss in instruction tuning can be a reliable indicator of overall downstream performance.

## BoostLoRA: Growing Effective Rank by Boosting Adapters

DevFeed: [BoostLoRA: Growing Effective Rank by Boosting Adapters](<https://devfeed.tech/articles/boostlora-growing-effective-rank-by-boosting-adapters-134976.md>)

Original publisher: [Read original article](<https://arxiv.org/abs/2604.27308>)

Author: Raviteja Anantha, Nick Levato, Layne C. Price

Published: 2026-11-16T05:00:00Z

Content type: article

Language: en

Sources: [arXiv: Artificial Intelligence](<https://devfeed.tech/sources/arxiv-artificial-intelligence.md>), [arXiv: Machine Learning](<https://devfeed.tech/sources/arxiv-machine-learning.md>)

Topics: [peft](<https://devfeed.tech/topics/peft.md>), [lora](<https://devfeed.tech/topics/lora.md>), [Fine-tuning](<https://devfeed.tech/topics/fine-tuning.md>), [Enhancing AI model performance](<https://devfeed.tech/topics/enhancing-ai-model-performance.md>)

Tags: [adapters](<https://devfeed.tech/tags/adapters.md>), [architecture](<https://devfeed.tech/tags/architecture.md>), [boosting](<https://devfeed.tech/tags/boosting.md>), [classification](<https://devfeed.tech/tags/classification.md>), [code-generation](<https://devfeed.tech/tags/code-generation.md>), [cs-ai](<https://devfeed.tech/tags/cs-ai.md>), [cs-lg](<https://devfeed.tech/tags/cs-lg.md>), [entropy](<https://devfeed.tech/tags/entropy.md>), [fine-tuning](<https://devfeed.tech/tags/fine-tuning.md>), [humaneval](<https://devfeed.tech/tags/humaneval.md>), [inference](<https://devfeed.tech/tags/inference.md>), [math-500](<https://devfeed.tech/tags/math-500.md>), [mbpp](<https://devfeed.tech/tags/mbpp.md>), [peft](<https://devfeed.tech/tags/peft.md>), [rank](<https://devfeed.tech/tags/rank.md>), [training](<https://devfeed.tech/tags/training.md>)

### AI overview

BoostLoRA grows a parameter-efficient adapter's cumulative effective rank across training rounds while keeping each round's adapter ultra-low-rank. It iteratively trains adapters on examples the current model gets wrong, assigns rounds to orthogonal subspaces using a ROTATE SVD basis, and merges each adapter before discarding it, leaving no inference overhead. On Qwen2.5-3B, the abstract reports 89.1% on GSM8K and 68.8% on MATH-500, exceeding both TinyLoRA and full fine-tuning; on MBPP and HumanEval, it reports 57.2% and 80.4%, respectively, while full fine-tuning falls below the zero-shot baseline. It also reports transfer to protein binding classification with ESM2-650M. These are abstract-level claims; the supplied excerpt does not include experimental details.

### Source excerpt

arXiv:2604.27308v2 Announce Type: replace-cross Abstract: Parameter-efficient fine-tuning (PEFT) methods face a tradeoff between adapter size and expressivity: ultra-low-parameter adapters are confined to fixed low-rank subspaces, capping performance even with extended training. We propose BoostLoRA, a gradient-boosting framework that overcomes this limit by iteratively training and merging minimal adapters on the examples the current model gets wrong. A ROTATE SVD basis strategy assigns each round to an orthogonal subspace, so cumulative effective rank grows linearly with the number of rounds while each adapter remains ultra-low-rank. After merging, adapters are discarded, leaving zero inference overhead. On Qwen2.5-3B, BoostLoRA reaches 89.1% on GSM8K and 68.8% on MATH-500, surpassing both the best single-shot ultra-low parameter adapter (TinyLoRA) and full fine-tuning; on code generation it reaches 57.2% on MBPP and 80.4% on HumanEval while full fine-tuning drops below the zero-shot baseline. We also demonstrate cross-architecture transfer on protein binding classification with ESM2-650M and cross-entropy training. BoostLoRA is, to our knowledge, the first PEFT method whose effective rank grows with training, separating per-round parameter cost from total representational capacity.

