Compiling Deterministic Structure into SLM Harnesses

• URL: http://arxiv.org/abs/2604.17450v1
• Date: Sun, 19 Apr 2026 14:04:29 GMT
• Title: Compiling Deterministic Structure into SLM Harnesses
• Abstract summary: Small language models (SLMs) are costly and face data sovereignty limits for high-volume use.<n>We propose Semantic Gradient Descent (SGDe), a teacher-student framework that compiles agentic improvement into discrete execution plans.
• Score: 0.6117371161379209
• License: http://creativecommons.org/licenses/by/4.0/
• Abstract: Enterprise deployment of small language models (SLMs) is constrained by epistemic asymmetry: SLMs cannot self-correct reasoning errors, while frontier LLMs are prohibitively costly and face data sovereignty limits for high-volume use. We propose Semantic Gradient Descent (SGDe), a teacher-student framework that compiles agentic workflows into discrete execution plans comprising DAG topologies, system prompts, and deterministic executable code. The trailing "e" distinguishes SGDe from stochastic gradient descent. SGDe operates in a discrete semantic space where a frontier teacher generates natural-language critiques acting as directional gradients to iteratively refine the SLM's workflow artefacts. We formalise SGDe within a PAC learning framework, establishing sample-complexity bounds that enable convergence with as few as three training examples on targeted synthetic tasks by leveraging the teacher as a statistical prior. On a GSM-Hard-derived test set built via adversarial synthesis, compiled workflows reach 91.3% accuracy at m=5 and 99.3% at m=3 within the small-m regime motivated by Corollary 1, a +26.3% to +34.3% absolute improvement over state-of-the-art prompt optimisers. In the emerging paradigm of harness engineering, SGDe treats placement of deterministic code (which subtasks to delegate to a Python runtime versus retain as LLM calls) as a trace-driven, per-node optimisation target, generalising the whole-problem offloading of PAL and PoT. The teacher compiles two complementary deterministic structures: capability offloading, which delegates subtasks to Python when the SLM cannot execute them reliably, and structural consensus, which wraps variance-limited reasoning steps in fan-out/fan-in subgraphs aggregated by deterministic voting.

