Small Language Models

Text-to-SQL is a fundamental task in natural language processing that enables users to interact with structured databases using natural language. While large language models (LLMs) have demonstrated remarkable performance on this task, their substantial computational requirements hinder deployment in resource-constrained settings. In this paper, we introduce SQuaD-SQL (Small-Qualified and Distilled for SQL), a novel approach that empowers small language models (SLMs) to approach the performance of LLMs on the Text-to-SQL task while significantly improving efficiency through knowledge distillation and synthetic data generation. Our method comprises three key components: (1) LLM-based synthetic data generation, where structured knowledge is extracted from LLMs via carefully designed prompting strategies; (2) parameter-efficient fine-tuning, enabling full model training on a single consumer-grade GPU; and (3) domain-adaptive fine-tuning, where domain-specific synthetic data further enhances performance in targeted domains. Experiments on the WikiSQL dataset demonstrate that SQuaD-SQL achieves an execution accuracy of 86.9% on the test set, approaching the performance of LLMs while offering faster inference and lower memory usage. These results suggest that, with proper training strategies, SLMs can serve as practical and efficient alternatives for Text-to-SQL applications in resource-limited environments.

Large Language Models achieve strong code generation for high resource languages like Python and Java but suffer sharp performance drops on Low-Resource Programming Languages~(LRPLs) such as Julia. Improving Small Language Models~(SLMs) for these languages faces a trilemma: Supervised Fine-Tuning~(SFT) is bottlenecked by data scarcity, inference-time scaling is too expensive for deployment, and Reinforcement Learning from scratch yields near zero advantages. We propose a three-phase pipeline that resolves this trilemma by decoupling syntax acquisition from algorithmic reasoning. First, we \emph{left-shift} inference-time compute to an offline data synthesis engine that uses iterative compiler and test feedback to generate verified training examples. Second, we fine-tune an SLM on this synthetic, verified data to embed strong syntactic priors. Third, we apply Reinforcement Learning with Verifiable Reward~(RLVR) grounded by language-agnostic Input/Output tests, where the SFT prior constrains exploration away from syntax errors. Applied to Qwen3-8B, our pipeline improves pass@1 by up to +7.6 points on MultiPL-E and +14.2 points on the Agnostics LiveCodeBench for Julia compared to SOTA results. Furthermore, the pipeline only used $\frac{1}{3}$ data and $\frac{1}{6}$ cost over the previous state-of-the-art. We further demonstrate that the pipeline generalizes to Ballerina achieving 49.7% MultiPL-E Pass@1, a language with near-zero pretraining representation. Ablations confirm that both the SFT phase and execution-grounded rewards are necessary for stable training.

We present LLM4SDM, the first study of open-source smaller language models (OS-sLLMs) for automated assessment of shared decision making (SDM) using the Observer OPTION12 framework. Unlike previous work that relies on large commercial models and the shorter OPTION5 instrument, our study focuses on privacy-preserving locally deployable models and Dutch melanoma consultation transcripts. Using expert-annotated clinical consultations, we evaluate three general-domain and two medical-domain OS-sLLMs during a development-phase pilot study. Results show that general-domain models outperform medical-domain models, which exhibit substantial hallucination and instruction-following failures. Gemma3:12b achieves the strongest agreement with human annotations (Pearson r=0.51, Spearman \r{ho}=0.59). Item-level and qualitative analyses reveal systematic challenges related to temporal discourse reasoning, conversational role attribution, and evidence grounding. We further introduce a Judge-LLM consensus framework designed to support disagreement resolution among multiple models. Our findings suggest that while current OS-sLLMs cannot replace human annotators, they offer a promising foundation for privacy-preserving human-in-the-loop SDM assessment.

Off-the-shelf TTS systems are poorly adapted to Taiwanese Mandarin. Their accent defaults to other Mandarin variants, their tokenizers over-segment common Taiwanese text, and their pronunciation degrades at code-switching boundaries where Chinese and English alternate within one utterance. These problems share one root: the text side lacks adaptation to the Taiwanese context. We address the text side from the bottom up. PangolinTokenizer, a byte-level BPE tokenizer trained on Taiwan-context data, reaches the lowest token rate (0.485 tokens/character) with the smallest vocabulary among nine tokenizers. Barbet, a billion-parameter Traditional-Chinese language model trained on PangolinTokenizer, serves as the text-semantic frontend and ranks first among comparable public models on a 14-task evaluation. BlueMagpie-TTS attaches Barbet to the pretrained acoustic stack of VoxCPM2 through a learned bridge, keeping the acoustic stack fixed. On a 1000-sentence Taiwan-localized test set, it lowers CER from 11.45% to 4.81% and WER from 14.83% to 5.36%, relative reductions of 58.0% and 63.9%. In a blind listening study on 500 of these sentences with ten listeners, 65.6% of majority votes prefer BlueMagpie-TTS.

Dynamic Voltage Frequency Scaling (DVFS) on resource-constrained embedded GPU platforms is essential for energy-efficient small language model (SLM) fine-tuning, as privacy- and personalization-driven adaptation increasingly requires local execution and involves repeated forward-backward optimization over many mini-batches, making it substantially more time- and energy-intensive than single-pass inference. To this end, 1) we first characterize the fine-tuning behavior of representative encoder-only SLMs of BERT variants, and autoregressive decoder-only SLMs of Pythia variants on GLUE benchmarks. In addition to the characterizations, 2) we propose a simple yet effective ML-based model selection that selects energy-optimal GPU DVFS settings on resource-constrained embedded platforms. Our results on NVIDIA Jetson AGX Orin demonstrate average 13.11% energy savings (up to 26.73%) over MAXN Mode 0, which has no explicit power cap.

Large language models are increasingly explored as AI tutors, yet deploying them in K-12 settings raises concerns around privacy, cost, and reliance on proprietary models. Small language models (SLMs) offer a promising alternative, but selecting the right model for a specific educational context remains difficult, particularly when the target domain, such as block-based programming, is largely absent from model training data. We introduce CSTutorBench, a benchmark for evaluating language models as CS tutors in VEX VR, a block-based robotics environment. The benchmark comprises 17 scenario-based questions scored against a pedagogical rubric grounded in established tutoring and feedback research, with a human-in-the-loop LLM-as-judge pipeline for evaluation. Preliminary findings across 11 models (4B-120B parameters) reveal that models perform well on surface-level criteria such as vocabulary and tone but struggle with deeper pedagogical behaviors, particularly avoiding answer leakage and engaging with student debugging histories. In our sample, model family and instruction-tuning approach appear to be better predictors of tutoring quality than parameter count alone, though the small number of models limits the strength of this conclusion. A targeted prompt revision grounded in recent educational prompt engineering research improved scores for 10 of 11 models. These results underscore the value of context-specific, pedagogically grounded benchmarks for SLM selection in educational deployment.

Physics reasoning fails structurally in small language models: an error at any step propagates forward, corrupting every inference that follows. Limited domain knowledge, hallucination under multi-step derivation, and distributional sensitivity compound this failure. We propose a step-level reward framework that identifies the first reasoning error, generates targeted structured feedback, and trains the model to revise its solution via policy gradient with KL regularization, without exposing it to ground truth solutions as generation targets. Unlike annotation-dependent step-level methods, no preference data construction is required and the external verifier operates exclusively at training time. Across five physics benchmarks, our framework delivers accuracy gains of 17-20% over CoT prompting and 10-16% over the strongest baseline, reduces calculation errors from 56.9% to 23.5%, and reduces miscomprehension errors from 22.3% to 12.0% in the best observed cases. Conceptual errors reduce from 89.7% to 68.7%, yet persist as the hardest failure mode across all conditions.