## Direction or Magnitude? Learning Scalars over Task-Calibrated Directions

DevFeed: [Direction or Magnitude? Learning Scalars over Task-Calibrated Directions](<https://devfeed.tech/articles/direction-or-magnitude-learning-scalars-over-task-calibrated-directions-135830.md>)

Original publisher: [Read original article](<https://arxiv.org/abs/2611.17784>)

Author: Ziqing Wen, Xiaohui Song, Jiahuan Wang, Ping Luo, Junlin Zeng, Dongsheng Li, Kele Xu, Tao Sun

Published: 2026-11-16T05:00:00Z

Content type: article

Language: en

Sources: [arXiv: Machine Learning](<https://devfeed.tech/sources/arxiv-machine-learning.md>)

Topics: [peft](<https://devfeed.tech/topics/peft.md>), [Fine-tuning](<https://devfeed.tech/topics/fine-tuning.md>), [Machine learning](<https://devfeed.tech/topics/machine-learning.md>)

Tags: [code-generation](<https://devfeed.tech/tags/code-generation.md>), [costs](<https://devfeed.tech/tags/costs.md>), [cs-lg](<https://devfeed.tech/tags/cs-lg.md>), [efficiency](<https://devfeed.tech/tags/efficiency.md>), [evaluation](<https://devfeed.tech/tags/evaluation.md>), [fine-tuning](<https://devfeed.tech/tags/fine-tuning.md>), [lora](<https://devfeed.tech/tags/lora.md>), [mathematical-reasoning](<https://devfeed.tech/tags/mathematical-reasoning.md>), [memory](<https://devfeed.tech/tags/memory.md>), [parameter](<https://devfeed.tech/tags/parameter.md>), [peft](<https://devfeed.tech/tags/peft.md>), [performance](<https://devfeed.tech/tags/performance.md>), [reasoning](<https://devfeed.tech/tags/reasoning.md>), [resource](<https://devfeed.tech/tags/resource.md>), [scalar](<https://devfeed.tech/tags/scalar.md>), [sharing](<https://devfeed.tech/tags/sharing.md>), [throughput](<https://devfeed.tech/tags/throughput.md>), [training](<https://devfeed.tech/tags/training.md>)

### AI overview

SCALAR adapts a model by learning layer-specific scalar coefficients over a fixed input basis built from task activations, paired with random output dictionaries. Grouped computation combines output atoms before applying them to token activations, and sharing dictionaries across compatible layers reduces fixed-basis storage. The abstract reports effective adaptation across commonsense and mathematical reasoning, code generation, and knowledge evaluation. On commonsense reasoning, SCALAR reaches accuracy comparable to LoRA with about 4% as many trainable parameters; on Qwen2.5-7B, measurements show 20% lower peak training memory and 6% higher throughput. These results are reported for the described settings and do not establish performance across other models or tasks.

### Source excerpt

arXiv:2611.17784v1 Announce Type: new Abstract: Parameter-efficient fine-tuning (PEFT) aims to preserve adaptation quality while reducing training costs. We propose SCALAR, which achieves performance comparable to LoRA with substantially fewer trainable parameters, lower peak training memory, and higher throughput. SCALAR constructs a fixed input basis from task activations and assigns a random output dictionary to each input direction. Fine-tuning learns only layer-specific scalar coefficients, reducing gradient and optimizer-state storage. Grouped computation aggregates output atoms before applying the adapter to token activations, while dictionary sharing across compatible layers reduces fixed-basis storage. Experiments on commonsense reasoning, mathematical reasoning, code generation, and knowledge evaluation after instruction tuning demonstrate effective adaptation at small trainable budgets. On commonsense reasoning, SCALAR achieves accuracy comparable to LoRA with about 4% of its trainable parameters. System measurements on Qwen2.5-7B show 20% lower peak training memory and 6% higher throughput than LoRA. The combination of adaptation quality, a compact trainable state, and training efficiency makes SCALAR a promising approach to resource-constrained fine-tuning.

