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.

Speculative decoding to speed up an LLM v1.0.0

Use this resource to speed up LLM inference without changing the model architecture

New

  • 349 downloads
  • 247 likes
  • 192 saves

Free

Install with the CLI

In a terminal, from the folder you want:

npx --yes palatekit@latest install llm-speculative-decoding-speedup

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 helps you choose between a draft model, Medusa and Lookahead, then writes the speculative generation code with transformers or vLLM. You get installation steps, a quick start and the checks to run before deployment.

Mostly picked by

Updated November 2, 2026

Card for the “Speculative decoding to speed up an LLM” skill

Details

Works in
Install guides
Generates with
Your agent's model
Licence
MIT

Similar resources

View all

What you get

  • SKILL.md: use cases, installation and quick start
  • references/medusa.md: decoding with Medusa multiple heads
  • references/lookahead.md: Lookahead decoding with Jacobi iteration

Good fit for

  • Latency is too high in your chatbot or code generation feature
  • You serve a model on limited hardware

Not for

    Tags

    • inference
    • LLM
    • speculative decoding
    • latency

    About

    You say

    “Add speculative decoding with a small draft model to my vLLM server.”

    Card for the “Speculative decoding to speed up an LLM” skill

    Skill published by Orchestra Research under the MIT 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 speed up LLM inference without changing the model architecture
    3. Review and adjust the result

    Content preview

    Files

    • LICENSE1.1 KB
    • NOTICE-PALATEKIT.md0.5 KB
    • references/lookahead.md9 KB
    • references/medusa.md10 KB
    • SKILL.md14.2 KB

    Origin and licence

    Listed unmodified

    Author
    Orchestra Research
    Source
    github.com/orchestra-research/ai-research-skills/tree/773a52944ba4747a18bd4ae9ade53fff041adcbc/19-emerging-techniques/speculative-decoding
    Licence
    MIT

    Notice

    « speculative-decoding » (Orchestra Research, MIT) 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.