CPU-ARC enables efficient on-CPU LLM reasoning reuse
CPU-ARC builds agentic reasoning caches from LLM traces matching frontier models on math tasks at 16-291x lower cost.
Stories per day · 30 days
2 RSS
Relatedablations1actor-critic1agentic-ai1chain-of-thought1constrained-decoding1cpu-optimization1generalization1grpo1model-optimization1neural-networks1open-source-models1out-of-distribution1output-formats1reasoning-caching1reinforcement-learning1rlvr1weight-space-programming1
13 stories, newest first · one card per story
Stories tagged #llm-reasoning
RLVR models abandon reasoning strategies under OOD shifts
Co-training with bijective inversion tasks restores derivation and OOD accuracy.
Astra reads and writes neural network weights
Frontier LLM Astra identifies trained behaviors from raw weights and edits Qwen3.5-0.8B weights to improve math performance without gradients.
1
Constrained decoding effects depend on the constraint type
Across eleven open-source models, apparent losses mostly came from mismatched prompts and formats rather than decoder masking.
S3AC adds a stabilized actor-critic for LLM reasoning
The paper reports gains from 1.5B to 397B parameters and stable single-rollout training at 35B.
Semifactual Credit-Augmented Policy Optimization
SCAPO adds semifactual token stability to GRPO, lifting Qwen3 AIME accuracy by up to 5.63 points.
Overcoming Scaling Limits in On-Policy Self-Distillation for LLM Reasoning
OASIS keeps on-policy self-distillation effective at scale by supervising verified reasoning trajectories.
An RL View of OPD: Least Square Policy Distillation for Sample-Efficient LLM Reasoning
Least-Square Policy Distillation beats on-policy distillation baselines by 1.59 Avg@16 on math reasoning.
SAGE: Mitigating Long-Horizon Reasoning Biases via Topological Guidance
SAGE uses algebraic and hyperbolic guidance to reduce long-horizon LLM reasoning biases, with up to an 8-fold gain.
Think Thrice Before Reranking: Multi-perspective Evidence and Reasoning Integration for Text Reranking
MERIT-Rank integrates multiple LLM reasoning trajectories for text reranking; its 4B model outperforms most 7B and 32B rerankers on BRIGHT.
Bellman Policy Optimization
Bellman Policy Optimization, a critic-free RLVR method derived from Policy Mirror Descent, improves LLM mathematical reasoning without intermediate state-value estimation.
Bellman Policy Optimization
Bellman Policy Optimization is a critic-free RLVR method that reformulates Policy Mirror Descent as a trajectory-level objective for LLM reasoning training.
AI models' written reasoning steps correspond to distinct internal patterns, a new study finds
KAIST and Naver AI Lab researchers show LLM reasoning steps like extraction and computation map to distinct activation patterns, strongest in middle layers.
3