HARTS (Hybrid‑Attention RL over Tree Structures) is a new system that jointly plans microbatches, data‑parallel replica assignments, and microbatch‑slot schedules to efficiently train agentic reinforcement learning models with hybrid attention over arbitrary rollout trees. It uses prefix compression and a linear‑time algorithm for chunkwise linear attention to avoid recomputing shared prefixes, enabling activation recomputation and bounded state replay while preserving trajectory‑wise training benefits. In experiments on an Agentic RL workload derived from SWE‑bench tasks, HARTS delivers 4.81–4.87× speedups in forward, backward, and gradient computations across multiple parallel configurations, with numerical differences comparable to baseline self‑rerun variation and a similar reward trend over the first 120 training steps.
arXiv:2608.28158v2 Announce Type: replace Abstract: Agentic reinforcement learning (RL) generates irregular rollout trees, but trajectory-by-trajectory training recomputes their shared interaction hi...
arXiv:2605. 15422v3 Announce Type: replace Abstract: Modern RL post-training methods such as GRPO and DAPO train on N response sequences of R tokens sampled from a shared prompt of P tokens, but standard FlashAttention replicates all P prompt tokens N times across both forward and backward passes -- duplicating compute and memory on identical hidden states.
arXiv:2608.30310v1 Announce Type: cross Abstract: Hybrid large language models interleave full-attention layers with linear-attention layers to reduce the cost of long-context inference. This structu...
arXiv:2608. 16798v1 Announce Type: cross Abstract: Agent harnesses have substantially improved performance on long-horizon tasks by coordinating agent interactions with the environment.
arXiv:2602. 00482v2 Announce Type: replace Abstract: Reinforcement learning (RL)-based post-training for large language models (LLMs) is computationally expensive, as it generates many rollout sequences that frequently share long token prefixes.
arXiv:2606. 03077v1 Announce Type: cross Abstract: Reinforcement learning (RL) has become a standard post-training paradigm for large language models (LLMs), extending beyond preference alignment to complex reasoning and multi-turn agentic behaviors.
ReFlow is a system‑algorithm co‑designed framework for efficient reinforcement learning (RL) post‑training of large language models (LLMs). It introduces replay‑driven decoupling, allowing recent trajectories to be reused multiple times to amortize rollout costs while limiting policy staleness. The architecture features a dual replay cycle, an elastic resource pool, and an adaptive scheduler, achieving up to 2.06× higher training throughput compared to existing synchronous and asynchronous RL systems without sacrificing model performance.
FlashNext‑RL is a model‑specific reinforcement learning training framework designed for the Qwen3.8‑Flash‑Next language model. It optimizes BF16 training of the full 48‑layer model on a single 8×B200 node by freezing early layers, offloading host memory, and tuning kernels for mixture‑of‑experts, sparse attention, and gated residual streams. The optimizations raise non‑padding training throughput from ~395 tokens/s/GPU to 4,438 tokens/s/GPU and enable a live software‑engineering RL run with 1,024 rollouts and a 131,072‑token context, completing iterations in about 22.4 minutes.
arXiv:2607. 14171v1 Announce Type: new Abstract: Reinforcement learning has emerged as the dominant paradigm for training large language model (LLM) agents that interact with executable sandboxes.
The paper introduces Self‑Sparse Speculative Rollouts (S$^3$R), a method that uses sparse attention to propose tokens and then verifies them with full attention, thereby reducing KV‑cache traffic while preserving the dense‑policy distribution for reinforcement learning rollouts. S$^3$R eliminates the need for a separate drafter by sharing policy parameters between drafting and verification, and a KV‑aware scheduler limits cache thrashing to maintain parallelism. Experiments on long‑context mathematical reasoning with the Qwen3 model show that S$^3$R achieves 1.35–1.39× higher end‑to‑end training throughput with comparable task performance.
arXiv:2607. 17299v1 Announce Type: cross Abstract: Long-horizon rollout generation has become the dominant systems bottleneck in agentic reinforcement learning (RL).
arXiv:2607. 14952v1 Announce Type: new Abstract: A growing gap separates inference context lengths from RL post-training: inference systems are approaching million-token contexts, while post-training workloads often remain at 256K tokens or below and rely on length generalization at deployment.