Automation & Ops ✓ NVIDIA · Official jetsonspeculative-decodingvllmlatency-optimizationeagle-3draft-model

Jetson vLLM Speculative Decoding

Configure speculative decoding for lower single-stream generation latency on Jetson vLLM servers.

FollowSkills review · FSRS-2.0
Not recommended
54/ 100 5-point scale 2.7 / 5
Trust17 / 25 · 3.4/5

The skill clearly scopes Jetson families, concurrency, and memory prerequisites, and provides disable, OOM, and benchmark fallback guidance. No malware, credential theft, or destructive default is evident. Points are deducted because user confirmation is not explicit, data flows and trust boundaries for model repositories and container images are not disclosed, and use of a mutable latest image reduces control.

Reliability8 / 20 · 2.0/5

The instructions consistently cover EAGLE-3 and draft-model branches, version conditions, OOM handling, and performance fallback. Static files contain no executable test suite for this skill, and the benchmark report lacks its source dataset; several compatibility claims are therefore not reproducible from the supplied files. The score stays within the static-review cap.

Adaptability11 / 15 · 3.7/5

Use cases, Jetson models, TPOT/TTFT conditions, concurrency limits, and out-of-scope devices are clearly stated. Points are deducted for missing input/output examples and more precise exclusion triggers, no Chinese-language support, and no discussion of mainland-China reachability for containers, model repositories, or NVIDIA services.

Convention8 / 15 · 2.7/5

The documentation has clear purpose, prerequisites, runtime constraints, verification, limitations, error handling, and handoff sections; version, licensing, and NVIDIA ownership are visible. Points are deducted for the missing recommended Examples section, the author format finding, inconsistent license wording between frontmatter and the skill card, and no explicit changelog, maintenance contact, or update path.

Effectiveness6 / 15 · 2.0/5

The skill supplies both speculative-config paths, device-specific token settings, memory adjustment, and before/after acceptance criteria, so it plausibly completes the core configuration task. Points are deducted because repository IDs, the base serve command, and actual benchmark outputs remain user-supplied, and the skill has no independently reproducible result evidence. Static review therefore cannot establish performance benefit and the score remains at or below 7.

Verifiability4 / 10 · 2.0/5

The skill cites vLLM documentation and a Jetson AI Lab tutorial, and the supplied materials include a revision-pinned evaluation report. Points are deducted because the report provides no underlying dataset or trial records, the skill has no key-path tests, and major technical claims are not individually corroborated within the supplied files.

Evidence confidence:Low Reviewed Jul 20, 2026 Reviewed revision 55f18499943e
Before you use it
  • Before enabling, verify target-model and EAGLE-3-head or draft-model compatibility, image provenance, versions, and VRAM headroom; pin image and model versions instead of relying on latest.
  • Keep a non-speculative baseline and remeasure against the stated thresholds; remove speculative-config on startup OOM or performance regression.
  • Mainland-China reachability and organizational compliance for model repositories, container pulls, and NVIDIA-related services are not addressed.
See the full review method →

What does this skill do, and when should you use it?

This skill tunes an existing vLLM server on Jetson Thor or AGX Orin with EAGLE-3 or a draft model. It targets decode-bound workloads with low concurrency and treats available VRAM as the primary constraint. The skill appends the appropriate speculative-decoding configuration and compares performance before and after the change. It is generally unsuitable for Orin Nano/NX, high-concurrency serving, or deployments with little memory headroom.

Guides the user to append --speculative-config to an existing vllm serve command, choose eagle3 or draft_model, provide a model repository ID, and set speculative-token and draft tensor-parallel parameters; adjusts the starting token count for Thor or AGX Orin and lowers gpu-memory-utilization relative to baseline; runs jetson-llm-benchmark at concurrency 1 before and after activation; and removes the configuration, lowers memory utilization, or selects a smaller draft model when vLLM rejects the option, startup OOMs, or benchmarks regress.

  1. A Jetson Thor user wants faster single-stream generation and the target model has a compatible published EAGLE-3 head.
  2. An AGX Orin deployment has acceptable TTFT but high per-token latency at concurrency no higher than 2.
  3. An engineering team needs benchmark-based evidence before keeping speculative decoding enabled.
  4. A vLLM server runs out of memory after speculative decoding is enabled and needs a documented rollback path.

What are this skill's strengths and limitations?

Pros
  • Provides concrete EAGLE-3 and draft-model configuration templates.
  • Defines different starting speculative-token values for Thor and AGX Orin.
  • Documents handling for rejected arguments, startup OOMs, and benchmark regressions.
  • Uses before-and-after benchmarking instead of assuming a speedup.
Limitations
  • Designed for decode-bound, single-stream or low-concurrency workloads.
  • Orin Nano/NX usually lacks enough VRAM for both target and draft models.
  • Requires a compatible EAGLE-3 head or same-family draft model; no concrete model IDs are supplied.
  • The provided source does not show independent benchmark results or a dedicated test suite for this skill.

