LoRA, PEFT, instruction tuning and domain adaptation — adapting a pretrained model without paying to train one.
Whether a language model has improved itself is increasingly judged not by mean accuracy but by which individual problems it gains and loses. Tracking these transitions means differencing two noisy estimates, leaving them vulnerable to measurement artifacts.
LLM-based Multi-Agent Systems (MAS) achieve strong performance on complex reasoning tasks by coordinating multiple agents, but at the cost of substantial token consumption. Recent work on automatic topology design, ARG-Designer, has reframed this problem as autoregressive graph generation.
The paper introduces V-REX, a specialist vision‑language model (VLM) tailored for veterinary X‑ray interpretation. By redesigning the VLM pipeline—including text tokenisation, pre‑training, grounding, and inference—the authors demonstrate that a carefully engineered model can outperform larger generalist foundation models while using far fewer parameters, data, and compute. V-REX is the first VLM capable of generating diagnostic reports for veterinary radiographs, achieving significant gains over open foundation models on this task.
The article "How to Fine-Tune an LLM: An End-to-End Guide" offers a practical, hands‑on walkthrough for fine‑tuning large language models in real‑world scenarios. It covers the entire process from data preparation to deployment, providing readers with actionable steps to adapt LLMs to specific tasks. The guide is aimed at practitioners looking to implement fine‑tuning in a structured, end‑to‑end manner.
A lot of prior work addressed key-value (KV) cache selection and compression by sparse attention to enable long-context inference for transformer language models without excessive hardware budgets. We provide a new method for fine-tuning models with sparse attention.
The paper presents a method to align a language‑model‑based crowd agent with aggregate mobility data by fine‑tuning it to match observed destination compositions derived from origin‑to‑destination flows. The approach uses iterative proportional fitting to reweight the model’s destination distribution and corrects for dominant destination inflation by training a low‑rank adapter on resampled trajectories. Experiments on mobile network counts from two baseball games show a 25% reduction in destination‑share error while maintaining similar grid correlation across policies.
Latent reasoning has advanced multimodal reasoning through a two-stage training paradigm: (1) a helper image is encoded into latent tokens to teach visual chain-of-thought during a supervised fine-tuning (SFT) stage, and (2) these latent tokens are further refined with reward feedback during a reinforcement learning (RL) stage. In this paper, we identify two key limitations of this framework, one in each stage.
Text editing in product posters entails inserting new text or replacing existing text while preserving product appearance, background content, and global composition. Despite recent progress in instruction-based image editing, general-purpose models remain unreliable in this setting: they often omit or incorrectly render the target text, place it over salient products or pre-existing content, and produce structurally distorted or visually inconsistent glyphs.
The paper examines how multilingual medical adaptation affects the internal representations of Whisper ASR models. By comparing various fine‑tuning strategies—zero‑shot decoding, English‑only, German‑only, two‑stage EN→EN+DE, and direct EN+DE fine‑tuning—it shows that fine‑tuning significantly improves performance, with the best model varying by setting. Layer‑wise encoder analysis reveals that English medical fine‑tuning drives the main representation shift, while multilingual continuation largely preserves the adapted space, and that domain and language signals remain recoverable across layers.
GigaBrain-WBC-0.5 is a Behavior World Model that uses a causal Transformer to predict next actions, states, and a distribution over latent behavior commands for humanoid whole-body control. It incorporates an automatic terrain-annotation pipeline to recover 3D contact geometry from motion data, allowing the model to learn how terrain and objects influence dynamics. The system detects implausible commands online, retracts them onto learned behaviors, and achieves high success rates in terrain interaction, command robustness, and fall recovery, with promising hardware trials on different robots.
The paper evaluates whether audio‑language models can use multimodal clinical context to improve dysarthric speech recognition. Using a benchmark built on the Speech Accessibility Project dataset, the authors test diagnosis labels, clinician ratings, and detailed clinical descriptions as prompts for nine models. They find that these prompts yield negligible or negative effects on word error rate, though fine‑tuning with LoRA and mixed prompt formats reduces WER by 52% and benefits certain subgroups such as Down syndrome and mild‑severity speakers.
MedStruct‑S is a benchmark for semi‑structured information extraction from OCR‑derived clinical reports, covering key discovery, key‑conditioned QA, and end‑to‑end key‑value extraction. It contains 3,582 annotated real‑world report pages and evaluates models under unknown keys and OCR noise. Experiments show encoder‑only models excel at non‑null key‑conditioned QA, while fine‑tuned decoder‑only models achieve the strongest overall performance across model sizes.
