REVIEW 12 cited by
Orca-Math: Unlocking the potential of SLMs in Grade School 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
read the original abstract
Mathematical word problem-solving has long been recognized as a complex task for small language models (SLMs). A recent study hypothesized that the smallest model size, needed to achieve over 80% accuracy on the GSM8K benchmark, is 34 billion parameters. To reach this level of performance with smaller models, researcher often train SLMs to generate Python code or use tools to help avoid calculation errors. Additionally, they employ ensembling, where outputs of up to 100 model runs are combined to arrive at a more accurate result. Result selection is done using consensus, majority vote or a separate a verifier model used in conjunction with the SLM. Ensembling provides a substantial boost in accuracy but at a significant cost increase with multiple calls to the model (e.g., Phi-GSM uses top-48 to boost the performance from 68.2 to 81.5). In this work, we present Orca-Math, a 7-billion-parameter SLM based on the Mistral-7B, which achieves 86.81% on GSM8k without the need for multiple model calls or the use of verifiers, code execution or any other external tools. Our approach has the following key elements: (1) A high quality synthetic dataset of 200K math problems created using a multi-agent setup where agents collaborate to create the data, (2) An iterative learning techniques that enables the SLM to practice solving problems, receive feedback on its solutions and learn from preference pairs incorporating the SLM solutions and the feedback. When trained with Supervised Fine-Tuning alone, Orca-Math achieves 81.50% on GSM8k pass@1 metric. With iterative preference learning, Orca-Math achieves 86.81% pass@1. Orca-Math surpasses the performance of significantly larger models such as LLAMA-2-70B, WizardMath-70B, Gemini-Pro, ChatGPT-3.5. It also significantly outperforms other smaller models while using much smaller data (hundreds of thousands vs. millions of problems).
Forward citations
Cited by 12 Pith papers
-
Multi-Turn On-Policy Distillation with Prefix Replay
ReOPD offline-distills multi-turn agentic LLMs via teacher-prefix replay plus step-decay sampling, matching online OPD accuracy at ≥4× speed with zero tool calls.
-
StepHint: Multi-level Stepwise Hints Enhance Reinforcement Learning to Reason
StepHint improves RLVR math reasoning by giving the model multiple prefix-level hints drawn from correct chains generated by stronger models, beating several RLVR baselines on six math benchmarks and two out-of-domain sets.
-
Thunder-LLM: Efficiently Adapting LLMs to Korean with Minimal Resources
A cost-effective recipe consisting of tokenizer extension, continual pretraining, FP8 training, and SFT/DPO post-training yields Korean-English bilingual 8B models with top Korean benchmark scores.
-
Excessive Reasoning Attack on Reasoning LLMs
A 10-token adversarial suffix, optimized with three custom losses, raises reasoning length 3x to 9x on open reasoning LLMs without hurting accuracy, and partly transfers to closed models.
-
Reinforce LLM Reasoning through Multi-Agent Reflection
DPSDP trains an actor-critic LLM pair with DPO-style preference learning on self-generated trajectories, improving iterative refinement accuracy on math benchmarks.
-
Self-Route: Automatic Mode Switching via Capability Estimation for Efficient Reasoning
Self-Route uses hidden-layer representations from a brief pre-inference plan to route each question to either short or long chain-of-thought, cutting tokens by 30-55% with under 2% accuracy loss.
-
EasyMath: A 0-shot Math Benchmark for SLMs
EasyMath, a new 0-shot math benchmark for small language models, shows accuracy rising with model size and training, modest chain-of-thought gains, and better consistency at larger scale.
-
RTTC: Reward-Guided Collaborative Test-Time Compute
RTTC uses a reward model to route each query to direct inference, retrieval-augmented generation, or test-time training, plus caching to cut redundant work.
-
Boosting LLM Reasoning via Spontaneous Self-Correction
SPOC trains LLMs to interleave self-verification and solution attempts in a single pass, reporting gains on math benchmarks, though most gains come from stronger first attempts.
-
MDPO: Multi-Granularity Direct Preference Optimization for Mathematical Reasoning
MDPO applies a SimPO-style length-normalized reward to preference pairs built at solution, inference, and step granularities, yielding small accuracy gains on math benchmarks.
-
HOFT: Householder Orthogonal Fine-tuning
HOFT and SHOFT fine-tune foundation models with two Householder-built orthogonal matrices, matching or beating LoRA, DoRA, OFT, BOFT and HRA on reasoning, translation, image generation and math.
-
MASTER: Enhancing Large Language Model via Multi-Agent Simulated Teaching
A multi-agent simulated teaching pipeline creates BOOST-QA, and fine-tuning on it lifts reported LLM benchmark scores by up to 31 points over the original data.
Discussion (0). Sign in to comment.