We define Tiny Language Models (TLMs) as models below roughly 3B parameters that fit on mainstream consumer devices. We study how to adapt them for and use them on verifiable multiple-choice tasks. We compare three LoRA-based fine-tuning paradigms (label generation, gold only, and our discriminative classification head) on a unified setup across several Qwen3 models from 0.6B to 8B and five benchmarks: HellaSwag, WinoGrande, PIQA, SciQ and ARC-C. Classification-head fine-tuning reliably outperforms label generation (+2-3%) at the 0.6B and 1.7B scales. Further, TLMs fine-tuned using the discriminative method are competitive to zero-/few-shot GPT-3 (175B), PaLM (540B) and GPT-4. The performance we report for Qwen3-0.6B and Qwen3-1.7B are SOTA on HellaSwag, WinoGrande, and PIQA.

Reinforcement Learning with Verifiable Rewards (RLVR) has emerged as a promising paradigm for improving mathematical reasoning in language models. Yet most RLVR work rewards only the final answer (outcome-based rewards), leaving the impact of step-level process supervision (process rewards) underexplored especially for small models that lack the capacity to self-correct under sparse feedback. We systematically compare five reward conditions applied to Qwen2.5-0.5B fine-tuned with Group Relative Policy Optimization (GRPO) on GSM8K: a no-RL baseline, process-only, outcome-only, and three hybrid weightings ($λ\in \{0.9, 0.5, 0.1\}$ process weight). Process-only supervision achieves 63.73% test accuracy versus 53.75% for outcome-only, a nearly 10-percentage point gap while yielding reasoning traces with higher step validity and lower deviation from ground-truth chain length. Hybrid rewards generally correlate positively with process weight, with one notable anomaly: the low-process / high-outcome configuration ($λ=0.1$) underperforms pure outcome supervision, suggesting conflicting optimization signals. Error analysis using GPT-4o as a judge reveals distinct failure mode distributions: process models generate structurally inconsistent but arithmetically grounded traces, while outcome models produce concise but derivation-error-prone chains. Our results demonstrate that reward granularity is a first-order design decision for RLVR, with process-level supervision substantially improving both accuracy and trace fidelity in small language models.

We present Wiola, a fully original Small Language Model (SLM) architecture built from first principles, sharing no structural lineage with any existing model family including GPT, LLaMA, Mistral, or Falcon. Wiola introduces five independently novel components: (i) Spiral Rotary Positional Encoding (SRPE), which embeds token positions on a three-dimensional helical manifold combining absolute, relative, and hierarchical positional signals; (ii) Gated Cross-Layer Attention (GCLA), providing each decoder layer with soft cross-attention access to compressed summaries of two preceding layers for inter-layer coherence; (iii) Adaptive Token Merging (ATM), which dynamically merges se mantically redundant adjacent tokens in middle network layers to reduce attention complexity without information loss; (iv) Dual Stream Feed-Forward (DSFF), replacing the conventional MLP with two parallel streams fused by a learned per-dimension gate; and (v) WiolaRMSNorm, a modified normalisation introducing a per-dimension learned offset vector that prevents representation collapse. We provide complete mathematical derivations, architectural block diagrams, complexity analyses, and systematic comparisons against GPT-2, LLaMA-2, and Mistral. Wiola is released in four sizes (120M, 360M, 700M, and 1.5B parameters) and is fully compatible with the HuggingFace Transformers ecosystem, with all 22 architectural unit tests passing.

While large language models have been dominating the research landscape recently, small language models remain highly relevant across various domains; yet, they receive far less attention. In this study, we investigate how smaller language models perform during the generation stage within a Retrieval-Augmented Generation (RAG) system. To benchmark these models effectively, we utilised both open-source and proprietary datasets covering diverse subject areas and question types. Our findings demonstrate that a RAG system with small language models can be executed directly on-device without requiring any GPU hardware within a reasonable time. The experimental code and links to the supplementary materials can be accessed through the GitHub repository: https://github.com/SibNN/SLM-RAG-EVAL.

CSIRTs increasingly fine tune language models on vulnerability scan records, but these records expose internal network topology and create privacy risks under regulations such as GDPR and LGPD. We present the first empirical study of how DP SGD and HMAC pseudonymization interact when fine tuning small language models with 1B to 3B parameters on structured CSIRT data. We evaluate 96 LoRA adapters across four SLMs and four training regimes, including raw fine tuning, QLoRA with large batch training, and DP SGD with epsilon equal to 2 and 8. We also audit memorization using 20 planted canaries, four extraction attacks, and a dual attack targeting HMAC pseudonymized identifiers. Our results show three main findings. First, matched update controls reproduce the observed reduction in memorization by reducing the number of optimizer updates alone, accounting for 66 percent to 132 percent of the measured effect, with a mean of 100 percent across three seeds and four models. In this setting, DP SGD provides the formal privacy guarantee but does not produce additional measurable reductions in memorization. Second, HMAC pseudonymization removes the original identifiers from the exposure surface, reducing exposure by 40 percent to 61 percent, while pseudonymized identifiers remain close to the expected random baseline and do not become a secondary memorization target. Third, F1 scores remain between 0.19 and 0.28 across all 96 adapters using four shot prompting, indicating that, under the evaluated training budget, 1B to 3B SLMs do not achieve operationally useful performance.

Small Language Models (SLMs) are increasingly being considered for deployment on edge devices such as laptops, enabling private, low-latency, and locally personalized applications. However, personalization requires models to adapt over time to evolving user- or task-specific data, placing them in a continual learning setting. This creates the risk of catastrophic forgetting, where learning new information degrades performance on previously learned tasks or broader model capabilities. Recent benchmarks such as TRACE have shown that continual fine-tuning can significantly degrade the general abilities of aligned large language models. In this work, we present a study for sequential LoRA personalization of SLMs. We save model checkpoints after each adaptation stage and evaluate them on current tasks, previously seen tasks, and a fixed reference set. This checkpoint-level protocol enables us to monitor task performance, forgetting, and reference set drift over time. We show that lightweight reference set distributional diagnostics can reveal model-specific instability patterns during sequential LoRA personalization of SLMs, including cases where task-level metrics alone hide harmful adaptation. We hope this can highlight new research avenues for monitoring stability of SLMs in a continual learning setting.

Social-physical human-robot interaction (spHRI) has grown rapidly across robotics, human-computer interaction, human-robot interaction, and haptics. Yet, fragmented terminology and inconsistent methodologies make systematic synthesis difficult. To support scalable review practices, we evaluated the extent to which small language models (SLMs; < 1.5B parameters) can assist with title and abstract screening for a large spHRI systematic review. While no SLMs matched human reviewers' performance, the models operated locally and screened papers orders of magnitude faster. The combined SLM ensemble identified 39 papers reviewers missed, representing 10.29% of the final relevant dataset. These results demonstrate that SLMs can augment, rather than replace, expert reviewers and make large-scale literature reviews accessible and sustainable.

A key step toward autonomous industrial operation is the ability to create and reconfigure control policies from natural-language requirement specifications, with minimal or no manual redesign. In this setting, policy generation by AI agents can be a credible path when paired with a plant-aware validator (e.g., a digital twin) that can check generated candidate actions before execution. However, practical deployment is constrained by inference latency and compute footprint: large cloud-based models are often too slow, opaque, or data-sensitive for edge closed-loop use. This work investigates whether a compact Small Language Model (SLM) can be retrained for control reasoning and embedded in a validator-guided correction loop. We use a Qwen2.5-1.5B model aligned via Group Relative Policy Optimization (GRPO), combined with (i) an action agent, (ii) a symbolic/digital-twin-style validation layer, and (iii) a reprompting agent that iteratively steers outputs toward valid actions. In randomized thermal-control simulations (30 experiments with 500 steps each), the framework achieves 91.5% average action-alignment accuracy (86.3%--100% across cases) at 3.84,s mean inference latency. Under symbolic re-mapping, it maintains a 95% in-range rate, indicating robust physical regulation despite reduced token-level agreement. These results support SLM+validator architectures as a practical path toward reconfigurable autonomous control at the edge.

