LongStraw: Long-Context RL Beyond 2M Tokens under a Fixed GPU Budget
:::info Stub — Full Engineering Breakdown Coming This paper was featured on Hugging Face Daily Papers on 2026-07-16 with 177 upvotes. A full breakdown with production viability rating, implementation notes, and honest limitations is being written. Subscribe to AI Letters → :::
| Authors | Changhai Zhou et al. |
| Year | 2026 |
| HF Upvotes | 177 |
| arXiv | 2607.14952 |
| Download | |
| HF Page | View on Hugging Face |
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. The gap is especially important for AI agents, whose observations, tool outputs, documents, and prior decisions accumulate over long trajectories. LongStraw is an architecture-aware execution stack for million-token RL post-training under a fixed GPU budget, instantiated with Group Relative Policy Optimization (GRPO). It evaluates the shared prompt without autograd, retains only model-specific state needed by later tokens, and replays short response branches one at a time, reducing the live training graph at the cost of additional replay time. We implement it for the hybrid recurrent and full-attention Qwen3.6-27B and the compressed-attention mixture-of-experts GLM-5.2. On eight H20 GPUs, LongStraw completes grouped Qwen scoring and response backward at 2.1M positions for groups of 2 and 8; increasing the group size adds only 0.21 GB of peak allocated memory, while a separate stress test reaches 4.46M positions. On 32 H20 GPUs, we validate the end-to-end LongStraw execution path for a 2.1M-token prompt across all 78 layers of GLM-5.2. These experiments establish execution capacity rather than complete training correctness because the captured prompt state is detached and some distributed forward and gradient composition paths remain incomplete.
Engineering Breakdown
The Problem
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. The gap is especially important for AI agents, whose observations, tool outputs, documents, and prior decisions accumulate over long trajectories.
The Approach
The gap is especially important for AI agents, whose observations, tool outputs, documents, and prior decisions accumulate over long trajectories.
Key Results
These experiments establish execution capacity rather than complete training correctness because the captured prompt state is detached and some distributed forward and gradient composition paths remain incomplete.
Research Areas
This paper contributes to the following areas of AI/ML engineering:
- Machine learning
- Deep learning
- Neural networks
- Model optimization
- AI systems
- Longstraw
:::tip Subscribe Get weekly breakdowns of papers like this in AI Letters - the newsletter for engineers building production AI systems. :::