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.

Activation recompute to reduce memory v1.0.0

Use this resource to reduce activation memory with selective or full recompute

New

  • 351 downloads
  • 124 likes
  • 156 saves

Free

Install with the CLI

In a terminal, from the folder you want:

npx --yes palatekit@latest install activation-recompute-memory-tuning

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 confirms that the memory pressure really comes from activations, chooses between per-module selective recompute and full per-layer recompute for the model, sets recompute_granularity and the modules, then compares measurements before and after. You get a configuration and memory measurements.

Mostly picked by

Updated November 2, 2026

Card for the “Activation recompute to reduce memory” 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: method and decision table
  • card.yaml: structured card
  • evals/evals.json: evaluation cases

Good fit for

  • An OOM comes from activations rather than fragmentation
  • You compare selective and full recompute on a model

Not for

    Tags

    • recompute
    • activations
    • memory
    • Megatron Bridge

    About

    You say

    “Is my OOM coming from activations? Tell me which module to recompute in selective mode.”

    Card for the “Activation recompute to reduce memory” 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 reduce activation memory with selective or full recompute
    3. Review and adjust the result

    Content preview

    Files

    • BENCHMARK.md4.9 KB
    • card.yaml16.8 KB
    • evals/evals.json5.4 KB
    • LICENSE11.3 KB
    • NOTICE-PALATEKIT.md0.5 KB
    • skill-card.md4.3 KB
    • SKILL.md18.7 KB
    • skill.oms.sig4.7 KB

    Origin and licence

    Listed unmodified

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

    Notice

    « nemo-mbridge-perf-activation-recompute » (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.