Small language models offer an efficient alternative to large-scale systems, but their ability to execute structured algorithms over multiple dependent decisions remains poorly understood. We study graph algorithm execution as a closed-loop prediction problem in which a model repeatedly selects the next action from the current graph and algorithmic state. Our evaluation framework covers several classical graph procedures, multiple synthetic graph families, and disjoint training, validation, and test partitions. It assesses both local decision quality and global execution behaviour using step accuracy, exact rollout accuracy, constraint validity, partial solution quality, prefix survival, and intervention-based diagnostics. The results show that adaptation can produce reliable policies for structural procedures such as traversal and coloring, while weighted algorithms remain substantially more sensitive to error accumulation. More broadly, the findings demonstrate that strong next-step prediction does not necessarily translate into reliable autonomous execution and motivate evaluating algorithmic language models through complete closed-loop rollouts rather than isolated decisions.

Linear mode connectivity (LMC) provides a promising foundation for understanding and merging independently trained neural networks, but existing methods typically optimize the interpolation path from only one model endpoint, limiting their scalability and effectiveness for large pretrained transformers. We propose a novel and scalable framework for enabling LMC-based model merging to {\em billion-parameter pretrained transformers}. Our method applies properly parameterized functionality-preserving weight transformations to align functionally equivalent solutions, and introduces a dual learning procedure in which both models jointly learn their corresponding transformations toward a shared linear interpolation path. This bidirectional optimization substantially reduces interpolation barriers and enables more reliable merging across large-scale architectures. Empirically, we show that our approach achieves near-zero loss barriers on WikiText for language models with medium-sized parameters, representing, to our knowledge, the first demonstration of near-barrier-free linear connectivity at this scale. In the vision domain, ViT-L maintains above 69% ImageNet top-1 accuracy throughout the interpolation path, while modern billion-parameter LLMs exhibit only small loss barriers. These results suggest that properly resolving parameter symmetries enables large pretrained Transformers to be connected and merged through simple linear paths with substantially improved interpolation performance. Code: https://github.com/VILA-Lab/Dual-Learned-Matching .

To make large-scale distributed training practical outside high-bandwidth datacenters, we must reduce blocking, high-volume synchronization. While DiLoCo communicates infrequently, its outer synchronization remains bandwidth-heavy and brittle to stragglers and transient failures. We relax exact synchronization to approximate synchronization via mixing/gossip, which degrades gracefully under delays and communication failures. This allows us to factorize DiLoCo synchronization into a non-blocking mixing step that overlaps computation with no staleness, and a blocking mixing step that tightens worker agreement, yielding a tunable trade-off between compute utilization and optimization stability. On up to billion-parameter language models in low-bandwidth settings, our framework substantially improves compute utilization compared to DiLoCo, with training progress ranging from comparable to closely matching it, and is more robust to failures.

Large language models (LLMs) achieve strong relation extraction (RE), but their computational demands and reliance on proprietary APIs limit deployment in resource-constrained or privacy-sensitive settings. We investigate how far small language models (SLMs) can close this gap across general-domain and literary text. We evaluate five models from 360M to 3B parameters under three domain-composition regimes and two prompt-conditioned tuning styles (30 configurations), comparing them with zero-shot frontier LLMs and a discriminative RoBERTa baseline. Across nine benchmarks, the best sub-billion model, Qwen2.5-0.5B fine-tuned on pooled general-domain data, achieves a general-domain positive-class micro-F1 of 0.83, versus 0.69 for GPT-5.4 and 0.66 for Claude Sonnet 4.6 evaluated zero-shot. This does not imply that SLMs are intrinsically stronger; rather, targeted task adaptation enables 4-bit models deployable on a single consumer GPU to outperform general-purpose frontier systems under this protocol. An in-domain RoBERTa baseline also exceeds both frontier models, indicating that the gain stems from task adaptation rather than generative decoding. On literary RE, tuned SLMs reach 0.92 on the human-annotated Biographical benchmark versus 0.83 for GPT-5.4, and 0.833 versus 0.578 on the two-benchmark literary average. A targeted domain-adaptive pretraining case study yields no practically meaningful gain over supervised fine-tuning, while the cleanest within-family scale comparison shows only marginal improvement. These results show that, when task-specific data are available, compact task-adapted models can provide accurate, private, and hardware-efficient RE.

What happens when LLM agents operate with no context outside a turn, minimal prompting, and simple tools? Inspired by swarm engineering, we give collectives of three agents the ability to send messages and manipulate a shared actively decaying text store, introducing evolutionary pressure. The agents spontaneously cooperate, develop storage management strategies, and generate complex evolving cultural artifacts, with no top-down engineering. Using tools from dynamical systems analysis, we show that these behaviours exhibit structured long-range coherence beyond the entropy horizon of the decaying store, consistent with emergent culture in the Sperberian sense.

Small language models are cheap to serve and feasible on local hardware, but strong public 135M-class systems are commonly trained with hundreds of billions to trillions of tokens on large clusters. We study a sharply resource-constrained regime: a complete 134.5M-parameter language-model pipeline executed on one NVIDIA L20 GPU. The released checkpoint, L20-Edu-135M, receives approximately 13B pretraining tokens: 10B FineWeb-Edu tokens followed by a 3B-token educational, mathematics, code, and reasoning mixture. We document the architecture, data gates, cross-source MinHash/LSH near-deduplication, segment deduplication, benchmark-overlap removal, throughput optimization, supervised fine-tuning (SFT) with weight interpolation, and reinforcement learning from verifiable rewards (RLVR) on GSM8K. In a self-run zero-shot six-task harness, L20-Edu-135M obtains a mean score of 0.4150. It trails SmolLM-135M (0.4767) and SmolLM2-135M (0.4917), but its mean is 87.1% of SmolLM-135M's while its nominal token count is 2.17% as large. This ratio is descriptive, not evidence of statistical equivalence or a controlled scaling law. The model exceeds several older 100M-160M public baselines under the same harness. Direct GRPO-style RLVR decreases GSM8K exact-match accuracy from 1.82% to 1.59% (192-token completions) and 1.21% (320-token completions). These single-run results identify a concrete failure mode rather than establishing a general lower bound on RLVR. The contribution is an auditable resource-constrained case study, not a state-of-the-art claim.

Clinical information for amyotrophic lateral sclerosis (ALS) care documented in unstructured clinical notes limits downstream analysis without extraction into structured formats. Open-source small language models with few-shot prompting for detecting the presence of ALS-relevant clinical terms in patient documentation were evaluated without task-specific training data. The detection task targeted 17 categories spanning functional scores, respiratory measures, medications, and related clinical and non-clinical attributes. Clinical note content was normalized from JSON-encoded discharge summaries and processed with a prompt template having structured JSON outputs. We compared 26 open-source models using aggregate, label-level, and manual-validation multilabel classification metrics. Manual validation showed that a regex rule baseline had higher overall micro-F1 and lower Hamming loss than any single SLM or TF-IDF baseline, while Qwen3-4B-Instruct-2507 was the highest-performing SLM by micro-F1. Model rankings varied by metric and label category, with the TF-IDF baseline showing high recall but low precision, some SLMs showing higher precision but lower recall, and Hammer2.1-7b showing strong performance for ALSFRS-R subscore detection. These findings support targeted hybrid extraction workflows rather than replacement of existing rule-based methods.

This paper evaluates the performance of twelve Small Language Models (SLMs) on Arabic natural language processing tasks. The study introduces a benchmark of 240 Arabic test items distributed across eight domains and ten language skills, covering both comprehension-oriented and generation-oriented tasks. All models were evaluated under a controlled zero-shot setting using a standardized Arabic-only prompt template. Model responses were assessed through a multi-model LLM-as-a-judge framework involving GPT-4.1 Mini, Claude Haiku 4.5, and DeepSeek-Chat, with scores aggregated across judges and analyzed by task, skill, and model family. The results show that Gemma 3 (12B) achieved the highest overall score (4.548/5), followed by Aya and C4AI Command Arabic. The observed results suggest that model size alone does not explain Arabic SLM performance. Models with stronger Arabic alignment and more reliable instruction-following behavior tended to perform better across tasks. Common failure patterns among lower-performing models include prompt leakage, hallucination, language drift, incomplete generation, and weak task adherence. Overall, the benchmark provides a structured reference for evaluating compact Arabic language models and supports future work on efficient, reliable, and culturally appropriate Arabic AI systems.

