# lora

LoRA is a low-rank adaptation method and Python package for efficiently fine-tuning large language models with PyTorch.

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

## Some Memories Must Fade: Persistence and Decay of User Memories in Model Parameters

DevFeed: [Some Memories Must Fade: Persistence and Decay of User Memories in Model Parameters](<https://devfeed.tech/articles/some-memories-must-fade-persistence-and-decay-of-user-memories-in-model-parameters-145026.md>)

Original publisher: [Read original article](<https://arxiv.org/abs/2611.25235>)

Author: Sanidhya Vijayvargiya, Rahul Lokesh

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

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: [AI Context Management](<https://devfeed.tech/topics/ai-context-management.md>), [lora](<https://devfeed.tech/topics/lora.md>), [Retrieval Augmented Generation (RAG)](<https://devfeed.tech/topics/retrieval-augmented-generation-rag.md>), [LLM evaluation / benchmarking](<https://devfeed.tech/topics/llm-evaluation-benchmarking.md>)

Tags: [adapter](<https://devfeed.tech/tags/adapter.md>), [behavioral-shifts](<https://devfeed.tech/tags/behavioral-shifts.md>), [capacity](<https://devfeed.tech/tags/capacity.md>), [cl](<https://devfeed.tech/tags/cl.md>), [continual-learning](<https://devfeed.tech/tags/continual-learning.md>), [controller](<https://devfeed.tech/tags/controller.md>), [cs-ai](<https://devfeed.tech/tags/cs-ai.md>), [cs-lg](<https://devfeed.tech/tags/cs-lg.md>), [dataset](<https://devfeed.tech/tags/dataset.md>), [horizon](<https://devfeed.tech/tags/horizon.md>), [knowledge](<https://devfeed.tech/tags/knowledge.md>), [language-models](<https://devfeed.tech/tags/language-models.md>), [large-language-models](<https://devfeed.tech/tags/large-language-models.md>), [lifecycle](<https://devfeed.tech/tags/lifecycle.md>), [lora](<https://devfeed.tech/tags/lora.md>), [manage](<https://devfeed.tech/tags/manage.md>), [management](<https://devfeed.tech/tags/management.md>), [memory](<https://devfeed.tech/tags/memory.md>), [memory-management](<https://devfeed.tech/tags/memory-management.md>), [model-weights](<https://devfeed.tech/tags/model-weights.md>), [persistence](<https://devfeed.tech/tags/persistence.md>), [user](<https://devfeed.tech/tags/user.md>)

### AI overview

The paper proposes managing user memories directly in model weights, with a fixed-capacity LoRA adapter for storing facts and a learned lightweight controller for deciding what to retain, update, reinforce, or let fade. The adapter uses k-hot rank combinations intended to reduce catastrophic interference and supports independently controlled temporal decay. Evaluation on controlled long-horizon user trajectories tests recall while facts remain valid and abstention after they expire. The learned policy reaches 86% of oracle performance by balancing recall and abstention; the authors report that continual-learning and retrieval-augmented-generation baselines tend to default mainly to one behavior. These findings are from the described dataset and evaluation.

### Source excerpt

arXiv:2611.25235v1 Announce Type: new Abstract: Parametric memory allows Large Language Models to internalize knowledge directly into their weights, offering a potentially powerful alternative to retrieval-based memory (RAG). However, existing approaches to parametric memory are difficult to deploy as they largely mirror the Continual Learning (CL) objective of retaining information indefinitely. This is neither sustainable nor sufficient for a memory system, which must govern knowledge across its lifecycle by retaining stable facts, reinforcing recurring observations, updating changed beliefs, and allowing transient information to fade. We propose a system for parametric memory management consisting of a memory substrate (handling information storage and forgetting) and a learned policy (what to remember, how deeply, and for how long). Our substrate is a fixed-capacity LoRA adapter that consists of k-hot rank combinations which store facts without catastrophic interference and maintain an independently controlled temporal decay, enabling outdated information to fade while long-term facts persist. On this substrate, we train a lightweight controller that governs each incoming user event to ensure the memory system reflects the user's true, updated belief state. To evaluate how memory systems manage user memory, we introduce a controlled dataset of long-horizon user trajectories with varying persistence of facts based on changing belief states and user-behavioral shifts, and probe queries that test recall while information is valid and abstention after it expires. Our policy learns to effectively manage user memory over its entire lifecycle, balancing recall and abstention to achieve 86% of oracle performance, unlike CL and RAG baselines that primarily default to one of the two behaviors. By providing the first system to manage memory directly in model weights, this work establishes a foundation for long-horizon parametric memory.

## Can LLMs Resolve Entity Conflicts in Domain Specific NLU Pipelines?

DevFeed: [Can LLMs Resolve Entity Conflicts in Domain Specific NLU Pipelines?](<https://devfeed.tech/articles/can-llms-resolve-entity-conflicts-in-domain-specific-nlu-pipelines-145095.md>)

Original publisher: [Read original article](<https://arxiv.org/abs/2611.25733>)

Author: Akansha Bhardwaj, Mingbo Cui, Victor Guimaraes, Dominik Kohler, Diana Nicoleta Popa, Meghana M. Reddy, Paolo Rosso

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

Content type: comparison

Language: en

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

Topics: [Large Language Model](<https://devfeed.tech/topics/llm.md>), [Entity resolution](<https://devfeed.tech/topics/entity-resolution.md>), [Fine-tuning](<https://devfeed.tech/topics/fine-tuning.md>), [lora](<https://devfeed.tech/topics/lora.md>), [Enhancing AI model performance](<https://devfeed.tech/topics/enhancing-ai-model-performance.md>)

Tags: [cases](<https://devfeed.tech/tags/cases.md>), [comparison](<https://devfeed.tech/tags/comparison.md>), [complex](<https://devfeed.tech/tags/complex.md>), [conflict-resolution](<https://devfeed.tech/tags/conflict-resolution.md>), [cs-ir](<https://devfeed.tech/tags/cs-ir.md>), [cs-lg](<https://devfeed.tech/tags/cs-lg.md>), [evaluation](<https://devfeed.tech/tags/evaluation.md>), [fine-tuning](<https://devfeed.tech/tags/fine-tuning.md>), [flexibility](<https://devfeed.tech/tags/flexibility.md>), [llama3](<https://devfeed.tech/tags/llama3.md>), [llms](<https://devfeed.tech/tags/llms.md>), [lora](<https://devfeed.tech/tags/lora.md>), [pipelines](<https://devfeed.tech/tags/pipelines.md>), [prompting](<https://devfeed.tech/tags/prompting.md>), [strategies](<https://devfeed.tech/tags/strategies.md>)

### AI overview

The study compares rule-based entity conflict resolution with two Llama3.1-8B-Instruct approaches in domain-specific NLU pipelines: multi-shot prompting and LoRA fine-tuning. It evaluates how well the models match existing pipeline outputs and, separately, their accuracy on human-labeled data. The abstract reports that prompting is easy to adapt and allows rapid behavioral changes, while fine-tuning performs better on complex, ambiguous cases. It suggests combining the approaches to pair prompting's flexibility with fine-tuning's broader accuracy gains.

### Source excerpt

arXiv:2611.25733v1 Announce Type: cross Abstract: Resolving entity conflicts is a critical step in Pipeline-based NLU architectures, especially when multiple extractors provide overlapping or conflicting outputs. Traditional rule-based entity conflict resolution methods often lack flexibility and are unscalable. In this work, we aim to study if the broad knowledge of LLMs can be leveraged to resolve entity conflicts in domain specific NLU pipelines. We present an experimental comparison between a traditional rule-based conflict resolution method, and two LLM-based approaches: a multi-shot prompting method using the Llama3.1-8B-Instruct model and a LoRA fine-tuned Llama3.1-8B-Instruct model. The evaluation was conducted under two setups: (1) a broad assessment of LLM's capability to match existing NLU pipeline output, and (2) a more controlled scenario assessing the LLM accuracy on human-labeled data. Results show that while multi-shot prompting offers ease of adaptation and rapid behavioral changes, the fine-tuned model delivers superior performance on complex, ambiguous cases. Ultimately, the best approach appears to be a combination of both strategies, leveraging multi-shot prompting's flexibility alongside the broad accuracy gains of LoRA fine-tuning for a more comprehensive performance.

## From "Weak" Signals to Strong Models: Preference Delta Aggregation with LoRA Merging

DevFeed: [From "Weak" Signals to Strong Models: Preference Delta Aggregation with LoRA Merging](<https://devfeed.tech/articles/from-weak-signals-to-strong-models-preference-delta-aggregation-with-lora-merging-145341.md>)

Original publisher: [Read original article](<https://arxiv.org/abs/2606.00357>)

Author: Qi Sun, Siyue Zhang, Yulin Chen, Yuxiang Xue, Ru Peng, Chen Zhao

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

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>), [rlhf](<https://devfeed.tech/topics/rlhf.md>), [post-training](<https://devfeed.tech/topics/post-training.md>), [llm-reasoning](<https://devfeed.tech/topics/llm-reasoning.md>), [qwen](<https://devfeed.tech/topics/qwen.md>)

Tags: [adapter](<https://devfeed.tech/tags/adapter.md>), [adapters](<https://devfeed.tech/tags/adapters.md>), [agentic](<https://devfeed.tech/tags/agentic.md>), [aggregation](<https://devfeed.tech/tags/aggregation.md>), [alignment](<https://devfeed.tech/tags/alignment.md>), [capabilities](<https://devfeed.tech/tags/capabilities.md>), [cs-ai](<https://devfeed.tech/tags/cs-ai.md>), [data](<https://devfeed.tech/tags/data.md>), [delta](<https://devfeed.tech/tags/delta.md>), [evaluations](<https://devfeed.tech/tags/evaluations.md>), [knowledge](<https://devfeed.tech/tags/knowledge.md>), [language-models](<https://devfeed.tech/tags/language-models.md>), [large-language-models-llms](<https://devfeed.tech/tags/large-language-models-llms.md>), [llms](<https://devfeed.tech/tags/llms.md>), [lora](<https://devfeed.tech/tags/lora.md>), [models](<https://devfeed.tech/tags/models.md>), [optimization](<https://devfeed.tech/tags/optimization.md>), [pairs](<https://devfeed.tech/tags/pairs.md>), [pda](<https://devfeed.tech/tags/pda.md>), [qwen3](<https://devfeed.tech/tags/qwen3.md>), [reasoning](<https://devfeed.tech/tags/reasoning.md>)

### AI overview

Preference Delta Aggregation (PDA) turns preference comparisons between weaker models into separate LoRA adapters trained through preference optimization, then merges them to improve a stronger model. Its Geometric Alignment Merging (GAM) aligns the adapters' left and right singular bases with orthogonal Procrustes rotations before averaging their factors. In evaluations on knowledge reasoning and agentic search, PDA-GAM outperformed the strongest individual delta. Three reasoning deltas and four agentic deltas improved Qwen3-8B's average scores by 6.8 and 7.3 points, respectively--2.1 and 4.3 points above the strongest single-delta baselines. The abstract reports similar gains on a second student model and attributes the results to combining complementary capabilities.

