Field Notes · Opening

Beyond One-Size-Fits-All: Sample-Adaptive Strategy Routing for Vision Token Pruning in MLLMs

This paper proposes sample-adaptive strategy routing for vision token pruning in multimodal large language models, enabling different pruning strategies to be applied to different inputs rather than using a fixed strategy across all inputs.

Show HN: Hydra – Open-source agentic terminal with a PTY daemon

Hydra is an open-source terminal application that uses a PTY daemon to enable agentic capabilities.

Training a 3.8B LLM to 0.384 CORE for $998 – Hugo Vergnes

This article describes training a 3.8 billion parameter language model to achieve a 0.384 CORE score for approximately $998.