DPO

Timeline

1. Oct 8, 2026 · Open-source release · 1 source
   huggingface/trl v1.15.0
   SFT, DPO, KTO, GRPO, RLOO and Distillation now score tokens with a fused LM head: a Triton kernel projects the hidden states through the LM head in tiles and reduces to per-token log-probs and entropy directly, so the [batch, seq, vocab] logits tensor is never built.
2. Oct 8, 2026 · Research paper · 1 source
   EgoVoice: Proactive Spoken Assistance from Egocentric Multimodal Streams
   We introduce EgoVoice, a framework for training and evaluating proactive egocentric spoken assistants.
3. Oct 7, 2026 · Research paper · 1 source
   Enabling Preference-driven Unlearning in Few-step Distilled Text-to-Image Diffusion Models
   Text-to-image diffusion models are increasingly distilled into few-step variants and being deployed to enable fast inference.
4. Oct 7, 2026 · Research paper · 1 source
   From Digital Human Interactions to Physics-Based Humanoid Skills: Physics-Grounded Post-Training of Interaction Generators
   In this paper, we introduce DIGHT, a co-adaptive framework that couples a Digital human Interaction Generator with a Humanoid Tracking policy.
5. Oct 7, 2026 · Research paper · 1 source
   How to train your model organism
   We re-visit two publicly released organism suites using this validation framework and show that (1) chat quality and CoT naturalness degrade substantially across training recipes, and (2) validation metrics predict how well interpretability methods recover the installed behavior, e.g., a logit lens readout covaries with an organism's general capabilities.
6. Oct 6, 2026 · Research paper · 1 source
   CM-DPO: Constraint-Margin Direct Preference Optimization for LLM Planning
   We introduce Constraint-Margin DPO (CM-DPO), which replaces DPO's binary preference signal with a continuous margin derived from a deterministic symbolic verifier and scaled by violation severity.
7. Oct 6, 2026 · Research paper · 1 source
   AutodidactWAM: Cross-Modal Self-Distillation from Generated Video to Robot Actions
   We propose AutodidactWAM, a hand-pose estimator trained without teleoperation, followed by inverse kinematics, that runs on the model's generated video to recover action estimates.

Often appears with