### Source excerpt

arXiv:2606.00357v3 Announce Type: replace Abstract: Training strong large language models (LLMs) requires high-quality supervision, which is often scarce. Recent work shows that paired preference data from weak-weaker model pairs (e.g., Qwen3 4B over 1.7B), despite the limited quality of individual responses, can provide an effective supervision signal through relative quality deltas, which we term a "weak" signal. This motivates a key research question: can multiple "weak" signals be constructively aggregated for improving strong models (e.g., Qwen3 8B)? To this end, we propose Preference Delta Aggregation (PDA), the first framework that derives a preference delta from each weak-weaker model pair, instantiates it as a LoRA adapter learned through preference optimization, and aggregates the resulting deltas via LoRA merging. To improve the composition of independently trained adapters, we introduce Geometric Alignment Merging (GAM), which aligns their left and right singular bases through orthogonal Procrustes rotations and averages the aligned factors. Evaluations on knowledge reasoning and agentic search show that PDA-GAM exceeds the strongest individual delta, with further gains along the evaluated sequences of additional deltas. Using three reasoning deltas and four agentic deltas, PDA-GAM improves Qwen3-8B by 6.8 and 7.3 points on average, respectively, exceeding the strongest single-delta baseline by 2.1 and 4.3 points. These configurations achieve the highest average scores among the evaluated merging methods, with similar gains observed on a second student model. Further analysis suggests that these gains come from composing complementary capabilities encoded across distinct preference deltas. Code is available at https://github.com/AlbertQiSun/Preference-Delta-Aggregation.

## Can an LLM Fix a Language Model Without Training It?

DevFeed: [Can an LLM Fix a Language Model Without Training It?](<https://devfeed.tech/articles/can-an-llm-fix-a-language-model-without-training-it-145551.md>)

Original publisher: [Read original article](<https://arxiv.org/abs/2611.25613>)

Author: Bharath Kumar Bolla, Raghavender Surya Upadhayula, Vishnu Surya Reddy Nandi, Ramesh Nandipalli

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

Content type: article

Language: en

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

Topics: [Large Language Model](<https://devfeed.tech/topics/llm.md>), [Transformers](<https://devfeed.tech/topics/transformers.md>), [LLM evaluation / benchmarking](<https://devfeed.tech/topics/llm-evaluation-benchmarking.md>), [lora](<https://devfeed.tech/topics/lora.md>)

Tags: [activation](<https://devfeed.tech/tags/activation.md>), [attention](<https://devfeed.tech/tags/attention.md>), [black-box](<https://devfeed.tech/tags/black-box.md>), [controller](<https://devfeed.tech/tags/controller.md>), [cs-ai](<https://devfeed.tech/tags/cs-ai.md>), [interactive](<https://devfeed.tech/tags/interactive.md>), [language-model](<https://devfeed.tech/tags/language-model.md>), [language-models](<https://devfeed.tech/tags/language-models.md>), [llm](<https://devfeed.tech/tags/llm.md>), [low-rank](<https://devfeed.tech/tags/low-rank.md>), [models](<https://devfeed.tech/tags/models.md>), [predictions](<https://devfeed.tech/tags/predictions.md>), [projection](<https://devfeed.tech/tags/projection.md>), [recovery](<https://devfeed.tech/tags/recovery.md>), [repair](<https://devfeed.tech/tags/repair.md>), [scripted](<https://devfeed.tech/tags/scripted.md>), [search](<https://devfeed.tech/tags/search.md>), [selection](<https://devfeed.tech/tags/selection.md>), [sessions](<https://devfeed.tech/tags/sessions.md>), [study](<https://devfeed.tech/tags/study.md>), [training](<https://devfeed.tech/tags/training.md>)

### AI overview

The study finds that an external language model can help repair selected Transformer faults by choosing numerical projection edits from predictions, activation summaries, and weight statistics, without gradient updates. Across 20 sessions and five models, live selection matched black-box search on SmolLM2 and Qwen2.5 for a two-layer fault; on Pythia it reached 65.62% completion accuracy, versus 0% for black-box search and 66.15% for a fixed projection. For a compound attention and feed-forward fault, scripted and LLM-guided edits using the same operator family raised the five-checkpoint mean from 24.89% to 71.25%. The comparisons also show that feedback does not always account for recovery: operator coverage can explain gains, supervised adaptation sometimes performs better, and scripted projections on unmodified checkpoints produced small gains or losses. These conclusions concern the tested faults and completion tasks.

### Source excerpt

arXiv:2611.25613v1 Announce Type: new Abstract: Although an external language model can select and test repairs to another model's weights, behavior can also be recovered through fixed numerical edits. We assessed the value of interactive selection in a controlled study of Transformer repair. After inspecting predictions, activation summaries, and weight statistics, the editor selects projection recipes to apply to the target without gradient updates. We analyzed 20 recorded sessions involving five masked and causal language models, comparing the selected edits with fixed projections, scripted and matched-budget black-box search, and low-rank adaptation. Under the core two-layer fault, live selection matches black-box search on SmolLM2 and Qwen2.5. For Pythia, live selection achieves 65.62% completion accuracy, compared with 0% for black-box search and 66.15% for a fixed projection. Under a separate compound fault affecting attention and feed-forward weights, both scripted and LLM-guided selection increase the descriptive five-checkpoint mean from 24.89% to 71.25% using the same operator family. On unmodified checkpoints, scripted projection yields small gains or losses on a separate set of framed questions. The experiments identify cases in which feedback changes an edit decision, operator coverage accounts for recovery, or supervised adaptation performs better. Combined with measurements of parameter restoration and preservation, these comparisons establish the scope of controller-guided repair for the tested faults and completion tasks.

## Self-Rewarded Reinforcement Learning for Better-Structured Medical Explanations in Small Language Models

DevFeed: [Self-Rewarded Reinforcement Learning for Better-Structured Medical Explanations in Small Language Models](<https://devfeed.tech/articles/self-rewarded-reinforcement-learning-for-better-structured-medical-explanations-in-small-language-models-142819.md>)

Original publisher: [Read original article](<https://arxiv.org/abs/2611.24650>)

Author: Prakhar Mishra, Bhavya Yajush Awasthi, Tanuja, Nishant

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

Content type: article

Language: en

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

Topics: [Reinforcement learning](<https://devfeed.tech/topics/reinforcement-learning.md>), [llm-reasoning](<https://devfeed.tech/topics/llm-reasoning.md>), [LLM Techniques](<https://devfeed.tech/topics/llm-techniques.md>), [post-training](<https://devfeed.tech/topics/post-training.md>), [lora](<https://devfeed.tech/topics/lora.md>)

Tags: [cs-ai](<https://devfeed.tech/tags/cs-ai.md>), [cs-cl](<https://devfeed.tech/tags/cs-cl.md>), [language-models](<https://devfeed.tech/tags/language-models.md>), [medical](<https://devfeed.tech/tags/medical.md>), [reinforcement-learning](<https://devfeed.tech/tags/reinforcement-learning.md>), [structured](<https://devfeed.tech/tags/structured.md>)

### AI overview

The study finds that a lightweight, self-rewarded reinforcement learning step can make medical explanations from Mistral-7B longer and more organized while preserving their underlying medical content. It applies PPO with LoRA to 2,000 Meerkat-derived prompts, using a reward that favors length, lexical diversity, and punctuation and penalizes very short responses. About 170 training steps on one NVIDIA A100 produced a stable increase in average reward. The authors discuss both the promise and limitations of using this approach to shape explanation style.

### Source excerpt

arXiv:2611.24650v1 Announce Type: new Abstract: Small language models (SLMs) can run on commodity hardware and support privacy-preserving, on-premises clinical deployment, but their medical explanations are often short and poorly structured compared with large language models. Recent supervised approaches such as Meerkat transfer chain-of-thought reasoning into 7B-8B backbones, yet their training objective does not directly target the style or organization of the generated rationales. Building on the instruction-tuned Mistral-7B backbone, we investigate whether a lightweight reinforcement learning (RL) step can improve explanation structure without human preference labels.We apply Proximal Policy Optimization (PPO) with LoRA adaptation to 2,000 Meerkat-derived medical prompts, guided by a simple self-supervised reward that favors longer, lexically diverse, and well-punctuated responses while penalizing very short ones. Training for about 170 PPO steps on a single NVIDIA A100 GPU yields a stable increase in the average reward. Quantitative and qualitative comparisons show that the RL-tuned model produces longer, more organized, multi-point explanations while preserving the underlying medical content.We discuss the promise and the limitations of self-rewarded RL for shaping the style of medical explanations.