Large language models (LLMs) exhibit substantial cross-lingual variation in mathematical reasoning performance, but it remains unclear whether these differences reflect language-specific parameters or a shared mechanism that manifests differently by language. We present a cross-lingual mechanistic analysis of mathematical reasoning in LLMs, enabling us to localize and compare model parameters that support mathematical reasoning across languages. We find that the extracted math-associated parameters exhibit partial cross-lingual overlap, with the strongest overlap concentrated in intermediate model layers. We further observe that English consistently produces the largest set of math-relevant parameters, whereas lower-resource languages reveal smaller sets of relevant parameters. These results suggest that math-related behavior in multilingual LLMs is neither fully language-invariant nor fully language-specific, but instead exhibits partial cross-lingual parameter overlap with systematic language-dependent differences.

While the NLP landscape is dominated by multi-billion parameter architectures, their deployment in low-resource, non-Latin scripts remains computationally prohibitive for edge configurations, mobile systems, and decentralized local hardware. This paper presents bangla-smollm-135m, a highly compact 135-million parameter decoder-only foundational model engineered explicitly for high-efficiency language modeling in the Bangla script. By leveraging a deterministic intersect-and-append token merging strategy between TituLLMs and SmolLM2-135M, the model overcomes subword script fragmentation without destabilizing early pretrained parameter states. In zero-shot multi-task benchmark evaluations (PIQA_bn, OpenBookQA_bn, CommonsenseQA_bn, and Bangla_MMLU), bangla-smollm-135m matches or outperforms models twice its size (Gemma-3-270m) and achieves parity with models in the 1B parameter tier. The model is available at rnnandi/bangla-smollm-135m

This technical report introduces VibeThinker-3B, a compact dense model with 3B parameters developed to investigate how far verifiable reasoning can be pushed within a strictly small-model regime. Building upon the Spectrum-to-Signal post-training paradigm, we systematically enhance the model through an optimized pipeline that includes curriculum-based supervised fine-tuning, multi-domain reinforcement learning, and offline self-distillation. Experimental evaluations demonstrate that VibeThinker-3B achieves frontier-level performance on highly demanding verifiable tasks. Specifically, it attains a score of 94.3 on AIME26 (improving to 97.1 with claim-level test-time scaling), an 80.2 Pass@1 on LiveCodeBench v6, and exhibits strong out-of-distribution generalization with a 96.1% acceptance rate on recent unseen LeetCode contests. This effectively places it in the performance band of first-tier reasoning systems, matching or exceeding flagship models that are orders of magnitude larger, such as DeepSeek V3.2, GLM-5, and Gemini 3 Pro. Furthermore, a score of 93.4 on IFEval confirms that this extreme reasoning enhancement does not compromise strict instruction controllability. Extending our previous 1.5B work, these findings motivate the Parametric Compression-Coverage Hypothesis, which views verifiable reasoning as compressible into compact reasoning cores, while open-domain knowledge and general-purpose competence require broad parameter coverage over facts, concepts, and long-tail scenarios. This perspective suggests that compact models are not merely deployment-efficient substitutes, but a complementary path toward frontier-level performance in parameter-dense capability regimes.

Pruning promises a shortcut to strong small language models. In this work, we examine this promise by pruning Llama-3.1-8B at pruning ratios of 0.5--0.8 with six methods spanning depth, width, and sparse granularities, under two controlled token-matched settings. (1) With the same training token budget, pruned initialization consistently outperforms random initialization. This shows that the parent model provides a strong starting point, although the advantage narrows as the training token budget grows and as the pruning ratio rises, nearly vanishing at the highest pruning ratio we study. (2) When training from scratch is instead given the full token budget consumed by the whole pipeline, pruning at finer granularities still retains an advantage, while coarser structured pruning can be matched or surpassed. This suggests that the parent model transfers knowledge that additional training tokens alone cannot fully recover, but only at fine granularity. Taken together, our results yield a clear recommendation: with a large pretrained model in hand and a limited training token budget, pruning is better than training from scratch; when the training budget is not limited, training from scratch can be competitive for coarser pruning, so a large pretrained parent is not always necessary.

We introduce SkMTEB, the first comprehensive MTEB-style text embedding benchmark for Slovak, a low-resource West Slavic language, comprising 31 datasets across 7 task types -- nearly 4$\times$ the depth of existing multilingual benchmark coverage for Slovak. Our evaluation of 31 embedding models reveals that large instruction-tuned multilingual models achieve the strongest performance, while existing Slovak-specific models trained for NLU tasks transfer poorly to embedding tasks. To address the need for efficient, locally-deployable Slovak embeddings, we develop \texttt{e5-sk-small} (45M parameters) and \texttt{e5-sk-large} (365M) by applying vocabulary trimming and fine-tuning to Multilingual E5 models. Despite size reductions of up to 62%, our open-source models achieve competitive performance with proprietary APIs while remaining locally deployable for semantic search and retrieval-augmented generation (RAG). We release the benchmark, models, datasets, and code openly, hoping our approach offers a replicable path for other under-resourced languages.

This study examines training dynamics in a small Llama-style language model trained under a fixed, compute-constrained token budget. Rather than evaluating efficiency solely through endpoint performance, the study uses a quantitative experimental repeated measures design to analyze how validation loss, validation perplexity, rolling volatility, backslide behavior, spike behavior, and between-seed variability change across token-based training intervals. Six independent training runs were conducted on a 4.26-million-parameter model using the TinyStories corpus, CPU-based full-precision training, and a target budget of approximately 20 million cumulative training tokens. Metrics were collected across 21 intervals, producing 126 seed-by-interval observations. Repeated measures ANOVA showed statistically significant interval effects for validation loss, validation perplexity, and rolling volatility. Descriptive trajectories revealed rapid early improvement followed by non-monotonic degradation during later training intervals. Mean validation loss decreased from 8.3552 at initialization to 2.7996 near 4 million tokens, but increased to 3.9010 by the final checkpoint. Validation perplexity followed the same pattern, falling sharply early in training before rising later. Derived telemetry further showed recurrent validation-loss backslides and no interval-summary evidence of a stable phase under the predefined criteria. These findings suggest that compute-aware language model evaluation should examine training trajectories rather than endpoint metrics alone. In constrained compute settings, additional token exposure may increase computational cost without producing proportional generalization gains, and interval-level telemetry can reveal instability, regression, and diminishing returns that final metrics may obscure.

Large Language Models such as GPT-4o and GPT-5 achieve strong zero-shot performance on biomedical claim verification, but cost and opacity limit scalable use. We fine-tune three small LLMs: Phi-3-mini (3.8B), Qwen2.5-3B, and Mistral-7B, via QLoRA on SciFact and HealthVer, providing the first study of QLoRA models against GPT-4o and fine-tuned BioLinkBERT encoders. Mistral-7B QLoRA surpasses both GPT-4o and GPT-5 (up to 12% F1 gain) at a fractional cost using just 1,008 training examples. We conduct extensive in-domain and cross-domain evaluation: models trained on SciFact tested on HealthVer and vice versa, at matched sizes to isolate dataset structure from data quantity. We identify a previously unreported structural artifact in SciFact that inflates in-domain scores, and show through bidirectional out-of-domain evaluation that training on structurally sound data enables robust cross-domain transfer. We plan to release all code and adapter checkpoints.

This paper investigates reinforcement learning (RL) methods for improving tool-calling capabilities in multimodal small language model (SLM) agents. While existing works have explored various reward designs to improve agentic tool-calling ability, these approaches face inherent limitations for SLM training, especially under multimodal scenarios. First, many existing methods evaluate tool use correctness through exact matching against certain ground-truth or predefined formats. However, this assumption is often unsuitable for multimodal tasks, where multiple tool use paths may be valid and annotated tool trajectories are typically unavailable. Second, such sparse and brittle binary rewards provide little guidance on how to improve the underlying decision process, making them particularly difficult for multimodal SLM to learn from. To address these issues, we propose Input Attribution-Aware Policy Optimization (IAPO), an RL algorithm for improving tool use in multimodal SLM by aligning the model's attribution across input components with that of a stronger teacher. Experiments on Qwen2.5-VL-3B show that the proposed method improves visual question answering accuracy by an average of 3% across six test sets compared with existing visual tool use work, by helping the model attend to the most relevant input evidence.

