Every premium and members-only resource for $9.99/month, as a no-commitment subscription. See the plans

PalateKit

Sign in
Menu

Request a resource

Describe the resource you are missing in one sentence. We will email you when it is published.

0 / 280

A page, a video or an image online.

Unlock PalateKit

Unlock everything

Plan

Paying means accepting our terms of sale.

Secure payment by Stripe. Receipts and invoices sent by email.

Welcome back

Sign in to install, download or save this resource. A free account is enough for free resources.

Reducing GPU memory in Megatron Bridge v1.0.0

Use this resource to diagnose and reduce GPU out-of-memory errors during training

New

  • 305 downloads
  • 155 likes
  • 174 saves

Free

Install with the CLI

In a terminal, from the folder you want:

npx --yes palatekit@latest install gpu-memory-reduction-techniques

With no option, the skill goes to ~/.agents/skills/; add --agent claude-code (or codex, cursor…) to target an agent.

First time: connect the CLI with

npx --yes palatekit@latest login

The agent first checks whether the OOM comes from allocator fragmentation, then proposes the levers: PYTORCH_CUDA_ALLOC_CONF with expandable_segments, the recommended first step, parallelism resizing, activation recompute and CPU offloading. You get settings and their compatibility constraints.

Mostly picked by

Updated November 2, 2026

Card for the “Reducing GPU memory in Megatron Bridge” skill

Details

Works in
Install guides
Generates with
Your agent's model
Licence
Apache-2.0

Similar resources

View all

What you get

  • SKILL.md: levers and constraints
  • card.yaml: structured skill card
  • evals/evals.json: evaluation cases

Good fit for

  • A training run fails with a CUDA memory error
  • You want to free memory without changing the model

Not for

    Tags

    • GPU memory
    • OOM
    • Megatron Bridge
    • recompute

    About

    You say

    “My Megatron Bridge run goes out of memory on 8 H100s: which settings should I try first?”

    Card for the “Reducing GPU memory in Megatron Bridge” skill

    Skill published by nvidia under the Apache-2.0 license, delivered in its original version.

    Show more

    How to use it

    1. Install the skill in your agent's skills folder
    2. Ask your agent to diagnose and reduce GPU out-of-memory errors during training
    3. Review and adjust the result

    Content preview

    Files

    • BENCHMARK.md4 KB
    • card.yaml10 KB
    • evals/evals.json1.5 KB
    • LICENSE11.3 KB
    • NOTICE-PALATEKIT.md0.5 KB
    • skill-card.md4.1 KB
    • SKILL.md14.4 KB
    • skill.oms.sig4.7 KB

    Origin and licence

    Listed unmodified

    Author
    nvidia
    Source
    github.com/nvidia/skills/tree/ca116d60beafe1b4459f0ab790b7b450912c56b0/skills/nemo-mbridge-perf-memory-tuning
    Licence
    Apache-2.0

    Notice

    « nemo-mbridge-perf-memory-tuning » (nvidia, Apache-2.0) livré dans sa version d'origine : fiche en français et en anglais par PalateKit.

    User reviews

    –

    No ratings yet

    Loading reviews…

    Delete your review?

    Your rating and comment will be removed from the page.