How multimodal language models should encode visual input remains up for debate. Most models use a pretrained vision encoder, while some recent recipes train the encoder from scratch with the language model or remove it entirely. Vision encoders bundle together three design choices: a visual prior from upstream pretrai...
Ellis Brown, Chris Hoang, Shengyi Qian et al.· 0 citations
Video reward models are trained and evaluated on human preferences between videos that often differ simultaneously on properties such as prompt alignment, visual fidelity, motion, and generator identity. As a result, aggregate accuracy does not reveal which video properties a reward model can reliably distinguish. We c...
S. Li, Jacqueline He, Jia-Hui Chen et al.· 0 citations
Across group-relative diffusion RL experiments, it is found that exploration is learnable, the shape of the exploration distribution outweighs its magnitude, and rollout quality is more effective than rollout quantity.
S. Li, Xiao-Chuang Han, Y. Tsvetkov et al.· arXiv.org· 0 citations
Verifiable Visual Rewards (VVR) is introduced, the first framework for programmatically verifiable image rewards, and it is shown that training on it generalizes to natural prompts, and gains extend to out-of-domain benchmarks.
S. Li, Xiao-Chuang Han, Y. Tsvetkov et al.· 1 citation
Anchored Decoding is proposed, a plug-and-play inference-time method for suppressing verbatim copying that enables decoding from any risky LM trained on mixed-license data by keeping generation in bounded proximity to a permissively trained safe LM.
Jacqueline He, J. Hayase, Wen-tau Yih et al.· arXiv.org· 0 citations
We use cookies to run the site and, with your consent, for analytics and to show ads.
See our Cookie Policy.