Pith. sign in

REVIEW 19 cited by

Rho-1: Not All Tokens Are What You Need

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

arxiv 2404.07965 v4 pith:KPASI7VI submitted 2024-04-11 cs.CL cs.AI

Rho-1: Not All Tokens Are What You Need

classification cs.CL cs.AI
keywords tokenslanguagemodelrho-1pretrainingtraininglosscontinual
verification ladder T0 review T1 audit T2 compute T3 formal T4 reserved
0 comments
Share X Bluesky LinkedIn Reddit HN
read the original abstract

Previous language model pre-training methods have uniformly applied a next-token prediction loss to all training tokens. Challenging this norm, we posit that "9l training". Our initial analysis examines token-level training dynamics of language model, revealing distinct loss patterns for different tokens. Leveraging these insights, we introduce a new language model called Rho-1. Unlike traditional LMs that learn to predict every next token in a corpus, Rho-1 employs Selective Language Modeling (SLM), which selectively trains on useful tokens that aligned with the desired distribution. This approach involves scoring pretraining tokens using a reference model, and then training the language model with a focused loss on tokens with higher scores. When continual pretraining on 15B OpenWebMath corpus, Rho-1 yields an absolute improvement in few-shot accuracy of up to 30% in 9 math tasks. After fine-tuning, Rho-1-1B and 7B achieved state-of-the-art results of 40.6% and 51.8% on MATH dataset, respectively - matching DeepSeekMath with only 3% of the pretraining tokens. Furthermore, when continual pretraining on 80B general tokens, Rho-1 achieves 6.8% average enhancement across 15 diverse tasks, increasing both efficiency and performance of the language model pre-training.

discussion (0)

Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.

Forward citations

Cited by 19 Pith papers

Reviewed papers in the Pith corpus that reference this work. Sorted by Pith novelty score.

  1. Bridging Compute- and Data-Optimal Pretraining

    cs.LG 2026-07 conditional novelty 7.0

    Pretraining loss obeys a single law in which repeated or paraphrased tokens count as η(N, data-per-parameter, expansion-ratio) fresh tokens, with total effective data saturating as derived tokens grow.

  2. CODEBLOCK: Learning to Supervise Code at the Right Granularity

    cs.LG 2026-06 unverdicted novelty 7.0

    CodeBlock partitions code responses into syntactically coherent blocks, scores them with generalized cross-entropy and data-flow signals, and applies sparse supervision to achieve higher pass@1 than full SFT using 1.9...

  3. Selective Contrastive Learning For Gloss Free Sign Language Translation

    cs.CL 2026-04 unverdicted novelty 7.0

    A pair selection strategy based on negative similarity dynamics strengthens contrastive supervision in gloss-free sign language translation by reducing noisy negatives.

  4. Understanding LoRA as Knowledge Memory: An Empirical Analysis

    cs.LG 2026-03 conditional novelty 7.0

    LoRA modules function as composable knowledge memories for LLMs with measurable storage capacity, internalization efficiency, and advantages in multi-module long-context reasoning.

  5. VocabTailor: Dynamic Vocabulary Selection for Downstream Tasks in Small Language Models

    cs.CL 2025-08 unverdicted novelty 7.0

    VocabTailor introduces a decoupled dynamic vocabulary selection framework that reduces vocabulary-related memory in SLMs by up to 99% with minimal task performance loss.

  6. Understanding LoRA as Knowledge Memory: An Empirical Analysis

    cs.LG 2026-03 conditional novelty 6.0

    LoRA modules are a complementary, finite-capacity parametric memory for LLMs: capacity grows with rank, small ranks are most parameter-efficient, synthetic QA data helps most, and practical multi-LoRA systems are bott...

  7. GradAlign: Gradient-Aligned Data Selection for LLM Reinforcement Learning

    cs.LG 2026-02 reject novelty 6.0

    GradAlign selects RL training problems whose policy gradients align with validation-set gradients, claiming more stable and better-performing LLM post-training.

