2.8m — Gmail.txt

: Increasing data from 2M to 2.8M results in no further performance gains, confirming the plateau [22]. Multimodal Structured Reinforcement Learning (MSRL) :

: Uses 22k data pairs focusing on textual accuracy ( 2.8M GMAIL.txt

To break the plateau, the authors implement a two-stage Reinforcement Learning (RL) process [11]. : Increasing data from 2M to 2

) used in the RL stages or the used to measure the success of the 2.8M dataset? : Uses 11k pairs with a balance of

The paper demonstrates that MSRL significantly outperforms pure SFT models by optimizing for both textual structure and visual fidelity, effectively surpassing the performance limit reached at 2.8M SFT samples [11, 25]. MSRL Stage Max Dataset Size 2.8 million samples [11, 22] 33k curated samples [11] GPU Requirement 16 H800 GPUs [11] 24 H800 GPUs [11] Training Goal Min. Negative Log-Likelihood [22] Hybrid Text-Visual Reward [11] Outcome Performance Plateaus [22] Breaks SFT Performance Limit [11]

) to ensure the generated code matches the visual intent [11].

: Uses 11k pairs with a balance of textual and visual rewards (