## Erased but Recoverable: On the Fragility of Concept Erasure Under Fine-Tuning

DevFeed: [Erased but Recoverable: On the Fragility of Concept Erasure Under Fine-Tuning](<https://devfeed.tech/articles/erased-but-recoverable-on-the-fragility-of-concept-erasure-under-fine-tuning-143146.md>)

Original publisher: [Read original article](<https://arxiv.org/abs/2611.24220>)

Author: Jun Li, Lizhi Xiong, Ziqiang Li, Yu Liu, Zenglin Shi

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

Content type: article

Language: en

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

Topics: [Fine-tuning](<https://devfeed.tech/topics/fine-tuning.md>), [stable-diffusion](<https://devfeed.tech/topics/stable-diffusion.md>), [lora](<https://devfeed.tech/topics/lora.md>), [Open Source Models & Datasets](<https://devfeed.tech/topics/open-source-models-datasets.md>)

Tags: [attention](<https://devfeed.tech/tags/attention.md>), [cs-cr](<https://devfeed.tech/tags/cs-cr.md>), [diffusion-models](<https://devfeed.tech/tags/diffusion-models.md>), [fine-tuning](<https://devfeed.tech/tags/fine-tuning.md>), [gradient](<https://devfeed.tech/tags/gradient.md>), [lora](<https://devfeed.tech/tags/lora.md>), [persistence](<https://devfeed.tech/tags/persistence.md>), [profiling](<https://devfeed.tech/tags/profiling.md>), [recovery](<https://devfeed.tech/tags/recovery.md>), [target](<https://devfeed.tech/tags/target.md>), [text-to-image](<https://devfeed.tech/tags/text-to-image.md>)

### AI overview

Concept erasure in text-to-image diffusion models can weaken or reverse after downstream fine-tuning such as DreamBooth, LoRA, or Custom Diffusion. The authors report that fine-tuning preserves certain high-contribution attention heads with stable output subspaces and response magnitudes, and that these heads in erased models can realign with their pretrained representations. Their proposed SHACE method profiles stability-critical heads and keeps their parameter blocks unchanged during erasure. Experiments reportedly show improved resistance to concept recovery across varied fine-tuning settings while maintaining target suppression and non-target generation quality.

### Source excerpt

arXiv:2611.24220v1 Announce Type: new Abstract: Existing concept erasure methods for text-to-image diffusion models primarily focus on suppressing target concepts while preserving non-target generation quality. However, in practical applications, open-source diffusion models are often further adapted through DreamBooth, LoRA, or Custom Diffusion. Such adaptations can reshape internal representations, thereby weakening or even reversing the erasure effects achieved by existing methods. We identify erasure persistence as a key yet underexplored dimension of concept erasure and analyze how downstream fine-tuning reshapes internal representations. Our analysis reveals that fine-tuning of the original model does not uniformly modify the model, but preserves certain high-contribution structures, particularly attention heads with stable output subspaces and response magnitudes. We also observe that, after downstream fine-tuning of concept-erased models, these heads' output representations more readily realign with their pretrained counterparts. Motivated by this observation, we propose Stable-Head-Aware Concept Erasure (SHACE), which improves erasure persistence by preserving stable attention structures during concept erasure. SHACE identifies stability-critical attention heads through Attention-Head Stability Profiling and keeps their corresponding parameter blocks unchanged during erasure via Stable-Head Gradient Preservation. Extensive experiments show that SHACE improves resistance to concept recovery across diverse fine-tuning settings while maintaining strong suppression and non-target generation quality.

## Finetuning VLAs for Dexterous Manipulation: The Right Checkpoint Is All You Need

DevFeed: [Finetuning VLAs for Dexterous Manipulation: The Right Checkpoint Is All You Need](<https://devfeed.tech/articles/finetuning-vlas-for-dexterous-manipulation-the-right-checkpoint-is-all-you-need-143430.md>)

Original publisher: [Read original article](<https://arxiv.org/abs/2611.23676>)

Author: Artur Kuramshin, Charlotte Cloutier, Michael Przystupa, Kirsty Ellis, Mohammad S. Akhter, Glen Berseth

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

Content type: article

Language: en

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

Topics: [Robot Manipulation](<https://devfeed.tech/topics/robot-manipulation.md>), [Fine-tuning](<https://devfeed.tech/topics/fine-tuning.md>), [lora](<https://devfeed.tech/topics/lora.md>), [Machine learning](<https://devfeed.tech/topics/machine-learning.md>)

Tags: [cs-lg](<https://devfeed.tech/tags/cs-lg.md>), [cs-ro](<https://devfeed.tech/tags/cs-ro.md>), [datasets](<https://devfeed.tech/tags/datasets.md>), [diffusion](<https://devfeed.tech/tags/diffusion.md>), [finetuning](<https://devfeed.tech/tags/finetuning.md>), [lora](<https://devfeed.tech/tags/lora.md>), [manipulation](<https://devfeed.tech/tags/manipulation.md>), [reinforcement-learning](<https://devfeed.tech/tags/reinforcement-learning.md>), [sft](<https://devfeed.tech/tags/sft.md>), [simulation](<https://devfeed.tech/tags/simulation.md>), [steering](<https://devfeed.tech/tags/steering.md>), [study](<https://devfeed.tech/tags/study.md>), [supervised-fine-tuning](<https://devfeed.tech/tags/supervised-fine-tuning.md>), [vision-language-action](<https://devfeed.tech/tags/vision-language-action.md>), [vision-language-model](<https://devfeed.tech/tags/vision-language-model.md>), [vla](<https://devfeed.tech/tags/vla.md>)

### AI overview

Using π₀.₅ and a LEAP hand, the study compares five VLA fine-tuning recipes on two simulation tasks with a handful of demonstrations. SFT rollout success poorly predicts which checkpoint will perform best after diffusion-steering reinforcement learning (DSRL). Freezing the vision-language model and fully tuning the action expert yields the most steerable policy, including for out-of-distribution placements; LoRA is least steerable. The authors recommend selecting checkpoints without relying on SFT rollout success. These findings are reported for the study's two simulated tasks.

### Source excerpt

arXiv:2611.23676v1 Announce Type: cross Abstract: Vision-Language-Action models have demonstrated strong manipulation capabilities. A limitation of VLA models is that most manipulation datasets focus on parallel-jaw gripper systems, making transfer to multi-finger hands challenging, particularly as teleoperation datasets are harder to collect with multi-fingered robots. In this work, we study how to best adapt VLAs to dexterous hands by pairing supervised fine-tuning (SFT) with diffusion-steering reinforcement learning (DSRL). Using $\pi_{0.5}$ and a LEAP hand, we evaluate five fine-tuning recipes across two simulation tasks using only a handful of demonstrations. Our empirical analysis yields three key insights: (1) SFT success is poorly correlated with post-DSRL success, meaning the best SFT checkpoint is rarely the most effective to steer; (2) freezing the vision-language model (VLM) and tuning only the action expert produces the most steerable policy for both training and out-of-distribution placements; and (3) low-rank adaptation (LoRA) is the least steerable approach. Consequently, we recommend an adaptation recipe that freezes the VLM, fully trains the action expert, and eschews SFT rollout success as a checkpoint selection metric.

## Same Rank, Different Needs: Rethinking LoRA Redistribution in Federated Learning

DevFeed: [Same Rank, Different Needs: Rethinking LoRA Redistribution in Federated Learning](<https://devfeed.tech/articles/same-rank-different-needs-rethinking-lora-redistribution-in-federated-learning-143440.md>)

Original publisher: [Read original article](<https://arxiv.org/abs/2611.23742>)

Author: Jinqian Chen, Jihua Zhu, Chang Liu

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

Content type: article

Language: en

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

Topics: [Federated Learning](<https://devfeed.tech/topics/federated-learning.md>), [lora](<https://devfeed.tech/topics/lora.md>), [Machine learning](<https://devfeed.tech/topics/machine-learning.md>), [AI Foundation Models](<https://devfeed.tech/topics/ai-foundation-models.md>), [Fine-tuning](<https://devfeed.tech/topics/fine-tuning.md>)

Tags: [cs-lg](<https://devfeed.tech/tags/cs-lg.md>), [federated](<https://devfeed.tech/tags/federated.md>), [federated-learning](<https://devfeed.tech/tags/federated-learning.md>), [learning](<https://devfeed.tech/tags/learning.md>), [lora](<https://devfeed.tech/tags/lora.md>), [rank](<https://devfeed.tech/tags/rank.md>)

### AI overview

In federated LoRA, keeping an aggregated update's leading singular directions may poorly preserve how that update responds to a particular client's inputs. The paper proposes Receiver-Aware Subspace Adaptation (RASA), which uses compact diagonal input second moments to select a rank-constrained readout by uncentered PCA of receiver-weighted responses. The authors report that this readout is optimal for their receiver objective and preserves at least as much response energy as spectral truncation under the diagonal metric. RASA also adds spectral-reference aggregation to remove the personalized initialization offset while retaining the locally learned increment. Experiments across DomainNet-100, NICO++, GLUE, and SuperNI report gains, including 3.23 percentage points in DomainNet-100 accuracy with FlexLoRA and up to 8.33 ROUGE-L points on seen SuperNI tasks and 4.37 on unseen tasks, with modest overhead.

