KernelBench, two TritonBenches, and the denominator behind your scoreSix public benchmark projects, pinned source revisions, and a four-task example showing how correctness and missing results change the score.4 min read·nodus-compute.hashnode.dev20m ago
Test weight and bias separately: a normalization ablationA fixed CPU ablation shows why varying weights alone or biases alone leaves blind spots. Includes all 504 observations and reproduction files.4 min read·nodus-compute.hashnode.dev2h ago
Keeping training state when the GPU node changesA controlled Nodus node replacement preserved a checkpoint and advanced training from step 8 to 9. Evidence, timing boundaries and limitations.6 min read·nodus-compute.hashnode.dev1d ago
Six wrong normalization formulas passed a correctness checkSix incorrect normalization formulas passed with identity inputs and failed with varied parameters in separate synthetic CPU and T4 CUDA runs.9 min read·nodus-compute.hashnode.dev2d ago
The reward is part of the datasetSix wrong normalization formulas passed a CPU checker with identity inputs. What this means for kernel RL feedback and held-out evaluation.3 min read·nodus-compute.hashnode.dev2d ago
