Tag
The paper analyzes why most RL post-training methods for language models never make it into open-source frameworks, tracing seven frameworks against 6,939 arXiv papers and finding only 37 of 1,598 proposed methods were adopted, while releasing a tracker dataset linking training failures to framework implementations.
The paper presents the first experiments on isolated sign language recognition for Icelandic Sign Language using a low-resource dataset, showing that cross-lingual transfer from other sign languages significantly improves performance.
An in-depth analysis of ChatGPT Dreaming V3's memory architecture, explaining how it synthesizes a coherent memory state from raw sources and comparing it to other open-source memory frameworks like mem0, supermemory, and Letta.