Recently, language models have made rapid progress across various domains and applications. However, their capability for self-improvement, i.e., whether they are adept at recognising and correcting flaws in their own reasoning, remains dubious. In this study, we address this question by constructing a sufficiency test to rigorously examine the self-correction capabilities of small language models (SLMs). We propose a minimal three-step self-correction pipeline that collects initial SLM answers, prompts the same model to generate hints for its incorrect responses given the ground truth, and feeds the model the same question with its own feedback to refine the initial answer. We evaluate a variety of instruction-tuned and reasoning SLMs in this experimental setup on arithmetic and logical reasoning benchmarks. Our findings show that SLMs with injected hint sentences yield only a 4.4 percent gain over initial question-answering accuracy. Even though the correct answer was provided alongside the model's incorrect reasoning, the evaluated SLMs fail to understand what was missing in their reasoning and show minimal semantic difference between hints that lead to corrections and ones that do not. Furthermore, our experiments show that longer hints are positively correlated with incorrect final answers, suggesting that longer deliberation on problems can hinder the reasoning process, meaning that SLMs do not necessarily scale in performance with a larger compute budget.

Current advancements in Audio Reasoning rely on massive Large Audio-Language Models (LALMs), hindering deployment in resource-constrained environments. We introduce TinyGiantALM, a compact 1.5B efficiency-oriented alternative. Instead of brute-force scaling, we propose an Instruction-Aware Feature Refinement framework using a Query-guided Projector and Semantic Gating to filter acoustic signals based on user intent. On the MMAR benchmark, TinyGiantALM achieves 46.4% zero-shot accuracy, significantly outperforming 7B-13B baselines. While a reasoning gap in logical narrative remains versus 30B+ models and certain trade-offs exist in overly dense or spatial scenes, our approach notably surpasses models up to 8x larger in disentangling mixed-modality environments. These findings demonstrate that architectural precision offers a tangible pathway to secure robust perception capabilities on edge-friendly scales.

Effective skill grounding is essential for deploying reusable skills in embodied agents, as even minor embodiment or environmental differences can render an entire skill incompatible. This challenge is particularly pronounced in embodied settings, where agents must operate in dynamic, partially observable environments without access to large language models (LLMs). In this setting, reliance on LLMs is impractical, while small language models (sLMs) remain insufficient for the effective skill grounding required for reliable long-horizon control. We present RECENT, a refactoring-centric agent framework that enables efficient skill grounding with sLMs by decoupling skill semantics from embodiment- and environment-specific execution binding. By representing skills as executable code, RECENT preserves the semantic intent encoded in a skill's control structure while grounding it by modifying only execution bindings through localized refactoring, rather than regenerating code from scratch. We evaluate RECENT across diverse skill grounding scenarios spanning multiple robot embodiments in dynamic environments, demonstrating robust long-horizon performance when deployed with an sLM. Across all scenarios, RECENT achieves the best performance among sLM-based Code-as-Policies (CaP) methods and matches the task performance of LLM-based CaP.

The rise of edge-based machine learning has enabled distributed adaptation of language models across mobile and IoT devices, offering privacy preservation and real-time responsiveness. However, distributed fine-tuning of language models on untrusted or heterogeneous edge nodes introduces new vulnerabilities. Compromised or unreliable devices can inject poisoned updates, leading to stealthy model manipulation or convergence degradation. Classical defenses such as robust aggregation or temporal anomaly detection operate on a single global model and are therefore limited in detecting coordinated or persistent poisoning. This work proposes a new system-level defense based on model multiplicity. Instead of maintaining one global model, the system rotates or concurrently trains multiple small language models (e.g., DistilGPT-2), each updated by independently sampled subsets of edge nodes. These models evolve under distinct training trajectories, creating multiple independent views of the same distributed population. Divergence between models quantified through gradient similarity, loss evolution, or parameter variance serves as a signal of anomalous or adversarial behavior. When one model deviates significantly from the ensemble mean, the system flags its contributing nodes for isolation or re-weighting. We implement this framework and evaluate it on edge-scale simulations of Small Language Model (SLM) training under varying heterogeneity and attack conditions. Results show that model multiplicity enables earlier and more reliable detection of poisoning compared to classical single-model defenses such as Flanders and Robust methods. Our findings demonstrate that diversity in model evolution can serve as a practical and effective defense mechanism for secure distributed learning on resource-constrained edge devices.

Large language models (LLMs) are widely used as judges for evaluating model outputs, but their high cost, latency, and opacity limit scalability. We introduce SLMJury, a framework for evaluating small language models (SLMs) as judges across two paradigms: closed-ended binary correctness and open-ended quality scoring. We benchmark 16 SLM judges (0.6B-14B parameters) from four model families across ten benchmarks: eight closed-ended tasks spanning mathematical, scientific, and general reasoning (N=64,824 judgments per configuration), plus SummEval and MT-Bench for summarization and conversational scoring. We formalize judging as a budget-conditioned function and study five dimensions. Four findings emerge. (1) The overthinking effect is domain-dependent: for most judges quick 10-token verdicts match or beat extended reasoning on mathematical judging (by 2-7% where they help), while reasoning wins on general tasks by up to 23%. (2) Domain generalization separates model families, with math-to-general accuracy gaps ranging from under 10% to nearly 40%. (3) Closed-ended and open-ended judging draw on different capabilities: the best binary judge (Phi-4) drops to rank 9 on MT-Bench, while reasoning-trained models invert this ordering. (4) Under the Reflect-Critique-Refine (RCR) debate protocol, multi-agent debate degrades accuracy across all tested configurations, whereas the top judges resist six adversarial personas with <=0.55% variance. Reliable automated evaluation does not require large proprietary models, yet no single SLM dominates. The leaderboard is available at https://anishh15.github.io/SLMJury/, and our framework code and pip package are publicly available at https://github.com/anishh15/SLMJury and https://pypi.org/project/slmjury/.

Deploying Small Language Models (SLMs) on edge devices requires efficient fine-tuning strategies that adapt models to new tasks without degrading their general capabilities. In this study, we benchmark five sub-1B models (135M-1B) on mathematical reasoning tasks and uncover a critical vulnerability: Full Fine-Tuning (Full FT) actively harms performance in models under 300M parameters, often dropping accuracy below zero-shot baselines. This "negative transfer" makes Parameter-Efficient Fine-Tuning (PEFT) not just an efficiency preference, but a stability requirement. We find that while Low-Rank Adaptation (LoRA) and Weight-Decomposed LoRA (DoRA) perform comparably, their strengths vary by task; DoRA excels in complex reasoning (GSM8K), while LoRA dominates pattern matching (OrcaMath). In particular, Full FT is outperformed by LoRA on aligned models (Qwen2.5-0.5B) and even by simple 5-shot In-Context Learning on the smallest architectures (SmolLM2-135M). Based on these findings, we recommend defaulting to PEFT for all aligned sub-1B models and caution against Full FT for any architecture smaller than 500M parameters to prevent catastrophic forgetting. Reproduction of this work can be found at https://github.com/gulguluu/tiny-slm-finetune-compare.

Efficient large language model (LLM) serving is increasingly constrained by deployment cost. Quantization is a key technique for reducing serving cost, yet even state-of-the-art 4-bit quantizers exhibit a noticeable quality gap from FP16, particularly for smaller models where low-bit serving is most beneficial. We identify a fundamental cause of this gap: quantization error is highly input-dependent and varies substantially across tokens, while existing post-quantization compensation methods are static and apply identical corrections to all inputs. As a result, easy tokens are over-corrected while hard tokens remain under-corrected. We present SPEAR, a system for post-quantization error-adaptive recovery that improves low-bit LLM serving. SPEAR introduces lightweight Error Compensators (ECs) modulated by per-token gates and places them only at the most error-sensitive layers identified through a CKA-guided entropy-aware diagnostic. This focuses a small parameter budget where it is most effective. Efficient deployment of ECs presents several systems challenges, including additional computation, tensor-parallel synchronization caused by input-dependent gating, and latency instability across configurations. SPEAR addresses these issues through adaptive kernel-fusion dispatch, combining an epilogue-integrated peer-reduction kernel with P2P dual-write to fuse the post-EC computation into low-bit GEMMs, and an SLO-constrained EC-aware scheduler for predictable serving performance. Across challenging per-channel quantization settings, SPEAR recovers 56-75% of the perplexity gap between W4 and FP16 while adding less than 1% model memory overhead and maintaining latency comparable to a widely used 4-bit serving deployment.

