REVIEW 8 cited by
Phi-4-Mini-Reasoning: Exploring the Limits of Small Reasoning Language Models in Math
Not yet reviewed by Pith; the record is open.
This paper has not been read by Pith yet. Machine review is queued; the pith claim, tier, and objections will appear here once it completes.
SPECIMEN: schema-true, not a live event
T0 review · schema-true
One-sentence machine reading of the paper's core claim.
pith:XXXXXXXX · record.json · timestamp
Phi-4-Mini-Reasoning: Exploring the Limits of Small Reasoning Language Models in Math
read the original abstract
Chain-of-Thought (CoT) significantly enhances formal reasoning capabilities in Large Language Models (LLMs) by training them to explicitly generate intermediate reasoning steps. While LLMs readily benefit from such techniques, improving reasoning in Small Language Models (SLMs) remains challenging due to their limited model capacity. Recent work by Deepseek-R1 demonstrates that distillation from LLM-generated synthetic data can substantially improve the reasoning ability of SLM. However, the detailed modeling recipe is not disclosed. In this work, we present a systematic training recipe for SLMs that consists of four steps: (1) large-scale mid-training on diverse distilled long-CoT data, (2) supervised fine-tuning on high-quality long-CoT data, (3) Rollout DPO leveraging a carefully curated preference dataset, and (4) Reinforcement Learning (RL) with Verifiable Reward. We apply our method on Phi-4-Mini, a compact 3.8B-parameter model. The resulting Phi-4-Mini-Reasoning model exceeds, on math reasoning tasks, much larger reasoning models, e.g., outperforming DeepSeek-R1-Distill-Qwen-7B by 3.2 points and DeepSeek-R1-Distill-Llama-8B by 7.7 points on Math-500. Our results validate that a carefully designed training recipe, with large-scale high-quality CoT data, is effective to unlock strong reasoning capabilities even in resource-constrained small models.
Forward citations
Cited by 8 Pith papers
-
OPRD: On-Policy Representation Distillation
OPRD performs distillation in hidden-state space on on-policy data for deterministic gradients and better math benchmark performance, plus OPRD-Bridge for cross-architecture transfer via low-rank projectors.
-
DICE: Diffusion Large Language Models Excel at Generating CUDA Kernels
DICE, a family of diffusion LLMs trained with a bi-phase infilling-to-generation RL curriculum, achieves state-of-the-art CUDA kernel generation at data and token budgets far below prior specialized autoregressive models.
-
HintMR: Eliciting Stronger Mathematical Reasoning in Small Language Models
A cooperative system with one SLM distilling stepwise hints from a large model to guide another SLM's math reasoning yields consistent accuracy gains on benchmarks.
-
Reasoning Up the Instruction Ladder for Controllable Language Models
RLVR on ~7K verifiable system/user conflict examples teaches LLMs to prioritize higher-priority instructions, improving instruction-hierarchy and safety benchmarks.
-
Efficient Reasoning on the Edge
LoRA adapters, budget-forced GRPO, dynamic switching, parallel verification and FPTQuant enable practical chain-of-thought reasoning on quantized Qwen2.5-7B for edge devices.
-
DenseSteer: Steering Small Language Models towards Dense Math Reasoning
DenseSteer is an inference-time steering framework that improves small LLMs' accuracy on math reasoning by modulating representations toward dense reasoning patterns with fewer but higher-density steps.
-
Energy Use of AI Inference, Efficiency Pathways, and Test-Time Scaling
Bottom-up estimates give a median of 0.34 Wh per query for frontier LLMs on H100 hardware, rising 13x with test-time scaling, while combined efficiency gains could cut energy 8-20x.
-
Energy Use of AI Inference, Efficiency Pathways, and Test-Time Scaling
A bottom-up model of AI inference energy finds ~0.3 Wh per query on optimized H100 deployments, 13x more for long reasoning queries, and 8-20x potential efficiency gains.
discussion (0)
Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.