Source profileQuality 95/100

NVIDIA/skills/skills/jetson-speculative-decoding/SKILL.md

jetson-speculative-decoding

Add EAGLE-3 or draft-model speculative decoding to a Jetson vLLM server when TPOT is the bottleneck.

Source repository stars
3,106
Declared platforms
0
Static risk flags
0
Last source update
2026-08-25
Source checked
2026-08-26

Decision brief

What it does: where it fits

Speculative decoding lets a small "draft" model propose tokens that the target model verifies in a single forward pass, reducing per-token latency. On Jetson, the win/loss is dominated by VRAM headroom, not by the draft quality. This skill encodes the parts an LLM won't already…

Best for

  • TPOT/ITL is the bottleneck (TTFT is fine, output is just slow).
  • Workload is single-stream or low-concurrency (≤2). Speculation usually loses at high concurrency.
  • Jetson family is Thor or AGX Orin. Do not suggest EAGLE-3 on Orin Nano/NX — there is rarely enough VRAM headroom to host both target and draft, and you'll OOM at startup.

Not for

  • Speculative decoding improves decode-heavy workloads; it does not reduce TTFT-dominated latency.
  • High concurrency can erase the benefit because continuous batching already keeps the GPU busy.

Compatibility matrix

Platform support, with evidence labels

PlatformStatusEvidenceWhat to check
CodexNot declaredNo explicit evidencePortability before use
Claude CodeNot declaredNo explicit evidencePortability before use
CursorNot declaredNo explicit evidencePortability before use
Gemini CLINot declaredNo explicit evidencePortability before use
Open the compatibility checker

Installation

Inspect first. Install second.

The source command is displayed only when detected. A safe inspection prompt is always available so your agent can explain every action before execution.

Source-detected install commandSource
npx skills add https://github.com/NVIDIA/skills --skill "skills/jetson-speculative-decoding"
Safe inspection promptEditorial

Inspect the Agent Skill "jetson-speculative-decoding" from https://github.com/NVIDIA/skills/blob/994b87022af46deada9fdb79fc560a77aaf931ce/skills/jetson-speculative-decoding/SKILL.md at commit 994b87022af46deada9fdb79fc560a77aaf931ce. List every install step, command, network request, credential, file read/write, external action, and rollback step. Explain whether it fits my task. Do not install or execute anything until I approve.

Workflow

What the source asks the agent to do

  1. 01

    Instructions

    Append --speculative-config to the vllm serve command shown in jetson-llm-serve.

    numspeculativetokens: start at 5 on Thor, 3 on AGX Orin. Higher values pay off only if the draft acceptance rate is 0.6.Always pair with the same vLLM runtime path used by jetson-llm-serve: upstream vLLM 0.20+ (vllm/vllm-openai:latest) or validated native vLLM 0.20+ on Thor, upstream vLLM 0.20+ on Orin JetPack 7.2 / L4T r39+, or the NVID…Drop --gpu-memory-utilization by 0.05 vs the non-speculative baseline to give the draft model headroom.
  2. 02

    How to verify it actually helped

    1. Run jetson-llm-benchmark (vLLM path) at --concurrency 1 before and after enabling speculation. 2. Acceptance: target ≥30% improvement in throughputtoks and ≥20% drop in tpotmsp50 at concurrency 1. 3. If improvement is <10%, or throughputtoks regresses at concurrency 8, disabl…

    Run jetson-llm-benchmark (vLLM path) at --concurrency 1 before and after enabling speculation.Acceptance: target ≥30% improvement in throughputtoks and ≥20% drop in tpotmsp50 at concurrency 1.If improvement is <10%, or throughputtoks regresses at concurrency 8, disable speculation. The draft model is costing more than it returns.
  3. 03

    Purpose

    Tune an existing Jetson vLLM deployment for faster token generation by appending the right --speculative-config and validating whether it improves single-stream decode speed.

    Tune an existing Jetson vLLM deployment for faster token generation by appending the right --speculative-config and validating whether it improves single-stream decode speed.
  4. 04

    When to use

    TPOT/ITL is the bottleneck (TTFT is fine, output is just slow).

    TPOT/ITL is the bottleneck (TTFT is fine, output is just slow).Workload is single-stream or low-concurrency (≤2). Speculation usually loses at high concurrency.Jetson family is Thor or AGX Orin. Do not suggest EAGLE-3 on Orin Nano/NX — there is rarely enough VRAM headroom to host both target and draft, and you'll OOM at startup.
  5. 05

    When NOT to use

    High-concurrency serving (≥8): batched decode usually beats speculation; the draft model just steals VRAM.

    High-concurrency serving (≥8): batched decode usually beats speculation; the draft model just steals VRAM.Models without a published EAGLE-3 head — do not train one ad-hoc as a "fix".After applying jetson-inference-mem-tune flags that already pushed --gpu-memory-utilization near the ceiling. Free at least 2 GB first.

Permission review

Static risk signals and limitations

No configured static risk pattern was detected

This is not proof of safety. Runtime behavior, indirect dependencies, and hidden external systems are outside the static scan.