Brain2Qwerty v2 is a model that decodes natural sentences from real‑time magnetoencephalography (MEG) recordings, achieving an average word error rate of 39% across 22,000 sentences typed by nine subjects. The model uses character, word, and sentence‑level representations and shows that decoding accuracy improves log‑linearly with more data, narrowing the gap to intracranial brain‑computer interfaces. AI contributes by replacing hand‑crafted event detection with deep learning, fine‑tuning large language models for semantic extraction, and employing AI agents to refine the decoding pipeline through automated code development.
The paper presents a local traditional OCR pipeline that can be iteratively fine‑tuned on both layout and appearance levels of complex historical Sanskrit manuscripts. By adapting to the specific manuscript distribution, the pipeline reduces human annotation effort and improves transcription accuracy across subsequent pages. The authors apply this method to three manuscripts, release a dataset with detailed layout and Unicode annotations in PAGE‑XML format, and benchmark the pipeline against leading multimodal large language models.
The paper investigates the use of looped language models for compositional tool calling, where models must coordinate multiple API calls and maintain state across interactions. Experiments on API-Bank, BFCL, and NESTful show that recurrent computation generally improves compositional and dependency-aware tool use, with accuracy increasing as recurrent depth grows. Adaptive inference offers a better compute‑performance trade‑off by allocating extra computation only when necessary.
The paper introduces Top‑K prompting as a training and inference strategy to better capture the diverse, plausible predictions inherent in single‑step retrosynthesis. Using an ultra‑large dataset (CREED‑CCV‑2+USPTO‑XL) of ~45.6 million verified reactions, the authors train the Chemistry Constraint‑Consistent Language Model (C3LM). With fine‑tuning that incorporates ChemCensor‑based and novelty‑oriented rewards, C3LM achieves state‑of‑the‑art performance on the OOD URSA‑expert‑2026 benchmark and demonstrates complementary reaction space exploration compared to conventional models, suggesting benefits for ensemble‑based retrosynthesis systems.
The report details the winning approach for the AES AIMLA 2025 Challenge on querying sound effects by vocal imitation. It explores two fine‑tuning strategies: contrastive learning with a frozen, pretrained CED encoder, and joint contrastive‑triplet learning using a MobileNetV3 encoder with semi‑hard negatives. Updated post‑challenge, the document includes additional released details.
The study investigates how few expert-annotated cases are needed to fine‑tune MedSAM3 for abdominal organ segmentation using Low‑Rank Adaptation (LoRA). With only 10 annotated CT or MRI cases, the LoRA‑adapted models achieve performance comparable to specialist systems that require orders of magnitude more data, including reliable gallbladder segmentation and near‑state‑of‑the‑art results for liver, kidneys, and spleen. The approach also generalizes to cardiac segmentation on the Whole Heart dataset, and training takes only 3–5 hours per organ on a single GPU, roughly twice as fast as nnU-Net.
GS‑VLA introduces a lightweight, plug‑and‑play framework that uses a 4 M‑parameter 3D‑Gaussian canonicalizer to adapt frozen Vision‑Language‑Action (VLA) policies to viewpoint shifts without retraining the policy. By treating viewpoint changes as a localized novel‑view synthesis problem under a locality assumption, the method normalizes observations through a scene‑ and policy‑independent disocclusion task. Experiments on the LIBERO benchmark demonstrate that GS‑VLA recovers a large portion of performance lost due to camera displacement, improving results across different policy architectures, unseen task suites, and perturbation scales. whyItMatters":"The approach offers a computationally efficient alternative to costly fine‑tuning or generative augmentation, enabling robust VLA deployment in real‑world settings where camera configurations may vary."
The paper presents a method for extracting key information from OCR‑digitized clinical reports, addressing challenges posed by heterogeneous documents and noisy OCR output. It introduces an open key space that is iteratively mined, normalized, clustered, and verified to build a canonical key inventory, and defines key coverage as a metric for inventory completeness. Experiments on reports from over 20 hospitals using a 0.2B BERT model show that performance improves steadily with key coverage, achieving high F1 scores when the top 90 keys are covered and outperforming a fine‑tuned Qwen3‑0.6B baseline.