  8. Don't Ignore the Tail: Decoupling top-K Probabilities for Efficient Language Model Distillation

    cs.CL 2026-02 unverdicted novelty 6.0

    A modified divergence decouples top-K teacher probabilities from the distribution tail during distillation, yielding competitive performance on decoder models with standard compute.

  9. Training-Trajectory-Aware Token Selection

    cs.CL 2026-01 unverdicted novelty 6.0

    Training-Trajectory-Aware Token Selection (T3S) reconstructs the token-level training objective to overcome a performance bottleneck in continual distillation of reasoning capabilities from large to small language models.

  10. LightReasoner: Can Small Language Models Teach Large Language Models Reasoning?

    cs.CL 2025-10 unverdicted novelty 6.0

    LightReasoner distills supervision signals from SLM-LLM behavioral divergence to improve LLM reasoning on math benchmarks with up to 28.1% accuracy gains and 90-99% reductions in resources.

  11. GIFT: Guided Importance-Aware Fine-Tuning for Diffusion Language Models

    cs.CL 2025-09 unverdicted novelty 6.0

    GIFT weights tokens by entropy during fine-tuning of diffusion language models and reports better performance than standard SFT on reasoning benchmarks across multiple settings.

  12. ReGATE: Learning Faster and Better with Fewer Tokens in MLLMs

    cs.CV 2025-07 unverdicted novelty 6.0

    ReGATE introduces a teacher-student adaptive token elision method that reduces training tokens to 38% while matching or exceeding baseline accuracy on multimodal benchmarks.

  13. Step-DPO: Step-wise Preference Optimization for Long-chain Reasoning of LLMs

    cs.LG 2024-06 conditional novelty 6.0

    Step-DPO performs preference optimization on individual reasoning steps rather than complete answers, producing nearly 3% accuracy gains on MATH for 70B+ parameter models with 10K preference pairs.

  14. DataComp-LM: In search of the next generation of training sets for language models

    cs.LG 2024-06 unverdicted novelty 6.0

    DCLM-Baseline dataset lets a 7B model reach 64% 5-shot MMLU accuracy after 2.6T tokens, beating prior open-data models by 6.6 points on MMLU with 40% less compute.

  15. DomainPilot: Domain-Level Loss-Guided Two-Stage Data Mixture Optimization for Efficient Language Model Fine-Tuning

    cs.LG 2026-07 conditional novelty 5.0

    Reallocating fine-tuning data proportions according to fitted per-domain loss curves improved benchmark scores by 1.8–3.8 points on Qwen3-1.7B, in a single unseeded run; the planned second stage is not yet implemented.

  16. RadarPLM: Adapting Pre-trained Language Models for Marine Radar Target Detection by Selective Fine-tuning

    eess.SP 2025-09 unverdicted novelty 5.0

    RadarPLM adapts PLMs for marine radar target detection with lightweight adaptation and selective fine-tuning based on online learning values, reporting at least 6.35% average detection gains in low SCR conditions.

  17. SPREG: Structured Plan Repair with Entropy-Guided Test-Time Intervention for Large Language Model Reasoning

    cs.AI 2026-04 unverdicted novelty 4.0

    SPREG detects logical failures in LLM long-chain reasoning through real-time entropy spikes and performs structured plan repairs using historical distributions, reporting a 20% absolute accuracy gain on AIME25.

  18. Masked Diffusion Language Models with Frequency-Informed Training

    cs.CL 2025-09 conditional novelty 4.0

    Masked diffusion language models trained on 100M words match a hybrid GPT-BERT baseline on BabyLM tests, with a rare-word-focused masking variant.

  19. Learning to Reason at the Frontier of Learnability

    cs.LG 2025-02 unverdicted novelty 4.0

    A curriculum sampling questions with high variance in success rate improves reinforcement learning performance for LLM reasoning tasks.