Unlocking Lossless Speedups in LLMs via Discrete Diffusion (5000 Tk/S)s-sahoo.com·3 pts·E-Reverance·0
Matching Features, Not Tokens: Energy-Based Fine-Tuning of Language Modelsenergy-based-fine-tuning.github.io·1 pts·E-Reverance·0
Inverse Rendering for Modeling with Line Primitives (Good for Fur)kenji-tojo.github.io·1 pts·E-Reverance·0
Unified Representation for Continuous-Latent Diffusion Language Modelingarxiv.org·1 pts·E-Reverance·1
Explorative Modeling: Unlocking a Third Pretraining Axis and E2E Generationexplorative-modeling.github.io·7 pts·E-Reverance·2