arXiv subjects

Kaleem Ullah Qasim

Publications and source records attributed to Kaleem Ullah Qasim.

7 recordsLinked to original sources

arXiv2026

The Residual Stream Is All You Need: On the Redundancy of the KV Cache in Transformer Inference

The key-value (KV) cache is widely treated as essential state in transformer inference, and a large body of work engineers policies to compress, evict, or approximate its entries. We prove that this state is entirely redundant: keys and values at every layer are deterministic projections of the residual stream, and recomputing them from a single residual vector per token incurs exactly zero reconstruction error, not approximately, but bit-identically. We verify this across six models from four architecture families (135M to 4B parameters). Cross-task residual patching at every layer produces D_KL = 0 between patched and original output distributions, confirming that the residual stream satisfies a Markov property and is the sole information-carrying state. Removing the cache entirely and recomputing from scratch yields token-identical output under greedy decoding on all models tested. We build on this result with KV-Direct, a bounded-memory inference scheme that checkpoints residual vectors (5 KB per token on Gemma 3-4B) instead of full KV pairs (136 KB), recomputing keys and values on demand. Over 20 conversation turns, KV-Direct holds peak memory at 42 MB while the standard cache grows past 103 MB. Against five eviction baselines (H2O, StreamingLLM, SnapKV, TOVA, window-only), KV-Direct maintains 100% token match at every cache budget; all baselines degrade to 5-28%. A per-operation latency analysis shows recomputation runs up to 5x faster than reading cached tensors at moderate batch sizes. Code is available at https://github.com/Kaleemullahqasim/KV-Direct.

cs.LG↗

arXiv2026

VERIFY-RL: Verifiable Recursive Decomposition for Reinforcement Learning in Mathematical Reasoning

Training language models to solve complex mathematical problems benefits from curriculum learning progressively training on simpler subproblems. However, existing decomposition methods are often heuristic, offering no guarantees that subproblems are simpler, that solving them aids the parent task, or that their relationships are mathematically grounded. We observe that symbolic differentiation provides a natural structure for verified decomposition: calculus rules explicitly define how expressions reduce to simpler components with provable properties. We introduce Verify-RL, a framework where every parent-child decomposition satisfies three verifiable conditions: strictly decreasing structural complexity, solution containment, and formal rule derivation. Unlike heuristic methods where a significant fraction of decompositions are invalid our properties admit automatic verification through symbolic computation, achieving "verification by construction" Experiments demonstrate that eliminating invalid decompositions yields sizable gains, accuracy on the hardest problems more than doubles from 32% to 68%, with a 40% relative improvement overall.

cs.AI↗

arXiv2025

Accelerating Training Speed of Tiny Recursive Models with Curriculum Guided Adaptive Recursion

Background: Recursive reasoning models achieve strong performance through iterative refinement, allowing small networks to match large language models. However, training is computationally expensive, often requiring 36 GPU-hours for Sudoku extreme. Existing models use fixed recursion depth and uniform supervision weighting, leading to inefficient training. Objectives: We propose CGAR (Curriculum-Guided Adaptive Recursion), applying curriculum learning to architectural depth. CGAR introduces Progressive Depth Curriculum (PDC) to dynamically adjust recursion depth and Hierarchical Supervision Weighting (HSW) to apply exponentially decaying importance to supervision steps. Methods: PDC implements a three-stage schedule transitioning from shallow (2, 1) to full depth (6, 3) configurations, providing 41.4% FLOPs reduction. HSW applies exponential decay to supervision steps, achieving 40% gradient variance reduction and accelerated convergence. Results: On Sudoku-Extreme, CGAR achieves 1.71x training speedup (10.93 to 6.38 hours) with only a 0.63% accuracy drop (86.65% to 86.02%). PDC alone achieves 2.26x speedup with 85.47% accuracy, showing a Pareto improvement in efficiency and quality. HSW provides 1.61x speedup. CGAR-trained models show superior inference efficiency with 100% halting accuracy and 11% fewer reasoning steps. Conclusions: CGAR enables efficient training of recursive models on modest hardware. By treating depth as a scheduled parameter, it achieves substantial savings and prevents overfitting, making these models practical for neurosymbolic AI and program synthesis. https://github.com/Kaleemullahqasim/CGAR and huggingface.co/Kaleemullah/trm-cgar-sudoku.

cs.LG↗

arXiv2025

ASTIF: Adaptive Semantic-Temporal Integration for Cryptocurrency Price Forecasting

