EPISODE · Jun 7, 2025 · 44 MIN
ProxyThinker: Guiding Large Models with Small Reasoners
from Neural intel Pod · host Neuralintel.org
This academic paper introduces PROXYTHINKER, a novel inference-time method designed to enhance the visual reasoning abilities of large vision-language models (LVLMs). Unlike computationally expensive fine-tuning approaches like reinforcement fine-tuning (RFT), PROXYTHINKER allows larger models to inherit reasoning skills from smaller, pre-trained reasoning models. It achieves this by adjusting the large model's output based on the difference between a small RFT expert's output and a small base model's output. The paper demonstrates that this training-free techniquesignificantly improves performance on various visual reasoning benchmarks while being computationally efficient.
What this episode covers
This academic paper introduces PROXYTHINKER, a novel inference-time method designed to enhance the visual reasoning abilities of large vision-language models (LVLMs). Unlike computationally expensive fine-tuning approaches like reinforcement fine-tuning (RFT), PROXYTHINKER allows larger models to inherit reasoning skills from smaller, pre-trained reasoning models. It achieves this by adjusting the large model's output based on the difference between a small RFT expert's output and a small base model's output. The paper demonstrates that this training-free techniquesignificantly improves performance on various visual reasoning benchmarks while being computationally efficient.
NOW PLAYING
ProxyThinker: Guiding Large Models with Small Reasoners
No transcript for this episode yet
Similar Episodes
Mar 14, 2026 ·23m
Mar 11, 2026 ·16m
Feb 28, 2026 ·14m