Unverified50% confidenceSolutionExact time
The core idea of SPLIT-RL is to separate the training of visual reasoning and language reasoning into two non-overlapping stages.
1
Sources
50%
Confidence
Long-term
Relevance
10/9/2026
First Seen
Sources
Related Entities
Related Claims
UnverifiedThe policy trained by SPLIT-RL is used during inference in the same way as an ordinary GRPO model, requiring only a single CoT call and adding no inference cost.72% similarVerifiedEach RL training step requires running inference (generating rollouts) and backpropagation (updating weights) simultaneously, making the compute requirements far higher than ordinary fine-tuning.71% similarUnverifiedo1、DeepSeek-R1等模型将思维链内化为训练目标，在输出答案前生成大量内部草稿推理70% similarUnverifiedModern reasoning models generally use RLHF or its variant RLAIF for alignment training, where the model learns during training when to stop reasoning and when to output an answer.70% similarUnverifiedThe researchers propose a simple, training-free framework that enables reasoning-capable vision-language models to perform iterative search, dynamic reasoning, and evidence accumulation.69% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/990747API
curl https://kongchang.com/api/v1/knowledge/claims/990747MCP
get_claim(id=990747)