---
title: "Viswaretas Kotra (@viswa-kotra) | Hashnode"
description: "Founder @ Nodus Compute, previously in HFT, Distributed Systems, and AI. Read the latest articles by Viswaretas Kotra on Hashnode."
url: "https://hashnode.com/@viswa-kotra"
---

@viswa-kotra

# Viswaretas Kotra

@viswa-kotra · San Francisco, USA · Joined October 2026

Founder @ Nodus Compute, previously in HFT, Distributed Systems, and AI.

### [KernelBench, two TritonBenches, and the denominator behind your score](https://nodus-compute.hashnode.dev/kernelbench-two-tritonbenches-and-the-denominator-behind-your-score?utm_source=hashnode&utm_medium=feed)

Six 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.dev](https://nodus-compute.hashnode.dev/kernelbench-two-tritonbenches-and-the-denominator-behind-your-score?utm_source=hashnode&utm_medium=feed)

[30m ago](https://hashnode.com/posts/kernelbench-two-tritonbenches-and-the-denominator-behind-your-score/6af50381369f8264510be86f)

### [Test weight and bias separately: a normalization ablation](https://nodus-compute.hashnode.dev/test-weight-and-bias-separately-a-normalization-ablation?utm_source=hashnode&utm_medium=feed)

A 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.dev](https://nodus-compute.hashnode.dev/test-weight-and-bias-separately-a-normalization-ablation?utm_source=hashnode&utm_medium=feed)

[3h ago](https://hashnode.com/posts/test-weight-and-bias-separately-a-normalization-ablation/6af4df4f68250c91f86d3fe5)

### [Keeping training state when the GPU node changes](https://nodus-compute.hashnode.dev/keeping-training-state-when-the-gpu-node-changes?utm_source=hashnode&utm_medium=feed)

A 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.dev](https://nodus-compute.hashnode.dev/keeping-training-state-when-the-gpu-node-changes?utm_source=hashnode&utm_medium=feed)

[1d ago](https://hashnode.com/posts/keeping-training-state-when-the-gpu-node-changes/6af36ff0293fffa36e514a86)

### [Six wrong normalization formulas passed a correctness check](https://nodus-compute.hashnode.dev/six-wrong-normalization-formulas-passed-a-correctness-check?utm_source=hashnode&utm_medium=feed)

Six 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.dev](https://nodus-compute.hashnode.dev/six-wrong-normalization-formulas-passed-a-correctness-check?utm_source=hashnode&utm_medium=feed)

[2d ago](https://hashnode.com/posts/six-wrong-normalization-formulas-passed-a-correctness-check/6af1d05036a54cca5df5fe55)

### [The reward is part of the dataset](https://nodus-compute.hashnode.dev/the-reward-is-part-of-the-dataset?utm_source=hashnode&utm_medium=feed)

Six 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.dev](https://nodus-compute.hashnode.dev/the-reward-is-part-of-the-dataset?utm_source=hashnode&utm_medium=feed)

[2d ago](https://hashnode.com/posts/the-reward-is-part-of-the-dataset/6af192e0befea3fedc931883)