## Disentangling Forgetting from Utility: Lightweight LLM Unlearning via Orthogonal Null-Space Projection

DevFeed: [Disentangling Forgetting from Utility: Lightweight LLM Unlearning via Orthogonal Null-Space Projection](<https://devfeed.tech/articles/disentangling-forgetting-from-utility-lightweight-llm-unlearning-via-orthogonal-null-space-projection-135836.md>)

Original publisher: [Read original article](<https://arxiv.org/abs/2611.17864>)

Author: Xiuyuan Wang, Weiming Liu, Zongxin Yang, Fan Wang, Liemiao Jin, Rubin Qin, Wei Chen, Bin Shi, Chaochao Chen, Xiaolin Zheng

Published: 2026-11-16T05:00:00Z

Content type: article

Language: en

Sources: [arXiv: Machine Learning](<https://devfeed.tech/sources/arxiv-machine-learning.md>)

Topics: [LLM security](<https://devfeed.tech/topics/llm-security.md>), [Large Language Model](<https://devfeed.tech/topics/llm.md>), [Fine-tuning](<https://devfeed.tech/topics/fine-tuning.md>), [peft](<https://devfeed.tech/topics/peft.md>), [Embeddings](<https://devfeed.tech/topics/embeddings.md>), [Training AI Models](<https://devfeed.tech/topics/training-ai-models.md>)

Tags: [adapters](<https://devfeed.tech/tags/adapters.md>), [arxiv](<https://devfeed.tech/tags/arxiv.md>), [balance](<https://devfeed.tech/tags/balance.md>), [benchmarks](<https://devfeed.tech/tags/benchmarks.md>), [capabilities](<https://devfeed.tech/tags/capabilities.md>), [convex](<https://devfeed.tech/tags/convex.md>), [cs-lg](<https://devfeed.tech/tags/cs-lg.md>), [deployment](<https://devfeed.tech/tags/deployment.md>), [disrupt](<https://devfeed.tech/tags/disrupt.md>), [embeddings](<https://devfeed.tech/tags/embeddings.md>), [fine-tuning](<https://devfeed.tech/tags/fine-tuning.md>), [information](<https://devfeed.tech/tags/information.md>), [knowledge](<https://devfeed.tech/tags/knowledge.md>), [language](<https://devfeed.tech/tags/language.md>), [language-models](<https://devfeed.tech/tags/language-models.md>), [large](<https://devfeed.tech/tags/large.md>), [large-language-models-llms](<https://devfeed.tech/tags/large-language-models-llms.md>), [lightweight](<https://devfeed.tech/tags/lightweight.md>), [llm](<https://devfeed.tech/tags/llm.md>), [projection](<https://devfeed.tech/tags/projection.md>), [utility](<https://devfeed.tech/tags/utility.md>)

### AI overview

UUE frames LLM unlearning as controllable model editing: it projects updates into the orthogonal left null space of retained-knowledge representations to limit interference with general capabilities, and uses a convex untargeted objective with a closed-form solution that pushes forget-set embeddings toward antipodal points. Lightweight pluggable adapters, including low-rank UUE-L, operate without modifying the frozen backbone. The abstract reports that UUE and UUE-L outperform existing methods on unlearning efficacy in TOFU and WMDP experiments; it provides no quantitative results or details on utility tradeoffs.

### Source excerpt

arXiv:2611.17864v1 Announce Type: new Abstract: Large Language Models (LLMs) pose significant security risks by retaining private or hazardous information, creating an urgent need for unlearning techniques that excise specific knowledge without compromising general utility. However, prevailing fine-tuning-based unlearning methods struggle to achieve this balance, often suffering from computational inefficiency and unintended utility degradation due to the entangled nature of knowledge within model parameters. To disentangle forgetting from utility, we propose UUE, a theoretically grounded framework that reformulates unlearning as controllable model editing. First, UUE projects parameter updates into the orthogonal left null space of retained knowledge representations, strictly ensuring that the editing process causes minimal interference to the model's general capabilities. Second, distinct from targeted mapping, we introduce a convex untargeted objective that derives a closed-form analytical solution to effectively disrupt undesired knowledge by pushing forget-set embeddings toward their antipodal points. We implement UUE via lightweight, pluggable adapters (including a low-rank variant, UUE-L), enabling efficient deployment without modifying the frozen backbone. Extensive experiments on TOFU and WMDP benchmarks demonstrate that UUE and UUE-L achieve superior unlearning efficacy, significantly outperforming existing methods.