Proving Shannon-type entropy inequalities is a fundamental task in information theory that often requires constructing non-trivial linear combinations of known constraints, which is a combinatorial search problem that scales poorly with the number of random variables. We investigate whether small-scale large language models (0.6B--1.7B parameters), fine-tuned on atomic proof steps and combined with guided beam search, can automate this process. On a held-out test set of 60 inequalities spanning n=10 to 15 variables, our 0.6B fine-tuned model achieves an 85% proof success rate with tree search. GPT-5.5 solves 1.7% samples under zero-shot prompting while Psitip solves 33.3% samples. A systematic ablation study across training context length (4096 vs.\ 8192 tokens) and data distribution (n=9-skewed vs not skewed) reveals that a 4096-token not skewed training distribution yields the best performance, with extended context and skewed data providing no marginal benefit. We further identify two dominant failure modes -- format failures and step quality degradation -- and verify that the beam-scoring heuristic is essential via a controlled ablation (random scoring reduces success from 83% to 23%).

Small language models (SLMs) are efficient and scalable, but their multilingual capabilities degrade severely at sub-billion scales, especially for Southeast Asian (SEA) languages. We introduce DuDi, a dual-signal multilingual distillation framework that combines an online sequence-level signal with off-policy and on-policy token-level signals. DuDi further uses a cross-lingual verbalizer to refine teacher feedback and improve teacher-student transferability in multilingual settings. Experiments on SEA-HELM across multiple model families, scales, and teacher-student settings show that DuDi consistently outperforms competitive distillation baselines. Ablations and analyses confirm that sequence-level optimization, token-level supervision, and cross-lingual verbalization provide complementary and transferable learning signals for multilingual SLMs.

Post-training Small Language Models (SLMs) for reasoning typically follows an SFT-then-RL pipeline, yet existing work rarely considers what data should be learned at each stage. We argue that data strategy should be aligned with the distinct roles of SFT and RL: SFT is better suited for acquiring not-yet-mastered reasoning skills, while RL is better suited for consolidating skills that the model can already partially access. Based on this principle, we propose a difficulty-aware SFT-then-RL framework that organizes training data into stage-specific sets. For hard samples in the SFT stage, we introduce a Bridge mechanism that transforms raw teacher-generated reasoning traces into more learnable supervision for SLMs. For hard samples that remain unsolved during RL, we apply Critique Fine-Tuning by converting all-zero-reward failures into diagnostic, repair, and new reasoning trace supervision for the next SFT stage. Experiments on two SLMs across five reasoning benchmarks show that our method consistently improves over representative SFT, distillation, and RL baselines. Our results highlight the importance of coordinating data difficulty across SFT and RL for effective SLM reasoning post-training.

Aggressive weight quantization to 2-bit precision offers substantial throughput and memory gains for large language model (LLM) inference, but typically incurs severe accuracy degradation. These gains are particularly relevant for edge and on-device deployment, where memory capacity and bandwidth are primary constraints. In this work, we extend Recover-LoRA -- a lightweight, data-free accuracy recovery method originally developed for general model weight corruption -- to the setting of ultra-low-bit quantization. We propose a selective mixed-precision strategy in which only gate and up projection layers of the MLP are quantized to 2-bit (W2), while all other linear layers remain at higher precision, yielding a mixed-precision GateUp configuration. We demonstrate via roofline analysis across three model families (4B--20B) and two hardware platforms that a W4/W2-GateUp deployment (4-bit base with 2-bit gate/up) delivers 7.5--23.3% TPS improvement over uniform W4 depending on model and context length, while confining quantization error to a predictable subset of layers. We then apply Recover-LoRA -- training low-rank adapters on the quantized layers via logit distillation with synthetic data -- to recover accuracy lost from 2-bit quantization of the gate and up layers. In a case study on Qwen3-4B, Recover-LoRA achieves 80--95% accuracy recovery on 9 of 12 benchmarks, using only 10k synthetic training samples and no labeled data. We further demonstrate that synthetic data performs comparably to curated labeled data for distillation-based recovery, and that recovery generalizes to out-of-distribution evaluation tasks. Our results present Recover-LoRA as a practical post-quantization accuracy recovery tool for aggressive weight compression in deployment settings.

When prompting language models for psychometric assessment, researchers assume that the responses reflect the injected persona and the meaning of the survey item. We test this premise using a diagnostic design that crosses five semantically distinct baseline personas with five semantically equivalent variants of each of four prompt components (persona wording, task instruction, item wording, option symbol). Measuring the 1-Wasserstein distance between the resulting response distributions and partitioning the variation among the five components allows for the separation of target effects from prompt artifacts. We apply the framework to 13 open-weight small language models (0.6B to 14B) on the Big Five Inventory and the Short Dark Triad. We find that in most models, the task instruction and option symbol displace response distributions further than paraphrasing the persona description or the item itself. For a substantial share of items, the artifact share of explained variation exceeds 50%; non-semantic changes of the prompt account for more response variation than the baseline personas. Our framework lets researchers quantify these prompt artifacts before interpreting psychometric output.

Deploying language-driven agents in robotics requires evaluations that reflect real-world task demands: natural-language instructions with reproducible outcomes. Such agents must connect language models to callable perception and control tools, and be assessed using deployment-critical metrics including latency, accuracy, and error modes. We present SCOPE (Simulation and Camera Operations for Perception and Evaluation), a modular agent for natural-language, open-vocabulary pan-tilt-zoom (PTZ) camera control and visual scene understanding, designed explicitly for edge deployment. SCOPE operates both in a Blender-based simulation environment and on a physical PTZ camera, executing all perception, planning, and control locally at the deployment site using edge-accessible compute. We release a 536-task benchmark spanning QA, single- and multi-step commands, counting, spatial reasoning, descriptions, and optical character recognition in a Blender-based simulation environment that exposes realistic PTZ control affordances. Execution traces are combined with an LM-as-Judge to evaluate latency, accuracy, and error modes. We evaluate 19 planner-perception model combinations pairing Qwen3 small language models (SLMs) with Moondream and Qwen vision-language models (VLMs). Stronger SLMs substantially reduce hallucinations and improve tool routing, leading to more reliable closed-loop behavior. Once a sufficiently capable SLM is used, perception becomes the dominant performance bottleneck. Mixture-of-Experts models on both the planning and perception side consistently match or exceed dense alternatives at latencies and memory footprints comparable to much smaller networks. Quantization provides additional efficiency gains with minimal accuracy degradation, identifying a practical, sim-to-real validated design point for real-time, edge-feasible language-driven PTZ control.

We explore Riemannian optimization techniques for rank-factored matrix parameters, targeting contemporary deep learning applications. We examine ten points in the algorithm design space: two geometries for rank-$r$ matrices, three geometries for rank-$r$ partial isometries, and block-matrix variants of these five, where factors are shared across block-rows and block-columns. We apply our methods to the multihead attention parameters in small language models. After tuning learning rates, our methods do not conclusively outperform an AdamW baseline. Our implementations are available online.