How do you install this skill?

Install it using the repository README's skills CLI flow: npx skills add nvidia/skills --skill jetson-speculative-decoding --yes. The source does not document a dedicated installer or additional installation script for this skill.

How do you use this skill?

Prepare a working jetson-llm-serve vLLM recipe and a jetson-llm-benchmark baseline, then prompt an agent: "Tune my Jetson vLLM server with speculative decoding and benchmark the result." The agent should select EAGLE-3 or a draft model based on the Jetson platform, model compatibility, and VRAM headroom. The source does not provide actual EAGLE-3 head or draft-model repository IDs, so the user must supply them.

How does this skill compare with similar options?

The skill explicitly compares speculative decoding with continuous batching: at concurrency 8 or higher, batched decoding usually wins. For startup OOMs, it hands off to jetson-inference-mem-tune.

FAQ

Which Jetson platforms are supported?
The instructions target Jetson Thor and AGX Orin. They specifically advise against suggesting EAGLE-3 on Orin Nano/NX because of likely VRAM limits.
What must already be available?
A working jetson-llm-serve recipe, sufficient memory for the target plus draft component, a pre-change jetson-llm-benchmark baseline, and either a compatible EAGLE-3 head or same-family draft model.
What if the speedup is small or throughput falls?
Disable speculation if improvement is below 10% or throughput regresses at concurrency 8, then return to the non-speculative serving command.
Does the skill provide the model repository IDs?
No. Its command examples use placeholders for the EAGLE-3 head and draft model repositories.

More skills from this repository

All from NVIDIA/skills

Automation & Ops ✓ NVIDIA · Official

Jetson Package & Environment Advisor

Select Jetson-compatible containers and GPU package indexes by device architecture.

Automation & Ops ✓ NVIDIA · Official

Jetson LLM Serving

Deploy OpenAI-compatible LLM and VLM servers on Jetson with vLLM or SGLang.

Automation & Ops ✓ NVIDIA · Official

Jetson LLM Benchmark

Measure and compare LLM runtime performance on Jetson with structured JSON results.

Dev & Engineering ✓ NVIDIA · Official

Jetson Memory Optimizer

Reclaim unused DRAM on headless or no-camera Jetson deployments.

Dev & Engineering ✓ NVIDIA · Official

Jetson Inference Memory Tuning

Choose a Jetson serving runtime and generate memory flags from the device snapshot and workload.

Dev & Engineering ✓ NVIDIA · Official

Jetson Target Switcher

Switch the active Jetson target among existing platform profiles.

Dev & Engineering ✓ NVIDIA · Official

Jetson BSP Info Check

Verify Jetson BSP version, board configs, and rootfs state before flashing.

Dev & Engineering ✓ NVIDIA · Official

Jetson BSP Source Builder

Rebuild Jetson BSP device trees, kernels, and modules from source changes and produce a deployable artifact manifest.

Automation & Ops ✓ NVIDIA · Official

Jetson Device Info Snapshot

Capture a concise software, kernel, OS, and power-mode baseline from a running Jetson target.

Dev & Engineering ✓ NVIDIA · Official

Jetson PCIe Controller Customizer

Configure per-controller PCIe state, lane width, and link speed on custom Jetson Thor or Orin carriers.

Dev & Engineering ✓ NVIDIA · Official

Jetson Target Platform Initializer

Create a Jetson target-platform profile for a reference devkit or custom carrier and update the active pointer.

Dev & Engineering ✓ NVIDIA · Official

Jetson Clock Customizer

Lock or cap Jetson CPU, GPU, and EMC clock behavior before flashing a BSP image.

Automation & Ops ✓ NVIDIA · Official

Jetson Health Snapshot

Read-only diagnostics that consolidate a live Jetson’s hardware, resource, and service state.

Dev & Engineering ✓ NVIDIA · Official

Jetson BSP Image Flasher

Safely flash a promoted Jetson BSP image onto a Jetson device in RCM mode.

Dev & Engineering ✓ NVIDIA · Official

Jetson BSP Quick Start

Choose a Jetson BSP customization path and hand validated intake answers to the right setup skills.

Dev & Engineering ✓ NVIDIA · Official

Jetson USB Port Customization

Safely enable, disable, or change Jetson USB port roles through a kernel device-tree overlay.

Automation & Ops ✓ NVIDIA · Official

Jetson Headless Memory Optimizer

Reclaim memory on GUI-free Jetson devices through safe, reversible system-service changes.

Dev & Engineering ✓ NVIDIA · Official

Jetson Fan Curve Customizer

Safely maintain nvfancontrol fan profiles and boot defaults in a Jetson BSP.

Automation & Ops ✓ NVIDIA · Official

Jetson nvpmodel Power Mode Customizer

Safely customize Jetson BSP power modes, frequency caps, and the boot-default profile.

Dev & Engineering ✓ NVIDIA · Official

NVIDIA Skill Finder

Finds the right NVIDIA agent skill for product, hardware, and workflow requests.

Related skills