### Source excerpt

arXiv:2611.23742v1 Announce Type: new Abstract: Federated low-rank adaptation (LoRA) enables clients with different resource budgets to collaboratively fine-tune foundation models. To distribute an aggregated update within each client's rank budget, existing methods commonly retain its leading singular directions. However, directions that best reconstruct the update matrix need not best preserve its responses on a particular client's inputs: a lower-energy direction may matter more to that recipient. We formalize this mismatch and show that matrix-optimal truncation can incur arbitrarily larger response error than another readout of the same rank. Motivated by this result, we propose Receiver-Aware Subspace Adaptation (RASA), a plug-in framework for receiver-aware knowledge transfer. RASA uses compact diagonal input second moments to perform uncentered PCA of receiver-weighted update responses. The resulting readout is globally optimal for the rank-constrained receiver objective and preserves at least as much response energy as spectral truncation under the diagonal metric. RASA also introduces spectral-reference aggregation, which removes the personalized initialization offset while retaining the locally learned increment. Experiments on DomainNet-100, NICO++, GLUE, and SuperNI demonstrate RASA's effectiveness across methods under resource, task, and data heterogeneity. In particular, RASA improves DomainNet-100 accuracy by 3.23 percentage points with FlexLoRA and achieves gains of up to 8.33 ROUGE-L points on seen SuperNI tasks and 4.37 on unseen tasks with modest overhead.

## 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.

## When Do Foundation Models Pay Off? A Break-Even Analysis of Pretrained Time Series Forecasters

DevFeed: [When Do Foundation Models Pay Off? A Break-Even Analysis of Pretrained Time Series Forecasters](<https://devfeed.tech/articles/when-do-foundation-models-pay-off-a-break-even-analysis-of-pretrained-time-series-forecasters-143286.md>)

Original publisher: [Read original article](<https://arxiv.org/abs/2607.04919>)

Author: Nicholas Tan Jerome, Frank Simon

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 Foundation Models](<https://devfeed.tech/topics/ai-foundation-models.md>), [Time Series](<https://devfeed.tech/topics/time-series.md>), [lora](<https://devfeed.tech/topics/lora.md>), [decision-making](<https://devfeed.tech/topics/decision-making.md>), [Predictive AI](<https://devfeed.tech/topics/predictive-ai.md>)

Tags: [analysis](<https://devfeed.tech/tags/analysis.md>), [arima](<https://devfeed.tech/tags/arima.md>), [benchmark](<https://devfeed.tech/tags/benchmark.md>), [cs-lg](<https://devfeed.tech/tags/cs-lg.md>), [data](<https://devfeed.tech/tags/data.md>), [dataset](<https://devfeed.tech/tags/dataset.md>), [datasets](<https://devfeed.tech/tags/datasets.md>), [decision](<https://devfeed.tech/tags/decision.md>), [ets](<https://devfeed.tech/tags/ets.md>), [fine-tuning](<https://devfeed.tech/tags/fine-tuning.md>), [fm](<https://devfeed.tech/tags/fm.md>), [forecasting](<https://devfeed.tech/tags/forecasting.md>), [foundation-model](<https://devfeed.tech/tags/foundation-model.md>), [foundation-models](<https://devfeed.tech/tags/foundation-models.md>), [gpu](<https://devfeed.tech/tags/gpu.md>), [gpu-infrastructure](<https://devfeed.tech/tags/gpu-infrastructure.md>), [history](<https://devfeed.tech/tags/history.md>), [models](<https://devfeed.tech/tags/models.md>), [series](<https://devfeed.tech/tags/series.md>), [time](<https://devfeed.tech/tags/time.md>), [time-series](<https://devfeed.tech/tags/time-series.md>)

### AI overview

Across 30 datasets, zero-shot and LoRA-tuned time-series foundation models beat classical baselines at every tested training fraction on 16 datasets, while classical methods reached break-even with as little as 21-6,324 samples on six others and 24-22,177 on eight. No classical method dominated. For datasets with fewer than 700 training points and non-negligible seasonality, the study recommends zero-shot forecasting and skipping fine-tuning: at under 20% of history, classical methods lost on four of ten such datasets and could not be fitted on six. LoRA rarely helped and degraded short-series performance by up to 52%. The proposed process checks dataset length and seasonality, then uses a brief 5-10% pilot when needed; at this benchmark scale, simple dataset features did not predict break-even outside the short-seasonal-series rule.

### Source excerpt

arXiv:2607.04919v2 Announce Type: replace Abstract: Deploying a time series foundation model (FM) requires GPU infrastructure and engineering overhead, with no guarantee of improvement over classical forecasting methods. We provide the first systematic break-even analysis of when this investment pays off. Across 30 benchmark datasets, we compare zero-shot and LoRA fine-tuned FMs (Chronos-Bolt, Moirai, Lag-Llama) against classical baselines (Naive, ETS, ARIMA, XGBoost) at six training set sizes from 2% to 100% of available data. FMs outperform classical methods at every evaluated training fraction on 16 of 30 datasets. On 6 datasets, classical methods surpass zero-shot FMs with at most 5% of the training data (21--6,324 samples); on the remaining 8, break-even ranges from 24 to 22,177 samples. No classical method dominates: ARIMA, Naive, ETS and XGBoost are each best on 6--11 datasets. One robust rule requires no model training: if n_train < 700 and seasonality is non-negligible, FM zero-shot is the right low-data choice -- below 20% of the history, classical methods either lose to it (4 of these 10 datasets) or cannot be fitted at all (6) -- and it stays best at full history on 6 of them; fine-tuning should be skipped. Contrary to common practice, LoRA fine-tuning rarely helps and degrades short series by up to 52%. We operationalise these findings as a two-step decision framework -- compute dataset length and seasonality strength, run a brief 5--10% pilot only if needed. Cheap dataset features do not predict the break-even category at this benchmark scale, so outside the rule a pilot remains necessary. Code and benchmark: https://github.com/nicolaisi/fm_breakeven.

## Learning through Discernment: when a small language model learns from user feedback on the job, and when it decays

DevFeed: [Learning through Discernment: when a small language model learns from user feedback on the job, and when it decays](<https://devfeed.tech/articles/learning-through-discernment-when-a-small-language-model-learns-from-user-feedback-on-the-job-and-when-it-decays-142695.md>)

Original publisher: [Read original article](<https://arxiv.org/abs/2611.23716>)

Author: Sirisilp Kongsilp

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

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: [Machine learning](<https://devfeed.tech/topics/machine-learning.md>), [Training AI Models](<https://devfeed.tech/topics/training-ai-models.md>), [Language models](<https://devfeed.tech/topics/language-models.md>), [lora](<https://devfeed.tech/topics/lora.md>), [rlhf](<https://devfeed.tech/topics/rlhf.md>), [On-device AI](<https://devfeed.tech/topics/on-device-ai.md>)

Tags: [adapter](<https://devfeed.tech/tags/adapter.md>), [cs-ai](<https://devfeed.tech/tags/cs-ai.md>), [cs-cl](<https://devfeed.tech/tags/cs-cl.md>), [cs-lg](<https://devfeed.tech/tags/cs-lg.md>), [feedback](<https://devfeed.tech/tags/feedback.md>), [job](<https://devfeed.tech/tags/job.md>), [language-model](<https://devfeed.tech/tags/language-model.md>), [learning](<https://devfeed.tech/tags/learning.md>), [small-language-model](<https://devfeed.tech/tags/small-language-model.md>), [user](<https://devfeed.tech/tags/user.md>)

### AI overview

In experiments with Qwen3 0.6B and 1.7B plus LoRA adapters, nightly learning from user feedback depended on whether the judge filtered out failures and whether users could see outcomes. A feedback sense that treated polite give-ups as success collapsed in 3 of 5 lives. A version that first checked for a clear user verdict stayed near oracle performance in 4 of 5 lives for two weeks, then faded to self-judging levels by day 60 as clean lessons became scarce. On τ²-bench, it interpreted reactions correctly, but 80% of lessons were failed attempts users had thanked; across 97 lives, the share of failed lessons strongly predicted outcomes (Spearman ρ = −0.81). The abstract concludes that user-guided learning held up only when failures were excluded and users could observe outcomes, and warns that skewed lessons may narrow the model.

### Source excerpt

arXiv:2611.23716v1 Announce Type: cross Abstract: Small language models on devices could keep improving by learning, overnight, from their own daily work. Without a true outcome, however, something must decide which attempts are worth learning from, and both the model's own judgment and the user's reaction can be wrong. We study a small model (Qwen3 0.6B and 1.7B) with a LoRA adapter, fine-tuned each night on the attempts a judge accepts, over 127 lives of up to a year on WebShop and $\tau^2$-bench. We compare the true outcome, the user's stated verdict and the model's own judgment with a separate, frozen "feedback sense" that reads the user's reaction, built in three versions. Judging itself, the model improved and then decayed, over a year and whatever its adapter. A sense that took polite give-ups for success collapsed 3 of 5 lives; one built for discernment, which first asks whether the user gave a clear verdict, kept 4 of 5 near the oracle for two weeks, then faded on clean but scarce lessons, to the level of self-judging by day 60. When the user cannot see the outcome ($\tau^2$-bench), the sense read reactions correctly, yet 80% of its lessons were failures the customer had thanked. Across 97 lives, the share of failed attempts among the lessons predicted the outcome (Spearman $\rho$ = -0.81). In our experiments, learning from users held up only where the judge kept failures out and the user could see the outcome; lopsided lessons may also narrow the model.