Related papers

• The Imitation Game: When LLMs Learn to Reason Like Programs via Code-Centric Reasoning Data Synthesis [50.473217152006875]
  MIMIC is a framework that leverages executable code as a rigorous medium for reasoning data synthesis.<n>MIMIC transforms algorithms into verifiable reasoning trajectories through narrative fusion, code-guided test synthesis, and dynamic code instrumentation.<n>Our method significantly elevates accuracy across general reasoning, complex mathematical benchmarks, and fine-grained deterministic tasks.
  arXiv  Detail & Related papers  (2026-09-13T17:04:16Z)
• Prompt Codebooks: Discrete Compositional Optimization for Language Model Instruction Refinement [10.441136483320205]
  We propose Prompt Codebooks, a novel compositional prompt optimization framework.<n>It recasts APO as discrete learning over a finite vocabulary of natural-language instincts.<n>It improves over zero-shot by up to +30.36 points, surpasses the strongest prior baseline (GEPA) by +3.34 on HotpotQA and +1.11 in aggregate.
  arXiv  Detail & Related papers  (2026-05-27T11:57:12Z)
• PRISM: : Planning and Reasoning with Intent in Simulated Embodied Environments [59.07829883257003]
  Built on five multi-room apartments, PRISM structures 300 human-verified tasks into three capability tiers.<n>PRISM exposes an agent-agnostic executable action API that allows arbitrary agents to be evaluated end-to-end.
  arXiv  Detail & Related papers  (2026-05-12T04:59:47Z)
• $\mathcal{S}^2$IT: Stepwise Syntax Integration Tuning for Large Language Models in Aspect Sentiment Quad Prediction [51.165432266846096]
  syntactic structure information is underutilized in the generative paradigm of large language models (LLMs)<n>S2IT progressively integrates syntactic structure knowledge into LLMs through a multi-step tuning process.<n> Experiments demonstrate that S2IT significantly improves state-of-the-art performance across multiple datasets.
  arXiv  Detail & Related papers  (2026-04-25T13:24:29Z)
• From Agent Loops to Structured Graphs:A Scheduler-Theoretic Framework for LLM Agent Execution [1.8222732878503212]
  The dominant paradigm for building LLM based agents is the Agent Loop, an iterative cycle where a single language model decides what to do next by reading an ever growing context window.<n>This perspective places Agent Loops and graph based execution engines on a single semantic continuum.<n>We propose SGH, which lifts control flow from implicit context into an explicit static DAG.
  arXiv  Detail & Related papers  (2026-04-13T12:16:45Z)
• Training LLMs for Multi-Step Tool Orchestration with Constrained Data Synthesis and Graduated Rewards [76.49428173793386]
  LLMs must invoke multiple dependent APIs in the correct order while propagating intermediate outputs.<n>Training models to handle such faces two obstacles: existing environments focus on simple per-turn function calls with simulated data, and binary rewards.<n>We present a framework addressing both challenges. First, we construct a reinforcement learning environment backed by a large-scale cache of real API responses, enabling a data synthesis pipeline that samples valid multi-step orchestration traces.<n>Second, we propose a graduated reward design that decomposes correctness into atomic validity and orchestration.
  arXiv  Detail & Related papers  (2026-03-25T18:31:39Z)
• LOGIGEN: Logic-Driven Generation of Verifiable Agentic Tasks [4.6880826836662814]
  We introduce textbfLOGIGEN, a logic-driven framework that synthesizes verifiable training data.<n>On $2$-Bench, LOGIGEN-32B(RL) achieves a textbf79.5% success rate, substantially outperforming the base model.
  arXiv  Detail & Related papers  (2026-02-28T08:35:30Z)
• Prism: Efficient Test-Time Scaling via Hierarchical Search and Self-Verification for Discrete Diffusion Language Models [96.0074341403456]
  Inference-time compute has re-emerged as a practical way to improve LLM reasoning.<n>Most test-time scaling (TTS) algorithms rely on autoregressive decoding.<n>We propose Prism, an efficient TTS framework for dLLMs.
  arXiv  Detail & Related papers  (2026-02-02T09:14:51Z)
• Improving LLM Reasoning with Homophily-aware Structural and Semantic Text-Attributed Graph Compression [55.51959317490934]
  Large language models (LLMs) have demonstrated promising capabilities in Text-Attributed Graph (TAG) understanding.<n>We argue that graphs inherently contain rich structural and semantic information, and that their effective exploitation can unlock potential gains in LLMs reasoning performance.<n>We propose Homophily-aware Structural and Semantic Compression for LLMs (HS2C), a framework centered on exploiting graph homophily.
  arXiv  Detail & Related papers  (2026-01-13T03:35:18Z)
• Managing the Stochastic: Foundations of Learning in Neuro-Symbolic Systems for Software Engineering [0.27195102129094995]
  Current approaches to AI coding agents blur the lines between the Large Language Model and the agent itself.<n>This paper proposes setting the control boundary such that the LLM is treated as a component of the environment environment.
  arXiv  Detail & Related papers  (2025-12-18T15:28:21Z)
• Zero-shot 3D Map Generation with LLM Agents: A Dual-Agent Architecture for Procedural Content Generation [8.398818816613806]
  We propose a training-free architecture that utilizes LLM agents for zero-shot PCG parameter configuration.<n>Our system pairs an Actor agent with a Critic agent, enabling an iterative workflow where the system autonomously reasons over tool parameters.
  arXiv  Detail & Related papers  (2025-12-11T10:22:02Z)
• ReflecSched: Solving Dynamic Flexible Job-Shop Scheduling via LLM-Powered Hierarchical Reflection [4.101501114944147]
  ReflecSched is a framework that empowers the LLM beyond a direct scheduler.<n>It distills simulations across multiple planning horizons into a concise, natural-language summary.<n>This summary is then integrated into the prompt of a final decision-making module, guiding it to produce non-myopic actions.
  arXiv  Detail & Related papers  (2025-08-03T11:26:35Z)
• SatLM: Satisfiability-Aided Language Models Using Declarative Prompting [68.40726892904286]
  We propose a new satisfiability-aided language modeling (SatLM) approach for improving the reasoning capabilities of large language models (LLMs) We use an LLM to generate a declarative task specification rather than an imperative program and leverage an off-the-shelf automated theorem prover to derive the final answer. We evaluate SATLM on 8 different datasets and show that it consistently outperforms program-aided LMs in the imperative paradigm.
  arXiv  Detail & Related papers  (2023-05-16T17:55:51Z)
• How Fine-Tuning Allows for Effective Meta-Learning [50.17896588738377]
  We present a theoretical framework for analyzing representations derived from a MAML-like algorithm. We provide risk bounds on the best predictor found by fine-tuning via gradient descent, demonstrating that the algorithm can provably leverage the shared structure. This separation result underscores the benefit of fine-tuning-based methods, such as MAML, over methods with "frozen representation" objectives in few-shot learning.
  arXiv  Detail & Related papers  (2021-05-05T17:56:00Z)

This list is automatically generated from the titles and abstracts of the papers in this site.

This site does not guarantee the quality of this site (including all information) and is not responsible for any consequences.