## Adaptable Watermarking: Transforming Image Watermarking Models via Lightweight Adapters

DevFeed: [Adaptable Watermarking: Transforming Image Watermarking Models via Lightweight Adapters](<https://devfeed.tech/articles/adaptable-watermarking-transforming-image-watermarking-models-via-lightweight-adapters-135888.md>)

Original publisher: [Read original article](<https://arxiv.org/abs/2611.18363>)

Author: Mehrdad Saberi, Vishal Asnani, Shruti Agarwal, Soheil Feizi, John Collomosse

Published: 2026-11-16T05:00:00Z

Content type: article

Language: en

Sources: [arXiv: Cryptography and Security](<https://devfeed.tech/sources/arxiv-cryptography-and-security.md>), [arXiv: Machine Learning](<https://devfeed.tech/sources/arxiv-machine-learning.md>)

Topics: [peft](<https://devfeed.tech/topics/peft.md>), [Image processing](<https://devfeed.tech/topics/image-processing.md>), [Image](<https://devfeed.tech/topics/image.md>)

Tags: [adapter](<https://devfeed.tech/tags/adapter.md>), [adapters](<https://devfeed.tech/tags/adapters.md>), [alignment](<https://devfeed.tech/tags/alignment.md>), [cs-cr](<https://devfeed.tech/tags/cs-cr.md>), [cs-cv](<https://devfeed.tech/tags/cs-cv.md>), [cs-lg](<https://devfeed.tech/tags/cs-lg.md>), [image](<https://devfeed.tech/tags/image.md>), [lightweight](<https://devfeed.tech/tags/lightweight.md>), [models](<https://devfeed.tech/tags/models.md>), [watermarking](<https://devfeed.tech/tags/watermarking.md>)

### AI overview

The paper presents lightweight adapters for converting encoder-decoder image watermarking models to other watermarking styles without retraining full models. Fine-tuning aligns residuals across spatial, gradient, and frequency domains, while a secret-loss term preserves message bit accuracy. Evaluations involving TrustMark, InvisMark, RoSteALS, HiDDeN, and WAM report that adapted encoders closely reproduce target residuals and adapted decoders achieve near-optimal message recovery under standard augmentations and distortions. The approach uses as few as 1-2 million trainable parameters per component; the described multi-decoder inference pipeline uses a 2-million-parameter decoder adapter.

### Source excerpt

arXiv:2611.18363v1 Announce Type: cross Abstract: We explore the adaptability of neural image watermarking models; the extent to which one watermarking method can be transformed into another through lightweight adaptation. We introduce a framework that enables encoder-decoder watermarking architectures to become modular and upgradable, supporting multiple watermarking styles without re-training full models. Adaptation is performed by attaching a compact parameter-efficient adapter and fine-tuning under a multi-domain residual alignment loss (spatial, gradient, and frequency) while preserving bit accuracy via a secret-loss term. We evaluate this approach across a suite of recent watermarking models, including TrustMark, InvisMark, RoSteALS, HiDDeN, and WAM, demonstrating that adapted encoders closely reproduce target residuals and that adapted decoders recover messages with near-optimal accuracy under standard augmentations and distortions. We show that cross-model adaptation, with applications in interoperable and upgradable watermarking, can be achieved with as few as 1-2M trainable parameters per component (e.g., the 2M decoder adapter used in our multi-decoder inference pipeline).