## LusoClin: A Synthetic Clinical Note Dataset for Medical Information Retrieval in European Portuguese

DevFeed: [LusoClin: A Synthetic Clinical Note Dataset for Medical Information Retrieval in European Portuguese](<https://devfeed.tech/articles/lusoclin-a-synthetic-clinical-note-dataset-for-medical-information-retrieval-in-european-portuguese-142657.md>)

Original publisher: [Read original article](<https://arxiv.org/abs/2611.23454>)

Author: Carla Teixeira Lopes, Daniel F\\'elix

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

Content type: article

Language: en

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

Topics: [information retrieval](<https://devfeed.tech/topics/information-retrieval.md>), [AI-driven information retrieval](<https://devfeed.tech/topics/ai-driven-information-retrieval.md>), [Fine-tuning](<https://devfeed.tech/topics/fine-tuning.md>), [lora](<https://devfeed.tech/topics/lora.md>), [llama](<https://devfeed.tech/topics/llama.md>), [Large Language Model](<https://devfeed.tech/topics/llm.md>), [managing large-scale text data in AI](<https://devfeed.tech/topics/managing-large-scale-text-data-in-ai.md>), [Open Source Models & Datasets](<https://devfeed.tech/topics/open-source-models-datasets.md>), [Internationalization (i18n)](<https://devfeed.tech/topics/i18n.md>)

Tags: [bert](<https://devfeed.tech/tags/bert.md>), [clinical-search](<https://devfeed.tech/tags/clinical-search.md>), [cs-ai](<https://devfeed.tech/tags/cs-ai.md>), [cs-ir](<https://devfeed.tech/tags/cs-ir.md>), [dataset](<https://devfeed.tech/tags/dataset.md>), [document-retrieval](<https://devfeed.tech/tags/document-retrieval.md>), [fine-tuning](<https://devfeed.tech/tags/fine-tuning.md>), [information-retrieval](<https://devfeed.tech/tags/information-retrieval.md>), [languages](<https://devfeed.tech/tags/languages.md>), [llama](<https://devfeed.tech/tags/llama.md>), [low-resource-languages](<https://devfeed.tech/tags/low-resource-languages.md>), [medical](<https://devfeed.tech/tags/medical.md>), [note](<https://devfeed.tech/tags/note.md>), [resource](<https://devfeed.tech/tags/resource.md>), [synthetic](<https://devfeed.tech/tags/synthetic.md>)

### AI overview

LusoClin is a publicly released dataset of synthetic clinical notes in European Portuguese, intended to support privacy-preserving research on clinical information retrieval in a language with limited access to real clinical text. It was created by preprocessing real clinical data, fine-tuning Llama 3.2 (3B) with Quantized Low-Rank Adaptation, and generating notes from prompts; a BERT-based model automatically annotates clinical entities for search and indexing. Evaluations using automatic metrics and human review indicate that the dataset captures relevant clinical language and may support document retrieval and domain-specific model adaptation. However, annotators can often distinguish its notes from real ones, and some medical inconsistencies remain.

### Source excerpt

arXiv:2611.23454v1 Announce Type: new Abstract: Access to large-scale clinical text collections is crucial for advancing medical information retrieval research, including clinical note search, patient cohort identification, and domain-adapted retrieval models. However, real clinical notes are rarely available due to privacy regulations, which particularly limits progress for low-resource languages such as European Portuguese. We present LusoClin, a publicly released dataset of synthetic clinical notes in European Portuguese, created to support privacy-preserving research on clinical text retrieval. LusoClin was generated by pre-processing real clinical data, fine-tuning the open-source Llama 3.2 (3B) instruction-tuned model using Quantized Low-Rank Adaptation, and applying prompt-based clinical note generation. To facilitate semantic indexing and clinical search, the notes are automatically annotated with clinical entities using a BERT-based model. We evaluate LusoClin in terms of utility, fidelity, diversity, and privacy using automatic metrics and human evaluation. Results indicate that the dataset captures clinically relevant language patterns and can support downstream IR applications such as clinical document retrieval and domain-specific model adaptation, although annotators can often distinguish synthetic from real notes and some medical inconsistencies persist.

## Atomic MathReasoner: Instance-Level Composition of Atomic Reasoning Skills for Mathematical LLMs

DevFeed: [Atomic MathReasoner: Instance-Level Composition of Atomic Reasoning Skills for Mathematical LLMs](<https://devfeed.tech/articles/atomic-mathreasoner-instance-level-composition-of-atomic-reasoning-skills-for-mathematical-llms-143572.md>)

Original publisher: [Read original article](<https://arxiv.org/abs/2611.24846>)

Author: Jiashu Li, Tianyu Gao, Qinglong Liu, Jinping Wang, Ye Wei, Zhiqiang Gao

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

Content type: article

Language: en

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

Topics: [llm-reasoning](<https://devfeed.tech/topics/llm-reasoning.md>), [Parallel reasoning](<https://devfeed.tech/topics/parallel-reasoning.md>), [lora](<https://devfeed.tech/topics/lora.md>), [Fine-tuning](<https://devfeed.tech/topics/fine-tuning.md>)

Tags: [amr](<https://devfeed.tech/tags/amr.md>), [atomic](<https://devfeed.tech/tags/atomic.md>), [benchmark](<https://devfeed.tech/tags/benchmark.md>), [capabilities](<https://devfeed.tech/tags/capabilities.md>), [composition](<https://devfeed.tech/tags/composition.md>), [confidence](<https://devfeed.tech/tags/confidence.md>), [cs-lg](<https://devfeed.tech/tags/cs-lg.md>), [gradient](<https://devfeed.tech/tags/gradient.md>), [inference](<https://devfeed.tech/tags/inference.md>), [llms](<https://devfeed.tech/tags/llms.md>), [performance](<https://devfeed.tech/tags/performance.md>), [prototypes](<https://devfeed.tech/tags/prototypes.md>), [reasoning](<https://devfeed.tech/tags/reasoning.md>), [router](<https://devfeed.tech/tags/router.md>), [skills](<https://devfeed.tech/tags/skills.md>)

### AI overview

Atomic MathReasoner (AMR) composes specialized reasoning skills per problem instead of treating mathematical reasoning as one capability. It trains capability-specific LoRA experts, searches held-out validation subsets without gradients for low-loss composition prototypes, scores them by token-level negative log-likelihood of reference solutions, filters uncertain assignments with a confidence margin, and distills the resulting labels into a lightweight router. At inference, the router selects a prototype without costly online optimization. On AtomicMath, the abstract reports consistent gains across general and math-specialized LLMs; for Qwen2.5-Math-Instruct-7B, average field accuracy rose from 64.4% to 68.1% at Level 1 and 59.3% to 63.7% at Level 2, while average logical reasoning performance rose from 39.4% to 46.8%.

### Source excerpt

arXiv:2611.24846v1 Announce Type: new Abstract: Most existing approaches enhance mathematical reasoning in large language models (LLMs) as a monolithic capability, despite the fact that different mathematical problems require different reasoning skills. Recent work has explored a more fine-grained perspective by decomposing mathematical reasoning into atomic skills and developing specialized modules for individual capabilities. However, how to effectively utilize these atomic abilities remains underexplored. To address this challenge, we propose Atomic MathReasoner (AMR), an instance-level framework for composing atomic reasoning skills. AMR first trains a library of capability-specific LoRA experts and performs gradient-free optimization over held-out validation subsets to discover a diverse set of low-loss composition prototypes. It evaluates the preference of each prototype using the token-level negative log-likelihood of reference solutions, filters ambiguous assignments through a confidence margin, and distills the resulting pseudo-labels into a lightweight router. During inference, the router selects an appropriate composition prototype for each problem, enabling problem-specific utilization of atomic skills without costly online optimization. Experiments on the AtomicMath benchmark across general and mathematics-specialized LLMs demonstrate consistent improvements. On Qwen2.5-Math-Instruct-7B, AMR increases average field accuracy from 64.4% to 68.1% at Level 1 and from 59.3% to 63.7% at Level 2, while improving average logical reasoning performance from 39.4% to 46.8%. These results demonstrate that effectively composing specialized atomic skills provides a promising approach to strengthening mathematical reasoning in LLMs.

## When Does Subspace Direction Matter for LoRA? Regime Analysis of the Magnitude Principle in Few-Shot Adaptation

DevFeed: [When Does Subspace Direction Matter for LoRA? Regime Analysis of the Magnitude Principle in Few-Shot Adaptation](<https://devfeed.tech/articles/when-does-subspace-direction-matter-for-lora-regime-analysis-of-the-magnitude-principle-in-few-shot-adaptation-143392.md>)

Original publisher: [Read original article](<https://arxiv.org/abs/2611.23428>)

Author: Nischal Subedi, Cencheng Shen, Peng Zhao

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

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>), [few-shot](<https://devfeed.tech/topics/few-shot.md>), [llm-reasoning](<https://devfeed.tech/topics/llm-reasoning.md>)

Tags: [activation](<https://devfeed.tech/tags/activation.md>), [alignment](<https://devfeed.tech/tags/alignment.md>), [bert](<https://devfeed.tech/tags/bert.md>), [classification](<https://devfeed.tech/tags/classification.md>), [completion](<https://devfeed.tech/tags/completion.md>), [covariance](<https://devfeed.tech/tags/covariance.md>), [cs-lg](<https://devfeed.tech/tags/cs-lg.md>), [dimension](<https://devfeed.tech/tags/dimension.md>), [evaluation](<https://devfeed.tech/tags/evaluation.md>), [few-shot](<https://devfeed.tech/tags/few-shot.md>), [llama](<https://devfeed.tech/tags/llama.md>), [lora](<https://devfeed.tech/tags/lora.md>), [magnitude](<https://devfeed.tech/tags/magnitude.md>), [span](<https://devfeed.tech/tags/span.md>), [stat-ml](<https://devfeed.tech/tags/stat-ml.md>), [structure](<https://devfeed.tech/tags/structure.md>), [subspace](<https://devfeed.tech/tags/subspace.md>), [theory](<https://devfeed.tech/tags/theory.md>), [training](<https://devfeed.tech/tags/training.md>)

### AI overview

The paper finds that LoRA initialization direction matters most when labeled data are scarce relative to representation dimension and the initialization subspace captures task structure. Its theory includes a finite-sample stability analysis and a first-step account that motivates an alignment coefficient, ρ, which tracks classification gains. PivotLoRA uses activation PCA at prediction-relevant positions, then applies an output-preserving orthogonal pivot early in training. In few-shot BERT-base classification it ranks best on average among 13 methods and beats vanilla LoRA in 29 of 30 settings despite a smaller input-factor initialization norm. Gains diminish with more labeled data; targeted experiments also report improvements on GSM8K and StrategyQA. The supplied text does not detail those evaluation setups or effect sizes.