Evidence record

Why each signal appears

EvidenceSourceComputedTestedEditorial
SignalValueEvidence typeMeaning
Quality score95/100ComputedDocumentation, specificity, maintenance, and trust rules
Repository stars3,106SourceRepository attention, not individual Skill quality
Compatibility0 platformsSourceDeclared in the catalog source record
Usage guideautomated source guideEditorialGenerated or reviewed according to the visible evidence level

Pinned source

Provenance and original SKILL.md

Repository
NVIDIA/skills
Skill path
skills/jetson-speculative-decoding/SKILL.md
Commit
994b87022af46deada9fdb79fc560a77aaf931ce
License
Apache-2.0
Collected
2026-08-26
Default branch
main
View the original SKILL.md

Jetson Speculative Decoding (vLLM)

Speculative decoding lets a small "draft" model propose tokens that the target model verifies in a single forward pass, reducing per-token latency. On Jetson, the win/loss is dominated by VRAM headroom, not by the draft quality. This skill encodes the parts an LLM won't already know.

Purpose

Tune an existing Jetson vLLM deployment for faster token generation by appending the right --speculative-config and validating whether it improves single-stream decode speed.

When to use

  • TPOT/ITL is the bottleneck (TTFT is fine, output is just slow).
  • Workload is single-stream or low-concurrency (≤2). Speculation usually loses at high concurrency.
  • Jetson family is Thor or AGX Orin. Do not suggest EAGLE-3 on Orin Nano/NX — there is rarely enough VRAM headroom to host both target and draft, and you'll OOM at startup.

When NOT to use

  • High-concurrency serving (≥8): batched decode usually beats speculation; the draft model just steals VRAM.
  • Models without a published EAGLE-3 head — do not train one ad-hoc as a "fix".
  • After applying jetson-inference-mem-tune flags that already pushed --gpu-memory-utilization near the ceiling. Free at least ~2 GB first.

Prerequisites

  • A working vLLM server recipe from jetson-llm-serve.
  • Enough memory headroom for the draft model or EAGLE-3 head in addition to the target model.
  • A benchmark baseline from jetson-llm-benchmark before enabling speculation.
  • A target model with a compatible EAGLE-3 head, or a small same-family draft model for the fallback path.

Instructions

Append --speculative-config to the vllm serve command shown in jetson-llm-serve.

EAGLE-3 (preferred when a head is published for the target model):

--speculative-config '{
  "method": "eagle3",
  "model": "<eagle3-head-repo-id>",
  "num_speculative_tokens": 5,
  "draft_tensor_parallel_size": 1
}'

Draft-model (fallback — pair a small same-family model):

--speculative-config '{
  "method": "draft_model",
  "model": "<small-draft-model-repo-id>",
  "num_speculative_tokens": 4,
  "draft_tensor_parallel_size": 1
}'

Jetson-specific tuning rules

  • num_speculative_tokens: start at 5 on Thor, 3 on AGX Orin. Higher values pay off only if the draft acceptance rate is >0.6.
  • Always pair with the same vLLM runtime path used by jetson-llm-serve: upstream vLLM 0.20+ (vllm/vllm-openai:latest) or validated native vLLM 0.20+ on Thor, upstream vLLM 0.20+ on Orin JetPack 7.2 / L4T r39+, or the NVIDIA-AI-IOT vLLM image on older Orin. Do not use an Orin NVIDIA-AI-IOT vLLM image on Thor. Older runtimes may lack EAGLE-3 or the current --speculative-config shape.
  • Drop --gpu-memory-utilization by ~0.05 vs the non-speculative baseline to give the draft model headroom.

How to verify it actually helped

  1. Run jetson-llm-benchmark (vLLM path) at --concurrency 1 before and after enabling speculation.
  2. Acceptance: target ≥30% improvement in throughput_tok_s and ≥20% drop in tpot_ms_p50 at concurrency 1.
  3. If improvement is <10%, or throughput_tok_s regresses at concurrency 8, disable speculation. The draft model is costing more than it returns.

Limitations

  • Speculative decoding improves decode-heavy workloads; it does not reduce TTFT-dominated latency.
  • High concurrency can erase the benefit because continuous batching already keeps the GPU busy.
  • Orin Nano/NX usually lack enough memory headroom for both target and draft models.
  • Acceptance rate and draft overhead are model-specific, so benchmark before and after instead of assuming a speedup.

Error handling

  • If vLLM rejects --speculative-config, verify that Thor and Orin JetPack 7.2 / L4T r39+ are using vLLM 0.20+ and that older Orin is using a JetPack-matched NVIDIA-AI-IOT vLLM image; then switch back to the non-speculative serving command if the runtime still rejects it.
  • If startup OOMs, lower --gpu-memory-utilization, use a smaller draft, or disable speculation and hand off to jetson-inference-mem-tune.
  • If benchmark throughput regresses, remove --speculative-config; a bad draft path is worse than no speculation.

Hand off to

  • jetson-llm-benchmark to quantify the change.
  • jetson-inference-mem-tune if startup OOMs after enabling speculation.

Source

vLLM speculative decoding docs and the Jetson AI Lab GenAI tutorial.

Frequently asked questions

What to verify before installation and use

What does the jetson-speculative-decoding source document cover?

Speculative decoding lets a small "draft" model propose tokens that the target model verifies in a single forward pass, reducing per-token latency. On Jetson, the win/loss is dominated by VRAM headroom, not by the draft quality. This skill encodes the parts an LLM won't already…

How do I install jetson-speculative-decoding?

The source record exposes this install command: npx skills add https://github.com/NVIDIA/skills --skill "skills/jetson-speculative-decoding". Inspect the command and pinned source before running it.