Small Language Models (SLMs) offer a balance between capability and computational feasibility. Neural scaling laws inform their optimal training, suggesting that they possess rich internal representations that scale with their size. However, deploying even these SLMs can be challenging under strict resource constraints. Language model probing provides methods for analyzing the linguistic knowledge encoded in a model's internals. We propose ProbScale, a framework that unifies insights from scaling laws and probing to identify parameter-efficient subnetworks within pre-trained SLMs. ProbScale utilizes the high-quality representations of well-scaled SLMs and uses task-specific probes to mathematically quantify the relevance of each layer for target downstream capabilities. This allows selecting subnetworks that optimally trade off performance against parameter size. We formulate the subnetwork selection as finding a layer subset maximizing aggregated, task-weighted probe performance under a parameter budget. Experiments on representative SLMs such as RoBERTa-Large and T5-Base demonstrate that ProbScale identifies subnetworks achieving significant parameter reduction, from 5 to 10 times, while maintaining high performance (95% to 98% of the original SLMs) on targeted tasks, outperforming heuristic baselines.

Short-form text rewriting is a constrained variant of paraphrasing in which limited context and high semantic density leave little room for variation. While large language models perform well on general paraphrasing, small language models (SLMs) often struggle with semantic fidelity and hallucination robustness in short-form settings. In this work, we present an empirical study of adapting an SLM, Phi Silica, for short-form rewrite through dataset curation, prompt distillation, parameter-efficient fine-tuning, and evaluation. We curate a dataset of short presentation-style text from public slide decks and use GPT-5-chat both to generate rewrite supervision and to conduct LLM-as-a-judge evaluation. Our results show that finetuning improves semantic fidelity, reduces hallucinations, and increases preference win rate against GPT-5-chat rewrites. The findings suggest that targeted adaptation for SLMs can substantially narrow the gap to cloud models and provide practical guidance for adapting SLMs to precision-critical rewrite tasks.

In real-world deployment, LLMs are often adapted continually across tasks to keep LLMs up-to-date in production, where new fine-tuning should preserve previously learned skills. However, indiscriminately mixing tasks can dilute task specialization, while sequential fine-tuning (full-parameter or low rank adaptation) often causes catastrophic forgetting due to destructive overwriting. Replay-based continual tuning and maintaining separate task-specific adapters can mitigate forgetting, but introduce additional compute, storage, and management overhead. Recognizing the redundancy of LLM parameters for any single task, we reframe continual task adaptation as task-specific parameter discovery via adaptation-aware probing: a short warm-start probe exposes a task's adaptation trace, enabling us to identify and isolate the small subset of parameters essential for each task to mitigate catastrophic forgetting. Building on this view, we introduce TRACE, a novel approach for discovering Task-specific paRameters via Adaptation-aware probing for Continual finE-tuning. We perform a short warm-start fine-tune to derive task-specific core parameters by comparing the warm-started and pre-trained models. Core parameters are identified via two strategies: importance scoring (L$_2$ norm and Fisher Information) and specificity analysis (cosine similarity of parameter updates). In continual fine-tuning settings, only the active task's core parameters are updated while others remain frozen, preserving prior knowledge. We conduct extensive experiments across multiple standard benchmarks to demonstrate the superior performance of our proposed method. Additionally, we validate the generalization of our method through a cross-model and scale transferability study, demonstrating a "small-to-large" paradigm that guides the fine-tuning of large-scale models under resource constraints.

Reinforcement learning with verifiable rewards (RLVR) trains language models using programmatically checkable signals such as unit-test outcomes, enabling direct optimization for functional correctness in code generation. We conduct an empirical study of RLVR for Python code generation on the MBPP benchmark using two small models (Qwen3-0.6B and Llama3.2-1B) with LoRA fine-tuning. Across multiple reward formulations such as: unit-test-only rewards, static-analysis-only shaping via the Ruff linter, and a combined reward, we compare group-based policy optimization variants (GRPO and GSPO) and evaluate both functional correctness and behavioral diagnostics. In our experimental setting, RLVR improves pass@1 on MBPP test by up to 13 percentage points under proposed combined reward configuration. However, we find that reward shaping can induce systematic behavioral shifts: using only static-analysis penalties may bias the policy toward shorter completions that reduce lint errors without reliably improving functional correctness. In contrast, combined rewards mitigate this degeneration and yield more stable trade-offs between correctness and style constraints. Overall, our results highlight that RLVR effectiveness for code generation is highly sensitive to reward design and optimization granularity, and that diagnostics beyond pass@1, including generation length, Ruff severity profiles, and execution error types are useful for identifying failure modes.

The design space of agentic AI inference spans two extremes: frontier large language models (LLMs), typically hosted in the cloud and offering strong performance across a wide range of tasks at substantially high cost, and more cost-efficient small language models (SLMs), which are amenable to on-device inference. Hybrid multi-agent systems (MASs) combining on-device and cloud models offer a promising middle ground, but they also introduce a complex and poorly understood design space in which task accuracy, monetary cost, and edge energy consumption are tightly coupled; in the absence of general design principles, hybrid components, although not the most prevalent choice, are typically introduced through ad hoc decisions tailored to specific domains. In this work, we examine this design space more systematically. We adapt two representative MAS architectures to support hybrid inference and study how individual design choices shift the operating point along the Pareto frontier of power, cost, and performance. Our findings paint a nuanced picture of hybrid MAS design: while SLMs can effectively benefit from LLM assistance, the optimal architecture is highly task-dependent, and greater frontier-level compute does not consistently translate to better performance.

Large language models (LLMs) demonstrate strong chain-of-thought (CoT) reasoning abilities, while smaller models (<= 3B parameters) significantly underperform on multi-step reasoning tasks. Based on empirical analyses of the Qwen-2.5 model family on math reasoning benchmarks, we find that more proficient reasoning is associated with fewer reasoning steps but higher information density per step, a property we term Dense Reasoning. Motivated by this observation, we propose DenseSteer, a training-free inference-time steering framework that enhances small-model reasoning by modulating internal representations toward dense reasoning patterns. Experiments show that our method yields consistent accuracy improvements without increasing token-level Negative Log-Likelihood, highlighting dense reasoning as an effective structural approach to mathematical problem solving.

Self-evolving scientific agents capable of conquering the hard tail of formal mathematics require Compositional Learning Behaviours (CLBs) -- the capacity to ground and recombine novel symbolic structures in context, beyond mere recombination of prelearned atoms. We propose S2B-LM, an adaptation of the CLB-evaluating Symbolic Behaviour Benchmark that removes numerical processing as a confound and adds chain-of-thought scaffolding to elicit rather than merely probe latent CLB competency. Cross-evaluating ten Lean~4 theorem provers on CLB competency in S2B-LM and miniF2F whole-proof performance, we find correlational and causal evidence of our claim: First, a necessary-condition analysis via quadrant test yields $p=0.004$, with model scale being ruled out as a confound. Second, extracting a CLB-encoding activation direction from DeepSeek-Prover-V2-7B using S2B-LM traces via Contrastive Activation Addition and applying it during miniF2F whole-proof generation on the AIME subset, CLB suppression collapses solve rate from $32.3\%$ to $2.9\%$, without loss of coherence, while suppressing a random activation direction of equal magnitude leaves it at $31.9\%$. Together, these results show that CLB competency is necessary but not sufficient for the hard tail of formal mathematical verification.

Recently, there has been increased interest in Small Language Models (SLMs), which are fast, show good performance, and have lower hardware demands than large language models (LLMs). However, SLMs hallucinate more frequently than LLMs, impacting their ability to solve complex multi-step reasoning problems as early mistakes cascade to the final response. To address this, existing works think-first followed by iterative retrieval to reduce hallucination. We argue that the think-first strategy is not always necessary as we find that: (i) SLMs are often accurately confident in their initial answer and, (ii) hallucinations can actually be beneficial for honing in on the true answer. As such, we position our work as an inversion of this strategy, i.e., answer first-reason later. We propose a cognitively-inspired framework where the model is first allowed to quickly answer the question (System-I (zero-shot)) and then resorts to deeper thinking (System-II) based on evidence retrieved from a knowledge source using the initial hypothesis. By combining System-I and System-II style thinking, we show that our method can outperform prior work that takes the traditional think-first route on various multi-step question-answering benchmarks.