### Source excerpt

arXiv:2611.23428v1 Announce Type: new Abstract: We show that the importance of a LoRA adapter's initial input-subspace direction is regime-dependent, and we characterize this dependence in theory and experiment. This refines the recent view that spectral initializers help through update magnitude alone. Direction matters most when labeled data are scarce relative to the representation dimension and the activation subspace used for initialization captures supervised task structure. In this regime, the adapter's row span remains sensitive to its starting point, making task-aligned initialization especially useful. A spiked-covariance perturbation analysis characterizes the finite-sample stability of the estimated subspace. A first-step analysis of LoRA training explains why the subspace's alignment with the task matters and motivates an alignment coefficient $\rho$ that closely tracks gains across classification tasks. We test this account with PivotLoRA, which initializes adapters from activation PCA at prediction-relevant positions and applies one output-preserving orthogonal pivot early in training. In few-shot BERT-base classification, PivotLoRA achieves the best average rank among 13 methods and improves over vanilla LoRA in 29 of 30 settings, even though it starts from a smaller input-factor initialization norm. Additional classification experiments extend the evaluation to 7B and 8B Llama decoders. Classification gains diminish with more labeled data, consistent with the proposed regime dependence. A targeted extension to completion-only reasoning also shows improvements on GSM8K and StrategyQA. These findings identify conditions under which initialization direction matters, with subspace-task alignment offering a low-cost diagnostic before training for classification.

## 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.

## Fine-Grained Japanese Clinical Named Entity Recognition: A Controlled Comparison of BERT and Open-Weight Large Language Models

DevFeed: [Fine-Grained Japanese Clinical Named Entity Recognition: A Controlled Comparison of BERT and Open-Weight Large Language Models](<https://devfeed.tech/articles/fine-grained-japanese-clinical-named-entity-recognition-a-controlled-comparison-of-bert-and-open-weight-large-language-models-139985.md>)

Original publisher: [Read original article](<https://arxiv.org/abs/2611.22219>)

Author: Feng Han, Ayako Yagahara, Kenji Hirata, Atsushi Yoshida, Kohsuke Kudo

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

Content type: article

Language: en

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

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

Tags: [bert](<https://devfeed.tech/tags/bert.md>), [comparison](<https://devfeed.tech/tags/comparison.md>), [cs-ai](<https://devfeed.tech/tags/cs-ai.md>), [cs-cl](<https://devfeed.tech/tags/cs-cl.md>), [fine-tuning](<https://devfeed.tech/tags/fine-tuning.md>), [japanese](<https://devfeed.tech/tags/japanese.md>), [json-schema](<https://devfeed.tech/tags/json-schema.md>), [language-models](<https://devfeed.tech/tags/language-models.md>), [large-language-models](<https://devfeed.tech/tags/large-language-models.md>), [llm](<https://devfeed.tech/tags/llm.md>), [llms](<https://devfeed.tech/tags/llms.md>), [medical](<https://devfeed.tech/tags/medical.md>), [named-entity-recognition](<https://devfeed.tech/tags/named-entity-recognition.md>), [ner](<https://devfeed.tech/tags/ner.md>), [performance](<https://devfeed.tech/tags/performance.md>), [prompting](<https://devfeed.tech/tags/prompting.md>), [qlora](<https://devfeed.tech/tags/qlora.md>)

### AI overview

On this historical Japanese case-report benchmark, fine-tuned open-weight LLMs with adequate output capacity outperformed BERT-CRF baselines, while zero-shot prompting and JSON Schema constraints alone did not. Across 433 test sentences and 113 entity categories, zero-shot relaxed F1 peaked at 0.337 versus 0.836-0.841 for BERT. With larger model-specific output limits, all eight fine-tuned LLMs with at least 13B parameters reached relaxed F1 of 0.882-0.914 and exact-surface F1 of 0.840-0.880. Increasing limits from 900 tokens improved relaxed F1 by 0.052-0.085, with higher recall and fewer truncated outputs. JSON Schema changed relaxed F1 only -0.014 to +0.008 for seven models; medical-specialized models showed no consistent edge over paired general models. Findings are limited to this benchmark and setup.

### Source excerpt

arXiv:2611.22219v1 Announce Type: new Abstract: Background: Generative large language models (LLMs) offer an alternative to encoder-based clinical named entity recognition (NER), but their performance on fine-grained Japanese extraction tasks and sensitivity to generation settings remain uncertain. Methods: We compared two fine-tuned BERT-CRF baselines with nine open-weight LLMs on iCorpus, which defines 113 entity categories. The evaluation used 433 sentences from 36 test documents; 1,672 sentences were used for training. LLMs were evaluated using zero-shot prompting with and without JSON Schema constraints and three-seed QLoRA fine-tuning. Fine-tuned models were assessed under different output-token limits, including 900 and larger model-specific limits of 2,600 or 4,096 tokens. Performance was measured using sentence-level label-surface matching with relaxed and exact-surface F1. Results: The highest zero-shot relaxed F1 was 0.337, compared with 0.836-0.841 for the BERT baselines. At the larger output limits, all eight fine-tuned LLMs with at least 13B total parameters exceeded both BERT baselines (relaxed F1, 0.882-0.914; exact-surface F1, 0.840-0.880). Across three-seed means, increasing the output limit improved relaxed F1 by 0.052-0.085, alongside higher recall and fewer length-limited outputs. JSON Schema constraints changed relaxed F1 by -0.014 to +0.008 among seven models outside the two reasoning-oriented configurations, while medical-general model pairs showed no consistent advantage for medical specialization. Conclusions: On this historical Japanese case-report benchmark, task-specific fine-tuning combined with adequate generation capacity was associated with stronger LLM extraction performance. Output-token limits materially affected comparisons with BERT, whereas enforcing output structure alone did not overcome low zero-shot accuracy.

## Watermark Evidence under Open-Weight LLM Modification: Exact Bounds for Low-Rank Output Watermarks

DevFeed: [Watermark Evidence under Open-Weight LLM Modification: Exact Bounds for Low-Rank Output Watermarks](<https://devfeed.tech/articles/watermark-evidence-under-open-weight-llm-modification-exact-bounds-for-low-rank-output-watermarks-139900.md>)

Original publisher: [Read original article](<https://arxiv.org/abs/2611.21702>)

Author: Prashant Shekhar, Dumindu Samaraweera, Caroline Howard

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

Content type: article

Language: en

Sources: [arXiv: Artificial Intelligence](<https://devfeed.tech/sources/arxiv-artificial-intelligence.md>), [arXiv: Cryptography and Security](<https://devfeed.tech/sources/arxiv-cryptography-and-security.md>)

Topics: [Large Language Model](<https://devfeed.tech/topics/llm.md>), [lora](<https://devfeed.tech/topics/lora.md>), [quantization](<https://devfeed.tech/topics/quantization.md>), [LLM security](<https://devfeed.tech/topics/llm-security.md>), [Open Source Models & Datasets](<https://devfeed.tech/topics/open-source-models-datasets.md>)

Tags: [attack](<https://devfeed.tech/tags/attack.md>), [covariance](<https://devfeed.tech/tags/covariance.md>), [cs-ai](<https://devfeed.tech/tags/cs-ai.md>), [cs-cl](<https://devfeed.tech/tags/cs-cl.md>), [cs-cr](<https://devfeed.tech/tags/cs-cr.md>), [experiments](<https://devfeed.tech/tags/experiments.md>), [false-positives](<https://devfeed.tech/tags/false-positives.md>), [fine-tuning](<https://devfeed.tech/tags/fine-tuning.md>), [int8](<https://devfeed.tech/tags/int8.md>), [language-models](<https://devfeed.tech/tags/language-models.md>), [llama](<https://devfeed.tech/tags/llama.md>), [lower-bound](<https://devfeed.tech/tags/lower-bound.md>), [mistral](<https://devfeed.tech/tags/mistral.md>), [robustness](<https://devfeed.tech/tags/robustness.md>), [sequence](<https://devfeed.tech/tags/sequence.md>), [span](<https://devfeed.tech/tags/span.md>), [theoretical](<https://devfeed.tech/tags/theoretical.md>), [watermark](<https://devfeed.tech/tags/watermark.md>)

### AI overview

The study derives bounds on how changes to open-weight LLM logits affect OpenStamp watermark evidence, including a residual-span lower bound, a covariance-based estimate of evidence drift, and sequence-level bounds on miss probability after attack selection. In tests on Llama-2-7B and Mistral-7B-v0.3, INT8 quantization and 30% pruning retained true-positive rates of 85.6-94.8%, while 500-step LoRA fine-tuning reduced rates to 28.8% and 11.6%. The covariance estimate tracked measured LoRA evidence drift, with no remainder-bound violations across 160,000 test prefixes. The experiments also recorded zero false positives among 500 base-model continuations per model.

