transformers

33d1bc04 - Add native masked MSE loss for Sapiens2ForPoseEstimation (#46764)

Add native masked MSE loss for Sapiens2ForPoseEstimation (#46764) * Add native masked MSE loss for Sapiens2ForPoseEstimation (#46518) * Fix trailing whitespace in test file * Add supervised fine-tuning documentation for Sapiens2 pose estimation * Fix method binding bug on _loss_function by wrapping in staticmethod * Switch test to floats_tensor and fix docs heatmap dimensions * Simplify loss function by relying on upstream weight expansion * Update dummy label_weights shape in tests to match heatmaps * Centralize loss function and test backward pass * don't use loss fn for now and just do it directly --------- Co-authored-by: vasqu <antonprogamer@gmail.com>