Many organizations lack computational resources to fine-tune large language models (LLMs) on private (unshareable) data for better utility, while fine-tuning tiny language models (TinyLMs) alone performs poorly. To address this bottleneck, we propose a data-free knowledge distillation framework that generates LLM update vectors based on TinyLMs fine-tuned on private data. An update vector is a vector of parameter changes from an initial model to its fine-tuned version on a dataset, capturing the effect of cumulative gradient steps during fine-tuning. The key idea of our framework is a novel Gradient Transformer that transforms TinyLM's update vectors into LLM's update vectors. As derived from shadow datasets, Grad-Transformer captures the correlation between TinyLM and LLM update vectors, enabling third-party providers to generate LLM update vectors given the organization's TinyLM update vectors without accessing the organization's private data. The framework supports multi-organization collaboration to jointly update LLMs, improving performance and cost-efficiency. Extensive experiments across language modeling and reasoning tasks show that Grad-Transformer remarkably outperforms state-of-the-art knowledge distillation baselines, even under strict differential privacy protection.

This paper examines the specialization of Small Language Models (SLMs) with up to 4 billion parameters for generating artifacts in domain-specific languages (DSL). Kubernetes manifests are chosen as the target domain. We propose the context-instrumental data distillation method: the source corpus is formed through synthetic generation and, in an extended scheme, through reverse instruction generation from real Kubernetes YAML files, with pairs included in training only upon passing external validators and matching the domain context model. Unlike classical KL-divergence knowledge distillation, the baseline implementation reduces to supervised fine-tuning on instrumentally verified examples. The experimental section presents a pilot implementation under resource-constrained conditions: the DeepSeek-V4 Flash API serves as the teacher for synthetic generation, while Qwen2.5-Coder-1.5B-Instruct is fine-tuned via LoRA on CPU. On the K8s-Distill-Pilot corpus (train_1200, validation_100, test_200), we achieved full-pass@1 = 91.5% (183/200) with a stricter prompt formulation and max_new_tokens=768. The key empirical finding is that for Kubernetes YAML, result quality in the pilot depended more on strict output format requirements than on simply increasing the number of training examples.

Large language models often generate confident but incorrect answers rather than abstaining when uncertain. This problem is particularly acute for small language models (SLMs), where computational constraints and autonomous operation amplify the need for reliable uncertainty detection. We propose Second Guess, a lightweight, parameter-free prompting technique for abstention in multiple-choice question answering (MCQA) that is well-suited for SLMs. Our key empirical insight is that models which truly know an answer will select it consistently, while uncertain models exhibit unstable behavior when an ``I don't know'' option is added. Evaluated on four open models (2B-8B parameters) and four benchmarks, Second Guess achieves the highest composite risk improvement of 10.81%. Notably, it maintains an 8% composite risk improvement on fine-tuned models where entropy-based methods degrade, and improves most for lower-performing models. All code and results required to reproduce this work is available in https://github.com/Mystic-Slice/second-guess

Chronic dermatologic diseases such as pemphigus require long-term follow-up, generating extensive longitudinal clinical documentation that is difficult to review comprehensively during routine visits and increasing clinician workload as well as the risk of missing critical historical information. We evaluated whether a locally deployed, privacy-preserving small language model (SLM) could retrieve structured clinical features and generate longitudinal summaries from long-term dermatology follow-up records. In this retrospective case series, thirty pemphigus patients contributed 541 visit notes that were aggregated into full longitudinal records (89,336 words); 56 clinically relevant features were annotated by two expert dermatologists. The locally deployed SLM (Qwen3 4B Thinking 2507) was queried with each complete record to retrieve 56 features and generate one final report summaries. Across 1,680 feature retrieval tasks, mean accuracy was 82.25%. Dermatologists' ratings of AI-generated summaries were high for overall quality (8.23-8.47), clinical accuracy (7.93-8.20), and usefulness (8.47-8.50), with no significant inter-evaluator differences and an overall preference for AI summaries in 53.3% of evaluations. These findings suggest that privacy-preserving, locally deployed SLMs can outperform medical experts and reliably generate clinically meaningful longitudinal summaries. SLMs may support clinical decision-making when integrated with appropriate oversight.

In the competitive landscape of sponsored search, balancing retrieval quality with production latency is a critical challenge. While large retrieval models based on Small Language Models (SLMs) such as Qwen3-Embedding-4B/8B set strong upper bounds on public benchmarks, their deployment in high-throughput, latency-sensitive environments remains impractical. In this paper, we present HARNESS-LM (HLM), a three-phase training framework for transferring the capabilities of large-scale retrievers into compact, cost-efficient models. The approach comprises: (1) training a high-performance reference ("teacher") retriever by fine-tuning a billion-parameter-scale SLM; (2) aligning query representations via an L2 objective to distill knowledge into a sub-600M parameter student encoder; and (3) applying a final contrastive refinement stage to optimize the student for retrieval performance. We also present a comprehensive empirical study of key design choices, including alignment objectives, embedding dimensionality, model scale, architecture, and optimization strategies, to identify configurations that are most effective in production settings. On a real-world Bing Ads evaluation benchmark, HLM recovers over 98% of the reference retriever's precision across multiple settings, while delivering up to 27x lower online query-encoder latency and 20x higher throughput on NVIDIA A100 GPUs. Online A/B testing on Bing Ads further shows a +1% Revenue, +0.6% Impression, and +0.4% Click uplift over the current ensemble of retrievers running in production with the deployed 190M parameter model, clearly highlighting the practical efficacy of the HLM recipe in a real-world sponsored search setting.

Transformer-based language models such as BERT having 110M+ parameters have revolutionized natural language understanding, yet their internal mechanisms remain largely opaque to researchers and practitioners. Traditional attention-based interpretability methods often emphasize structurally important but semantically weak tokens such as punctuation marks rather than meaningful semantic relationships. This work introduces a lightweight and model-agnostic framework for quantifying token-level representational importance using hidden-state activation strengths at Layer 8 of BERT. The proposed Activation Flow Network (AFN) framework computes Token Activation Strength using the L2 norm of Layer-8 hidden representations, enabling direct ranking of semantically salient tokens. The study further introduces a threshold-based activation bucket formulation that partitions tokens into HIGH-activation and LOW-activation groups using an empirical upper-quartile activation boundary. Experimental observations demonstrate that semantically meaningful content words consistently occupy the HIGH-activation bucket and dominate representational activation shifts, while structurally supportive tokens contribute comparatively less. The results suggest that Layer 8 acts as a critical semantic consolidation zone balancing structural and semantic information processing. By revealing how activation magnitudes concentrate around semantically informative tokens, this work provides an interpretable and computationally efficient alternative to attentioncentric analysis, contributing toward transforming BERT from a "black box" into a more transparent "glass box" model for natural language understanding.

Production LLM systems increasingly require machine-readable outputs: JSON objects, typed traces, regex-constrained fields, and tool-call schemas. This paper targets on-device and low-cost small language model (SLM) deployments, where sub-3B models are attractive for privacy, latency, and commodity hardware but have limited capacity to satisfy schemas while solving tasks. The usual engineering assumption is that hard output constraints improve reliability without changing the underlying answer. We show that this assumption is unsafe for small models. We introduce \emph{constraint tax}, a measurement protocol for isolating the answer and executable-accuracy loss caused by structured-output constraints at fixed model, fixed task distribution, and fixed problem instances. Across 15,000 commodity-GPU generations with Qwen2.5-0.5B, Qwen2.5-1.5B, and SmolLM2-1.7B, hard answer-only schema decoding raises schema validity from 61.5% to 100.0%, but lowers answer accuracy from 19.7% to 11.0% and increases wrong-valid-schema outputs from 49.5% to 88.9%. The strongest industry analogue is a deterministic calendar tool-call task: Qwen2.5-1.5B achieves 91.5% executable accuracy with prompt-only JSON but only 48.0% under the same hard tool-call schema, while both modes are 100.0% schema-valid. The error is semantic, not structural. We also show that the 3B boundary still pays a direct-schema tax and that delayed packaging supports a constructive design pattern: reason free, constrain late. The practical conclusion is direct: production systems should report schema validity, answer accuracy, executable accuracy, and wrong-valid-schema rate separately.