### Source excerpt

arXiv:2611.21702v1 Announce Type: cross Abstract: Open-weight language models can continue generating after their released weights are modified, while a watermark verifier still scores outputs against the original base and marked models. We study how such modifications change the evidence available to an OpenStamp-style verifier. At each shared prefix, we represent the modification as a residual in the next-token logits. We derive an exact finite-vocabulary lower bound on the expected log-likelihood-ratio evidence under a residual-span constraint, along with a sharp span-only deviation bound. We also characterize the first-order evidence shift through a covariance term with an explicit second-order remainder, and determine the residual span needed to reach a target evidence level. For sequence-level detection, we derive an exact conditional exponential-moment bound and combine it with a supermartingale argument and a distribution-free order-statistic tolerance bound. This yields a high-confidence upper bound on miss probability after attack selection on calibration data and evaluation on independent certification continuations. In experiments with OpenStamp on Llama-2-7B and Mistral-7B-v0.3, INT8 quantization and 30% pruning retain test true-positive rates of 85.6-94.8%, while 500-step LoRA fine-tuning lowers them to 28.8% and 11.6%. The covariance prediction tracks the measured mean LoRA evidence drift on both models, with no remainder-bound violations across 160,000 test prefixes. Additionally, we observe zero false positives among 500 base-model continuations per model. Through these theoretical and empirical results, the current research provides a better understanding of watermark robustness to model-editing/tuning attacks.

## RPMem: Learning Long-Term Recurrent Parametric Memory Across Sessions for LLM Agents

DevFeed: [RPMem: Learning Long-Term Recurrent Parametric Memory Across Sessions for LLM Agents](<https://devfeed.tech/articles/rpmem-learning-long-term-recurrent-parametric-memory-across-sessions-for-llm-agents-139817.md>)

Original publisher: [Read original article](<https://arxiv.org/abs/2609.23466>)

Author: Fanyu Zhao, Ruike Cao, Liang Dong, Fugen Yao, Jian Xu, Guanjun Jiang, Yifei Zhao, Yinsheng Li, Han Zhang

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

Content type: article

Language: en

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

Topics: [AI chatbot continuity](<https://devfeed.tech/topics/ai-chatbot-continuity.md>), [Large Language Model](<https://devfeed.tech/topics/llm.md>), [long-running](<https://devfeed.tech/topics/long-running.md>), [lora](<https://devfeed.tech/topics/lora.md>), [harvestbench](<https://devfeed.tech/topics/harvestbench.md>)

Tags: [agents](<https://devfeed.tech/tags/agents.md>), [cs-ai](<https://devfeed.tech/tags/cs-ai.md>), [cs-cl](<https://devfeed.tech/tags/cs-cl.md>), [learning](<https://devfeed.tech/tags/learning.md>), [llm](<https://devfeed.tech/tags/llm.md>), [llm-agents](<https://devfeed.tech/tags/llm-agents.md>), [memory](<https://devfeed.tech/tags/memory.md>), [recurrent](<https://devfeed.tech/tags/recurrent.md>), [sessions](<https://devfeed.tech/tags/sessions.md>)

### AI overview

RPMem addresses cross-session memory evolution for long-running LLM agents by compiling each session into a model-independent latent memory, then using a task-trained recurrent gate to integrate it with retained memory. It maps the consolidated memory to backbone-specific LoRA parameters, which the authors say allows memory encoding to transfer when the backbone changes. Across three benchmarks and five backbones, the abstract reports near-constant update cost and memory footprint. On PERMA with Qwen3-8B, RPMem scores 85.52%, exceeding the strongest parametric baseline by 5.32 percentage points and the strongest text-based baseline by 12.98 points. Ablations support the complementary roles of session compilation and consolidation; the gate's observed integration strategies vary by task.

### Source excerpt

arXiv:2609.23466v3 Announce Type: replace-cross Abstract: Long-running LLM agents require memory that persists and evolves across sessions. Text-based memory retrieves and reconstructs past interactions at every query, making long-horizon performance increasingly dependent on retrieval quality and contextual reasoning as histories grow. Parametric memory encodes experience directly into model computation, but existing approaches provide limited support for cross-session memory evolution. Their coupling to a specific backbone further restricts memory reuse after model replacement. We introduce RPMem, a two-stage architecture that compiles each session into a model-independent latent memory through forward computation and selectively integrates it with retained memory via a task-trained recurrent gate. The consolidated memory is then mapped to backbone-specific low-rank adaptation (LoRA) parameters, allowing the encoding capability to transfer when the backbone is replaced. Evaluation across three long-term memory benchmarks and five diverse backbones demonstrates broad generalization with near-constant update cost and memory footprint. With Qwen3-8B on PERMA, RPMem reaches 85.52%, outperforming the strongest parametric and text-based baselines by 5.32 and 12.98 percentage points, respectively. Ablations validate the complementary roles of session compilation and cross-session consolidation, while dynamics analyses reveal that the gate acquires task-specific memory integration strategies. These results establish RPMem as a lifecycle-independent parametric memory framework that maintains evolving cross-session memory that remains reusable across backbone replacements. Our code is available at https://github.com/Quark-Medical/rpmem/tree/main.

## Learning Stratified Policies for Inference-Time Objectives

DevFeed: [Learning Stratified Policies for Inference-Time Objectives](<https://devfeed.tech/articles/learning-stratified-policies-for-inference-time-objectives-140124.md>)

Original publisher: [Read original article](<https://arxiv.org/abs/2611.23154>)

Author: Mridul Mahajan, Xuezhou Zhang, Aldo Pacchiano

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>)

Topics: [llm-reasoning](<https://devfeed.tech/topics/llm-reasoning.md>), [Large Language Model](<https://devfeed.tech/topics/llm.md>), [Machine learning](<https://devfeed.tech/topics/machine-learning.md>), [lora](<https://devfeed.tech/topics/lora.md>)

Tags: [adapters](<https://devfeed.tech/tags/adapters.md>), [approximation](<https://devfeed.tech/tags/approximation.md>), [best-of](<https://devfeed.tech/tags/best-of.md>), [budget](<https://devfeed.tech/tags/budget.md>), [cs-ai](<https://devfeed.tech/tags/cs-ai.md>), [cs-lg](<https://devfeed.tech/tags/cs-lg.md>), [inference](<https://devfeed.tech/tags/inference.md>), [language](<https://devfeed.tech/tags/language.md>), [language-models](<https://devfeed.tech/tags/language-models.md>), [large-language-models](<https://devfeed.tech/tags/large-language-models.md>), [learning](<https://devfeed.tech/tags/learning.md>), [linear](<https://devfeed.tech/tags/linear.md>), [lora](<https://devfeed.tech/tags/lora.md>), [model](<https://devfeed.tech/tags/model.md>), [multiple](<https://devfeed.tech/tags/multiple.md>), [objectives](<https://devfeed.tech/tags/objectives.md>), [optimize](<https://devfeed.tech/tags/optimize.md>), [pass](<https://devfeed.tech/tags/pass.md>), [performance](<https://devfeed.tech/tags/performance.md>), [policies](<https://devfeed.tech/tags/policies.md>), [policy](<https://devfeed.tech/tags/policy.md>), [time](<https://devfeed.tech/tags/time.md>)

### AI overview

The paper reports that allocating a k-response inference budget across k learned policies can improve pass@k over repeatedly sampling one policy. It implements the policy population as separate LoRA adapters on a shared language-model backbone, with each adapter contributing one response, and optimizes for pass@k. The authors show consistent empirical gains over a single-policy baseline across multiple reasoning tasks. In a linear setting, their analysis also finds that a population of LoRAs can attain zero best-of-k approximation error even when every individual LoRA has positive prediction error. The supplied text does not specify the tasks, gain sizes, or experimental setup.

### Source excerpt

arXiv:2611.23154v1 Announce Type: cross Abstract: Inference-time scaling improves the performance of large language models by sampling multiple responses and evaluating their collective utility. However, many existing approaches obtain these samples by repeatedly sampling from a single policy. In this work, we study if the sample budget at inference time can instead be allocated across a population of policies. We introduce stratified policies, where each of $k$ learned policies contributes one response to a $k$-sample inference-time objective. We realize this policy population using separate LoRA adapters on a shared language model backbone and optimize it for pass@$k$. We theoretically motivate the approach by analyzing the linear setting; a population of LoRAs can achieve zero best-of-$k$ approximation error in settings where every single LoRA has strictly positive prediction error. Empirically, we demonstrate that stratified policies consistently improve pass@$k$ over a single-policy baseline across multiple reasoning tasks. Overall, our results highlight how distributing samples across complementary policies can be a more effective use of sample budget than repeated i.i.d. sampling from a single policy at inference time.

## MoE--LoRA Methods for Multilingual Adaptation: A Systematic Empirical Study

DevFeed: [MoE--LoRA Methods for Multilingual Adaptation: A Systematic Empirical Study](<https://devfeed.tech/articles/moe-lora-methods-for-multilingual-adaptation-a-systematic-empirical-study-141855.md>)

Original publisher: [Read original article](<https://arxiv.org/abs/2611.23158>)

Author: Nikit Srivastava, Joel Dag, Ren\\'e Speck, Axel-Cyrille Ngonga Ngomo

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

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>), [LLM evaluation / benchmarking](<https://devfeed.tech/topics/llm-evaluation-benchmarking.md>), [Large Language Model](<https://devfeed.tech/topics/llm.md>)

Tags: [adapters](<https://devfeed.tech/tags/adapters.md>), [cs-cl](<https://devfeed.tech/tags/cs-cl.md>), [cs-lg](<https://devfeed.tech/tags/cs-lg.md>), [evaluation](<https://devfeed.tech/tags/evaluation.md>), [language](<https://devfeed.tech/tags/language.md>), [language-model](<https://devfeed.tech/tags/language-model.md>), [language-models](<https://devfeed.tech/tags/language-models.md>), [llms](<https://devfeed.tech/tags/llms.md>), [lora](<https://devfeed.tech/tags/lora.md>), [low-rank](<https://devfeed.tech/tags/low-rank.md>), [low-resource-languages](<https://devfeed.tech/tags/low-resource-languages.md>), [performance](<https://devfeed.tech/tags/performance.md>), [reading](<https://devfeed.tech/tags/reading.md>), [study](<https://devfeed.tech/tags/study.md>), [translation](<https://devfeed.tech/tags/translation.md>)

### AI overview

The study compares eight LoRA methods on a frozen language model across 200 languages. A reduced-scale search found viable configurations but did not reliably predict relative performance at scale; the top-ranked configuration was never best at scale. LoRA improved fit on translation data, consistently worsened reading comprehension, and benefited low-resource languages most. The authors therefore characterize LoRA as a targeted tool rather than a general-purpose way to improve multilingual capabilities.