## LoRAWeaver: Directional Adapter Routing in LoRA Representation Space

DevFeed: [LoRAWeaver: Directional Adapter Routing in LoRA Representation Space](<https://devfeed.tech/articles/loraweaver-directional-adapter-routing-in-lora-representation-space-135056.md>)

Original publisher: [Read original article](<https://arxiv.org/abs/2611.17340>)

Author: Inish Melroy D Souza, Tom Jackson, Vitor Castro, Rebecca Higginson, Andrea Soltoggio

Published: 2026-11-16T05:00:00Z

Content type: article

Language: en

Sources: [arXiv: Artificial Intelligence](<https://devfeed.tech/sources/arxiv-artificial-intelligence.md>)

Topics: [lora](<https://devfeed.tech/topics/lora.md>), [peft](<https://devfeed.tech/topics/peft.md>), [Fine-tuning](<https://devfeed.tech/topics/fine-tuning.md>), [cosine similarity search](<https://devfeed.tech/topics/cosine-similarity-search.md>), [Embeddings](<https://devfeed.tech/topics/embeddings.md>)

Tags: [adapter](<https://devfeed.tech/tags/adapter.md>), [alignment](<https://devfeed.tech/tags/alignment.md>), [benchmark](<https://devfeed.tech/tags/benchmark.md>), [cosine-similarity](<https://devfeed.tech/tags/cosine-similarity.md>), [cs-ai](<https://devfeed.tech/tags/cs-ai.md>), [fine-tuning](<https://devfeed.tech/tags/fine-tuning.md>), [inference](<https://devfeed.tech/tags/inference.md>), [language-models](<https://devfeed.tech/tags/language-models.md>), [lora](<https://devfeed.tech/tags/lora.md>), [prototypes](<https://devfeed.tech/tags/prototypes.md>), [routing](<https://devfeed.tech/tags/routing.md>)

### AI overview

LoRAWeaver routes queries to task-specific LoRA adapters without training a router. It represents each query conditioned on candidate adapters, then selects the adapter whose task prototype has the highest cosine similarity. The approach targets dynamic adapter pools, where existing routing and encoder methods may need retraining as adapters are added. On LLaMA-2-7B and LLaMA-2-13B, evaluations across 48 FLAN-v2 tasks and out-of-distribution tasks report better cross-task performance than training-free baselines and results approaching a trained retrieval encoder. An ablation supports the value of directional alignment in adapter-output space.

### Source excerpt

arXiv:2611.17340v1 Announce Type: new Abstract: Large language models are expensive to update through full retraining. Parameter-efficient fine-tuning reduces the cost of adaptation, but repeated adaptation creates a growing pool of task-specific LoRA adapters. This pool complicates adapter selection during inference. Existing routing and encoder-based methods often require retraining when new adapters are added, which limits their use in dynamic settings. This paper treats adapter routing as a representation alignment problem in the adapter-output space. The proposed training-free method selects adapters by matching adapter-conditioned query representations against task prototypes using cosine similarity. Experiments with LLaMA-2-7B and LLaMA-2-13B on a 48-task FLAN-v2 benchmark and on out-of-distribution tasks show that the method outperforms training-free baselines in the cross-task setting and approaches a trained retrieval encoder without learning a router. An ablation confirms the benefit of directional alignment in the adapter-output space.

## Predicting Ethanol Drinking Status from Nonhuman Primate Liver Histology: A Proof of Concept for Parameter-efficient Fine-tuning in Few-Shot Vision-Language Modeling

DevFeed: [Predicting Ethanol Drinking Status from Nonhuman Primate Liver Histology: A Proof of Concept for Parameter-efficient Fine-tuning in Few-Shot Vision-Language Modeling](<https://devfeed.tech/articles/predicting-ethanol-drinking-status-from-nonhuman-primate-liver-histology-a-proof-of-concept-for-parameter-efficient-fine-tuning-in-few-shot-vision-language-modeling-132312.md>)

Original publisher: [Read original article](<https://arxiv.org/abs/2611.17105>)

Author: Amirmohammad Mohammadi, Rupak Khadka, Kathleen A. Grant, Mary Lauren Benton, Joshua Peeples