Financial time series forecasting is fundamentally an information fusion challenge, yet most existing models rely on static architectures that struggle to integrate heterogeneous knowledge sources or adjust to rapid regime shifts. Conventional approaches, relying exclusively on historical price sequences, often neglect the semantic drivers of volatility such as policy uncertainty and market narratives. To address these limitations, we propose the ASTIF (Adaptive Semantic-Temporal Integration for Cryptocurrency Price Forecasting), a hybrid intelligent system that adapts its forecasting strategy in real time through confidence-based meta-learning. The framework integrates three complementary components. A dual-channel Small Language Model using MirrorPrompt extracts semantic market cues alongside numerical trends. A hybrid LSTM Random Forest model captures sequential temporal dependencies. A confidence-aware meta-learner functions as an adaptive inference layer, modulating each predictor's contribution based on its real-time uncertainty. Experimental evaluation on a diverse dataset of AI-focused cryptocurrencies and major technology stocks from 2020 to 2024 shows that ASTIF outperforms leading deep learning and Transformer baselines (e.g., Informer, TFT). The ablation studies further confirm the critical role of the adaptive meta-learning mechanism, which successfully mitigates risk by shifting reliance between semantic and temporal channels during market turbulence. The research contributes a scalable, knowledge-based solution for fusing quantitative and qualitative data in non-stationary environments.

cs.AI↗

arXiv2025

Complexity Agnostic Recursive Decomposition of Thoughts

Large language models often fail on multi-step reasoning due to fixed reasoning strategies that ignore problem specific difficulty. We introduce CARD (Complexity Agnostic Recursive Decomposition), a framework that predicts problem complexity before generation and adapts decomposition accordingly. Our system comprises MRCE (Multi-dimensional Reasoning Complexity Estimator), a 0.6B Qwen model predicting 30 fine-grained features from question text and a two-stage recursive solver: (1) hierarchical decomposition into K steps based on task profile and (2) per-step thought budget allocation (1, 5-9, or 10 thoughts) via recursive MRCE profiling. Evaluated on three reasoning models (Qwen3-0.6B, DeepSeek-R1-Distill-Qwen-1.5B, Qwen3-1.7B), CARD achieves 81.4% to 89.2% accuracy on GSM8K while reducing token cost by 1.88x to 2.40x compared to fixed decomposition baselines. On MATH-500, CARD reaches 75.1 to 86.8% accuracy using 1.71x to 5.74x fewer tokens. Our results demonstrate that preemptive complexity estimation enables both higher accuracy and significant efficiency gains.

cs.CL↗

arXiv2025

MARBLE: A Multi-Agent Rule-Based LLM Reasoning Engine for Accident Severity Prediction

Accident severity prediction plays a critical role in transportation safety systems but is a persistently difficult task due to incomplete data, strong feature dependencies, and severe class imbalance in which rare but high-severity cases are underrepresented and hard to detect. Existing methods often rely on monolithic models or black box prompting, which struggle to scale in noisy, real-world settings and offer limited interpretability. To address these challenges, we propose MARBLE a multiagent rule based LLM engine that decomposes the severity prediction task across a team of specialized reasoning agents, including an interchangeable ML-backed agent. Each agent focuses on a semantic subset of features (e.g., spatial, environmental, temporal), enabling scoped reasoning and modular prompting without the risk of prompt saturation. Predictions are coordinated through either rule-based or LLM-guided consensus mechanisms that account for class rarity and confidence dynamics. The system retains structured traces of agent-level reasoning and coordination outcomes, supporting in-depth interpretability and post-hoc performance diagnostics. Across both UK and US datasets, MARBLE consistently outperforms traditional machine learning classifiers and state-of-the-art (SOTA) prompt-based reasoning methods including Chain-of-Thought (CoT), Least-to-Most (L2M), and Tree-of-Thought (ToT) achieving nearly 90% accuracy where others plateau below 48%. This performance redefines the practical ceiling for accident severity classification under real world noise and extreme class imbalance. Our results position MARBLE as a generalizable and interpretable framework for reasoning under uncertainty in safety-critical applications.

cs.AI↗

arXiv2025

Recursive Decomposition of Logical Thoughts: Framework for Superior Reasoning and Knowledge Propagation in Large Language Models

Enhancing the reasoning capabilities of Large Language Models remains a critical challenge in artificial intelligence. We introduce RDoLT, Recursive Decomposition of Logical Thought prompting, a novel framework that significantly boosts LLM reasoning performance. RDoLT is built on three key innovations: (1) recursively breaking down complex reasoning tasks into sub-tasks of progressive complexity; (2) employing an advanced selection and scoring mechanism to identify the most promising reasoning thoughts; and (3) integrating a knowledge propagation module that mimics human learning by keeping track of strong and weak thoughts for information propagation. Our approach was evaluated across multiple benchmarks, including GSM8K, SVAMP, MultiArith, LastLetterConcatenation, and Gaokao2023 Math. The results demonstrate that RDoLT consistently outperforms existing state-of-the-art techniques, achieving a 90.98 percent accuracy on GSM8K with ChatGPT-4, surpassing state-of-the-art techniques by 6.28 percent. Similar improvements were observed on other benchmarks, with accuracy gains ranging from 5.5 percent to 6.75 percent. These findings highlight RDoLT's potential to advance prompt engineering, offering a more effective and generalizable approach to complex reasoning tasks.

cs.CL↗