### Source excerpt

arXiv:2611.23158v1 Announce Type: cross Abstract: Routed low-rank adapters offer a parameter-efficient means to improve the performance of large language models (LLMs) across tasks. Hence, it is assumed that they can also be used to improve the support of low-resource languages in LLMs, which currently serve English far better than the thousands of other languages spoken worldwide. However, this assumption has never been verified empirically under a shared multilingual evaluation protocol. We address exactly this gap by comparing eight low-rank adaptation (LoRA) methods on a frozen base model using 200 unique languages in more than 9{,}000 evaluation settings. Our results suggest that a reduced-scale search can identify viable configurations of LoRA methods but does not reliably predict their relative performance, as the top-ranking configuration is never best at scale. LoRA reliably improves language-model fit on translation data, consistently degrades reading comprehension, and benefits low-resource languages most. These findings suggest that LoRA methods should be regarded as targeted tools rather than general-purpose means for improving the multilingual capabilities of LLMs.

## Efficient Self-Supervised Adaptation of 3D Abdominal Vision-Language Model for Institution-Specific HCC Classification via Full Fine-Tuning and PEFT

DevFeed: [Efficient Self-Supervised Adaptation of 3D Abdominal Vision-Language Model for Institution-Specific HCC Classification via Full Fine-Tuning and PEFT](<https://devfeed.tech/articles/efficient-self-supervised-adaptation-of-3d-abdominal-vision-language-model-for-institution-specific-hcc-classification-via-full-fine-tuning-and-peft-141782.md>)

Original publisher: [Read original article](<https://arxiv.org/abs/2611.22504>)

Author: Febryan Putra Kartika, Cheng-Yu Ma, Ying-Jia Lin, Chi-Tung Cheng, Kuan-Fu Chen

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

Content type: article

Language: en

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

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

Tags: [3d](<https://devfeed.tech/tags/3d.md>), [acquisition](<https://devfeed.tech/tags/acquisition.md>), [annotations](<https://devfeed.tech/tags/annotations.md>), [classification](<https://devfeed.tech/tags/classification.md>), [contrastive-learning](<https://devfeed.tech/tags/contrastive-learning.md>), [cross](<https://devfeed.tech/tags/cross.md>), [cs-cv](<https://devfeed.tech/tags/cs-cv.md>), [cs-lg](<https://devfeed.tech/tags/cs-lg.md>), [fine-tuning](<https://devfeed.tech/tags/fine-tuning.md>), [lora](<https://devfeed.tech/tags/lora.md>), [medical](<https://devfeed.tech/tags/medical.md>), [peft](<https://devfeed.tech/tags/peft.md>), [pretraining](<https://devfeed.tech/tags/pretraining.md>), [temperature](<https://devfeed.tech/tags/temperature.md>), [training](<https://devfeed.tech/tags/training.md>), [vision-language-model](<https://devfeed.tech/tags/vision-language-model.md>), [vision-language-models-vlms](<https://devfeed.tech/tags/vision-language-models-vlms.md>)

### AI overview

The study reports that a pretrained 3D medical vision-language model can be adapted to institution-specific HCC classification using self-supervised contrastive learning on paired CT scans and radiology reports, without lesion annotations or HCC labels as training targets. Under cross-institutional domain shift, full fine-tuning achieved AUC 0.82 and F1 0.71; PEFT achieved AUC 0.79 and F1 0.70. Full fine-tuning converged quickly but overfit within a few epochs. PEFT used ConvLoRA for the image encoder and LoRA for the text encoder, updated about 1% of model parameters, and required more training steps. Both improved on the pretrained model, though results remained below those reported for supervised HCC classification. The abstract also reports that contrastive temperature initialization significantly affected performance.

### Source excerpt

arXiv:2611.22504v1 Announce Type: cross Abstract: Medical vision-language models (VLMs) capture cross-modal relationships between image and text, yet their adaptation to institution-specific clinical tasks remains underexplored. In this study, we fine-tuned a pretrained 3D medical VLM for hepatocellular carcinoma (HCC) classification using paired abdominal CT scans and radiology reports from a different institution with acquisition characteristics that differ from the model's original pretraining corpus. We compared two adaptation strategies: full fine-tuning and parameter-efficient fine-tuning (PEFT), motivated by the common use of PEFT to reduce computational cost and enable adaptation under limited-data constraints. Both approaches achieve strong downstream HCC classification performance despite the cross-institutional domain shift, with PEFT reaching an AUC of 0.79 and F1 of 0.70, and full fine-tuning achieving an AUC of 0.82 and F1 of 0.71. Although these results remain below the performance reported by supervised HCC classification approaches, our method achieves improvements over the pretrained model through CT-report contrastive learning without requiring lesion-level annotations or HCC classification labels as training targets. Full fine-tuning converges rapidly but overfits within a few epochs, whereas PEFT (ConvLoRA for the image encoder and LoRA for the text encoder) attains comparable performance while updating approximately 1% of the model parameters, although requiring more training steps. We also examine the role of contrastive temperature, observing that temperature initialization significantly affects classification performance. This study demonstrates that a 3D medical VLM can be efficiently adapted to institution-specific HCC classification using self-supervised CT-report contrastive learning, while highlighting the practical trade-offs between full fine-tuning and parameter-efficient fine-tuning.

## 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.

## Localized LoRA-MoE: Mixture of Low-Rank Experts With Adaptive Routing

DevFeed: [Localized LoRA-MoE: Mixture of Low-Rank Experts With Adaptive Routing](<https://devfeed.tech/articles/localized-lora-moe-mixture-of-low-rank-experts-with-adaptive-routing-139787.md>)

Original publisher: [Read original article](<https://arxiv.org/abs/2607.05114>)

Author: Babak Barazandeh, Subhabrata Majumdar, Vinay Prithyani, George Michailidis

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>)

Topics: [lora](<https://devfeed.tech/topics/lora.md>), [Mixture of Experts (MoE)](<https://devfeed.tech/topics/mixture-of-experts-moe.md>), [Fine-tuning](<https://devfeed.tech/topics/fine-tuning.md>)

Tags: [adapter](<https://devfeed.tech/tags/adapter.md>), [benchmarks](<https://devfeed.tech/tags/benchmarks.md>), [cs-ai](<https://devfeed.tech/tags/cs-ai.md>), [cs-cl](<https://devfeed.tech/tags/cs-cl.md>), [cs-lg](<https://devfeed.tech/tags/cs-lg.md>), [fine-tuning](<https://devfeed.tech/tags/fine-tuning.md>), [language-model](<https://devfeed.tech/tags/language-model.md>), [lora](<https://devfeed.tech/tags/lora.md>), [moe](<https://devfeed.tech/tags/moe.md>), [peft](<https://devfeed.tech/tags/peft.md>), [routing](<https://devfeed.tech/tags/routing.md>), [spatial](<https://devfeed.tech/tags/spatial.md>)

### AI overview

Localized LoRA-MoE combines spatially localized low-rank adapters with context-conditioned routing to reduce interference when training streams mix operational contexts. Its Block-Wise variant uses one gate for the full grid; Cell-Wise assigns a separate gate to each grid cell. In the reported structured benchmarks, both beat static baselines at comparable adapter budgets, while Cell-Wise performed better when contexts affected different spatial regions. On six interleaved GLUE tasks, task-conditioned routing beat LoRA and static localized adapters at a fixed budget and matched per-task adapters using six times the budget. On Qwen2.5-1.5B across eight commonsense benchmarks, both variants beat static baselines; Cell-Wise scored 3.9 points above LoRA at matched active capacity.

### Source excerpt

arXiv:2607.05114v2 Announce Type: replace-cross Abstract: Parameter-efficient fine-tuning (PEFT) methods such as LoRA adapt large models through a single low-rank update. Such updates are susceptible to gradient interference: when a training stream interleaves multiple operational contexts, conflicting updates to shared adapter weights collapse the adapter into an unspecialized average. While recent methods introduce block-wise isolation to remedy this issue, their topologies are static and cannot adapt to context switching or localized sensor degradation. In this work, we introduce Localized LoRA-MoE, a framework that combines localized spatial blocking with context-conditioned expert routing. We propose and evaluate two architectural variants: Block-Wise LoRA-MoE (centralized routing), which modulates the entire structural grid through a single context-conditioned gate, and Cell-Wise LoRA-MoE (decentralized routing), which assigns an independent gate to every coordinate cell of the matrix grid. Across three structured transformation benchmarks--a high-dimensional SVD matrix simulation, context-switched transformations over real tabular data, and spatial reconstruction under simulated sensor occlusion--both variants outperform static baselines under comparable adapter budgets. Both variants perform similarly when context shifts the grid uniformly, while cell-wise routing is strictly better when contexts are spatially heterogeneous. When fine-tuning a language model on six interleaved GLUE tasks, task-conditioned routing at a fixed adapter budget outperforms LoRA and static localized adapters, matching per-task adapters with 6x budgets. At decoder scale on Qwen2.5-1.5B, both routed variants outperform static baselines across eight commonsense benchmarks, with cell-wise routing achieving +3.9 points over LoRA at matched active capacity.

[Next page](<https://devfeed.tech/topics/lora.md?cursor=WyIyMDI2LTExLTE4VDA1OjAwOjAwKzAwOjAwIiwgIjU2NDdhZjMxLTc2ZGQtNDRlZS1iNWM5LWZhODg4YWU0MmU5MiJd>)