Published: 2026-11-13T05:00:00Z

Content type: article

Language: en

Sources: [arXiv: Artificial Intelligence](<https://devfeed.tech/sources/arxiv-artificial-intelligence.md>)

Topics: [peft](<https://devfeed.tech/topics/peft.md>), [few-shot](<https://devfeed.tech/topics/few-shot.md>), [Fine-tuning](<https://devfeed.tech/topics/fine-tuning.md>), [vlm](<https://devfeed.tech/topics/vlm.md>), [image recognition](<https://devfeed.tech/topics/image-recognition.md>), [Machine learning](<https://devfeed.tech/topics/machine-learning.md>), [Computer vision](<https://devfeed.tech/topics/computer-vision.md>)

Tags: [abstract](<https://devfeed.tech/tags/abstract.md>), [adapter](<https://devfeed.tech/tags/adapter.md>), [adapters](<https://devfeed.tech/tags/adapters.md>), [after](<https://devfeed.tech/tags/after.md>), [alcohol](<https://devfeed.tech/tags/alcohol.md>), [all](<https://devfeed.tech/tags/all.md>), [analysis](<https://devfeed.tech/tags/analysis.md>), [approach](<https://devfeed.tech/tags/approach.md>), [arxiv](<https://devfeed.tech/tags/arxiv.md>), [attention](<https://devfeed.tech/tags/attention.md>), [based](<https://devfeed.tech/tags/based.md>), [cs-ai](<https://devfeed.tech/tags/cs-ai.md>), [cs-cv](<https://devfeed.tech/tags/cs-cv.md>), [few-shot](<https://devfeed.tech/tags/few-shot.md>), [fine](<https://devfeed.tech/tags/fine.md>), [fine-tuning](<https://devfeed.tech/tags/fine-tuning.md>), [language](<https://devfeed.tech/tags/language.md>), [modeling](<https://devfeed.tech/tags/modeling.md>), [parameter](<https://devfeed.tech/tags/parameter.md>), [proof-of-concept](<https://devfeed.tech/tags/proof-of-concept.md>), [status](<https://devfeed.tech/tags/status.md>), [vision](<https://devfeed.tech/tags/vision.md>)

### AI overview

In this proof of concept using a subset of the Monkey Alcohol Tissue Research Resource, parameter-efficient adapters improved mean balanced accuracy over non-PEFT baselines for few-shot prediction of ethanol drinking status from nonhuman-primate liver whole-slide images. In the four-shot setting, ABMIL alone achieved 43.5%, compared with 78.1% with an adapter. Textual pathway adaptations produced more localized attention heatmaps, while visual pathway adaptations produced more diffuse ones; a pathologist identified similar histological features across adaptations. The evidence is limited to the supplied abstract and does not establish performance beyond this dataset and study.

### Source excerpt

arXiv:2611.17105v1 Announce Type: cross Abstract: Combining vision-language models (VLMs) with parameter-efficient fine-tuning (PEFT) offers a promising approach to few-shot whole-slide image (WSI) classification in computational pathology. This study evaluates this approach for predicting ethanol drinking status from nonhuman primate liver histology. Adapters are evaluated against non-PEFT configurations on a subset of the Monkey Alcohol Tissue Research Resource (MATRR). All evaluated PEFT configurations achieve higher mean balanced accuracy than the non-PEFT baselines. For example, in the four-shot setting, attention-based multiple instance learning (ABMIL) alone achieves 43.5% mean balanced accuracy, compared with 78.1% after adding an adapter. Qualitative attention analysis shows that textual pathway adaptations produce localized attention heatmaps, while visual pathway adaptations produce more diffuse ones. A pathologist's review identified similar histological features across adaptations.

[Next page](<https://devfeed.tech/topics/peft.md?cursor=WyIyMDI2LTExLTEzVDA1OjAwOjAwKzAwOjAwIiwgImRhN2ViMGRiLTYyYzAtNGU0OS1hYTJiLTMyYzZmMzgwZTFiOCJd>)