LLM Digest

Story

arxiv_cs_lg · Nov 2, 2026 · paper

Source brief

FADE: Factored Ability--Difficulty Filtering for Online RLVR Scheduling

arxiv.orgNov 2, 2026
original source linked

In brief

Group-based reinforcement learning with verifiable rewards (RLVR) learns only from prompts whose sampled completions disagree: all-pass and all-fail groups produce zero relative advantage and burn generation budget without moving the policy. Adaptive schedulers try to steer rollo

Read the original at arxiv.org →Open in live feed Read that day’s brief

FADE: Factored Ability--Difficulty Filtering for Online RLVR Scheduling

Earlier in this thread 4 items

PIVOT: Progressive Intervention via Online Teacher Correction

Causal Attention on CPU: A Cache-Aware and Load-Balanced Scheduling Kernel for Prefill

EnvSched: Environment-Aware Scheduling for Rollout-Efficient Agent Reinforcement Learning

CRISP: Co-designing Pruning and Scheduling for Efficient MoE Inference on Edge Servers