skills.rest
IsNoobgrammer
IsNoobgrammerCOMMUNITY
Explore all skills by this author →
Website𝕏@ShauryaSharthakGitHub

ml-engine

Build reproducible distributed training pipelines for TPU hardware with PyTorch-XLA and JAX.

4|2|Updated Apr 26, 2026
Search Tags:#jax#machine-learning#distributed-training#tpu#pytorch-xla#spmd#fsdpv2
One-click install
npx skills add https://github.com/IsNoobgrammer/skills-for-agents --skill ml-engine
Or copy as Structured Prompt for Agent▼
Please help me install this Agent Skill.
Skill: ml-engine
Source: https://github.com/IsNoobgrammer/skills-for-agents/tree/main/skills/ml-engine
Command: npx skills add https://github.com/IsNoobgrammer/skills-for-agents --skill ml-engine

SYSTEM DOCUMENTATION & REQUIREMENTS

💡 This Skill requires torch_xla, jax, einops, datasets, wandb, and includes references (resource) components.

What problem does it solve?

This skill solves the complexity of managing distributed training on TPU hardware, eliminating silent recompilation bugs and manual sharding overhead.

Core Features & Use Cases

  • Distributed Training: Orchestrates SPMD and FSDPv2 training pipelines on TPU v2/v3/v5 pods.
  • Research Ablations: Provides a reproducible scaffold for testing novel architectures like MoE and custom attention kernels.
  • Use Case: Use this skill to rapidly prototype a sparse MoE model, automatically sharding the router and experts across a TPU v3-64 pod while ensuring deterministic data streaming.

Quick Start

Invoke the ml-engine skill to generate a full research scaffold for a sparse MoE model with dynamic clamping on a v3-8 TPU.

Frequently Asked Questions about ml-engine

High-intent search queries and answers about installing and using this skill.

FAQPage Schema
How do I set up distributed training on TPU pods using PyTorch-XLA?▼

Distributed training on TPU pods requires orchestrating SPMD sharding and FSDPv2 pipelines. This engine generates reproducible scaffolds for novel architecture ablations across TPU v2/v3/v5 pods, handling automated sharding and deterministic data streaming.

What is the best way to shard a sparse MoE model across a TPU v3-64 pod?▼

Sharding a sparse MoE model across a TPU v3-64 pod involves distributing the router and experts automatically. This skill provides a reproducible scaffold that handles SPMD sharding and deterministic data streaming for rapid prototyping.

Does this distributed training engine support multi-pod scaling with JAX?▼

Yes, multi-pod scaling is supported for JAX and PyTorch-XLA research workflows. The engine satisfies requirements for deterministic data streaming and multi-pod scaling on TPU hardware, eliminating manual sharding overhead.

Why does my PyTorch-XLA model trigger silent recompilation on TPU?▼

Silent recompilation bugs on TPU stem from complexity in managing distributed training. This skill eliminates these bugs by building reproducible pipelines that ensure deterministic data streaming and automated kernel selection.

Can I test custom attention kernels with FSDPv2 on TPU v5 pods?▼

Yes, you can test custom attention kernels using FSDPv2 on TPU v5 pods. The engine provides a reproducible scaffold for novel architecture ablations, featuring automated attention kernel selection and dynamic clamping.

Related Skills

Explore More Software Engineering Skills→
Tianyi-Billy-MaTianyi-Billy-Ma

pytorch-lightning

Automate PyTorch training workflows with LightningModule, Trainer, and LightningDataModule.

Community
Advanced
RheasilviaRheasilvia

weights-and-biases

Log machine learning experiment metrics and hyperparameters to Weights & Biases.

Community
Advanced
mvillmowmvillmow
18

train-model

Outline Mojo training loops with data handling, optimization, and checkpointing.

Community
Basic
simple-platformsimple-platform
1

scl-expressions

Explain SCL dynamic expressions with backtick syntax and built-in functions.

Official
Intermediate