TRM-Planner: Offline Target Planning and Distillation for Tiny Recursive Models (2026.findings-acl)
Copied to clipboard
| Challenge: | Tiny Recursive Models (TRMs) perform iterative reasoning with an Adaptive Computation Time (ACT)-style loop, but their supervised training targets can be brittle and their halting behavior difficult to tune. |
| Approach: | They propose a two-stage teacher-cache distillation recipe that shifts compute to offline teacher-caching stage. |
| Outcome: | The proposed model improves su-pervision while leaving student-time inference unchanged. |
Similar Papers
Can Compact Language Models Search Like Agents? Distillation-Guided Policy Optimization for Preserving Agentic RAG Capabilities (2026.acl-long)
Copied to clipboard
| Challenge: | Existing agentic RAG systems rely on large language models with billions of parameters. |
| Approach: | They propose a method to elicit agentic RAG behaviors from compact models . they propose ARC, which uses cold-start initialization and teacher guidance . |
| Outcome: | The proposed method outperforms the larger teacher model in some cases. |
Fast and Effective On-Policy Distillation from Reasoning Prefixes (2026.findings-acl)
Copied to clipboard
Dongxu Zhang, Zhichao Yang, Sepehr Janghorbani, Jun Han, Andrew Ressler II, Qian Qian, Gregory D Lyng, Sanjit Singh Batra, Robert E. Tillman
| Challenge: | On-policy distillation (OPD) requires expensive on-the-fly sampling of the student policy during training, which substantially increases training cost. |
| Approach: | They propose to use on-policy distillation to sample trajectories from student model . they propose to terminate the sampling early during distillation . |
| Outcome: | The proposed method matches the performance of full OPD in long reasoning outputs while reducing training FLOP by 2x–40x. |
Marco-o1 v2: Towards Widening The Distillation Bottleneck for Reasoning Models (2025.acl-long)
Copied to clipboard
Huifeng Yin, Yu Zhao, Minghao Wu, Xuanfan Ni, Bo Zeng, Huaiyu.wh Huaiyu.wh, Tianqi Shi, Liangying Shao, Chenyang Lyu, Longyue Wang, Weihua Luo, Kaifu Zhang
| Challenge: | Recent efforts to distill large reasoning models into smaller lightweight models have shown competitive performances. |
| Approach: | They propose to distill long Chain-of-Thought data to improve SFT and RL methods by constructing data from scratch using Monte Carlo Tree Search. |
| Outcome: | The proposed method significantly improves reasoning performance on various benchmarks such as math (GSM8K, MATH, AIME). |
Prompt-Level Distillation: A Non-Parametric Alternative to Model Fine-Tuning for Efficient Reasoning (2026.acl-industry)
Copied to clipboard
| Challenge: | Advanced reasoning typically requires Chain-of-Thought prompting, which is accurate but incurs prohibitive latency and substantial test-time inference costs. |
| Approach: | They propose to extract explicit reasoning patterns from a Teacher model and organize them into a structured list of expressive instructions for the Student model’s System Prompt. |
| Outcome: | Evaluated using Gemma-3 4B, the proposed model improves Macro F1 scores on StereoSet and Contract-NLI while increasing LogiQA accuracy to 70%. |
Distilling Instruction-following Abilities of Large Language Models with Task-aware Curriculum Planning (2024.findings-emnlp)
Copied to clipboard
| Challenge: | Instruction tuning aims to align large language models (LLMs) with open-domain instructions and human-preferred responses. |
| Approach: | They propose a multi-round distillation framework that uses an oracle LLM to select instructions that are difficult for a student LLM. |
| Outcome: | The proposed framework outperforms large language models and user-tuned models on several widely recognized benchmarks and multiple student LLMs. |
Harnessing Negative Signals: Reinforcement Distillation from Teacher Data for LLM Reasoning (2026.acl-long)
Copied to clipboard
| Challenge: | Recent advances in model distillation show that data from advanced reasoning models can effectively train smaller student models. |
| Approach: | They propose a method to use both positive and negative distilled reasoning traces to maximize LLM reasoning performance in offline settings. |
| Outcome: | The proposed model outperforms existing methods in the distillation context. |
SmartAD: Capacity-Aligned Agent Distillation for Small Language Models (2026.findings-acl)
Copied to clipboard
| Challenge: | Large language models (LLMs) show strong reasoning and decision-making ability, but their high inference cost motivates transferring agentic skills to small language models. |
| Approach: | They propose a capacity-aligned agent distillation framework that trains SLMs on full reason–act–observe trajectories from a tool-using teacher. |
| Outcome: | The proposed framework outperforms all baselines on multi-hop QA and math benchmarks with 1.5B and 3B models. |
Faster MoE LLM Inference for Extremely Large Models (2026.findings-acl)
Copied to clipboard
| Challenge: | Existing inference optimizations for coarse-grained Mixture-of-Experts models implicitly assume a fixed activation budget, which is poorly understood. |
| Approach: | They propose a training-free policy that adapts token-level activation using router confidence and entropy while remaining within the model’s original budget. |
| Outcome: | The proposed skipping policy can provide substantial throughput gains, but optimal static schedules vary significantly across models and routing mechanisms. |
Staged Knowledge Distillation Through Least-to-Most Prompting: Optimizing Teacher Guidance via Difficulty-Aware Training (2025.findings-emnlp)
Copied to clipboard
| Challenge: | Knowledge distillation (KD) enables the compression of large language models (LLMs) conventional methods suffer from training-inference mismatches and suboptimal performance due to expensive student-generated outputs. |
| Approach: | They propose a method that combines a CL strategy and adaptive loss design to reduce training mismatches and suboptimal performance. |
| Outcome: | L2M-KD outperforms existing white-box KD methods on instruction-following tasks . it outperformed existing methods, achieving superior student model performance with reduced overhead . |
Teaching-Assistant-in-the-Loop: Improving Knowledge Distillation from Imperfect Teacher Models in Low-Budget Scenarios (2024.findings-acl)
Copied to clipboard
| Challenge: | Large language models (LLMs) have demonstrated state-of-the-art (SOTA) performance across a wide spectrum of tasks. |
| Approach: | They propose a framework that leverages three signal types to improve efficiency within resource-constrained, imperfect teacher scenarios. |
| Outcome: | The proposed framework improves on four complex reasoning tasks by 20.79% compared to fine-tuning without any signals across datasets. |