dgx-station-inference
Skill1.5k starsChanged 2 months agoApache-2.0View on GitHub
Resolve, tune, preflight, launch, verify, inspect, and stop exact-model inference on NVIDIA DGX Station through dgx-assist. Use for vLLM or SGLang container selection, NGC versus upstream, GPU memory utilization, CPU or KV offload, HBM fit, KV-cache sizing, ISL or context length, prefix caching, chunked prefill, batching, concurrency, performance tuning, serving or deploying a named model, an OpenAI-compatible endpoint, Station recipe models, or an owned inference service. Require an exact model ID for recipe resolution or model-specific tuning, and never recommend or substitute a different model.
Trending
More talk lately.about the whole repository
Who's talking about NVIDIA/dgx-spark-playbooks
- CarstVaartjes/vonk-forge-recipes“Refresh needs review: gemma-4-e4b-it-nvidia-vllm-single”GitHub · Nov 3, 2026
- CarstVaartjes/vonk-forge-recipes“Refresh needs review: gemma-4-e2b-it-nvidia-vllm-single”GitHub · Nov 3, 2026
- laxshmate/genmodelfit“すべての環境で検出できるようにする(41 通りの環境の洗い出し・GB10・WSL2 の AMD/Intel・GPU 名は表でなく文法で)”GitHub · Oct 27, 2026
- CarstVaartjes/vonk-forge-recipes“Refresh qwen3-8b-nvfp4-nvidia-sglang-single to the newest upstream pins”GitHub · Oct 13, 2026
- CarstVaartjes/vonk-forge-recipes“Refresh qwen3-8b-fp8-nvidia-vllm-single to the newest upstream pins”GitHub · Oct 13, 2026
The biggest 5 of 50 mentions linking to the repository that our crawl found. Each line is their own headline.
See all 50
Talked about on GitHub (50 times).
On GitHub · 50
- Nov 3, 2026Refresh needs review: gemma-4-e4b-it-nvidia-vllm-single — CarstVaartjes/vonk-forge-recipes0 reactions · 1 comment
- Nov 3, 2026Refresh needs review: gemma-4-e2b-it-nvidia-vllm-single — CarstVaartjes/vonk-forge-recipes0 reactions · 1 comment
- Oct 27, 2026すべての環境で検出できるようにする(41 通りの環境の洗い出し・GB10・WSL2 の AMD/Intel・GPU 名は表でなく文法で) — laxshmate/genmodelfit0 reactions
- Oct 13, 2026Refresh qwen3-8b-nvfp4-nvidia-sglang-single to the newest upstream pins — CarstVaartjes/vonk-forge-recipes0 reactions · 1 comment
- Oct 13, 2026Refresh qwen3-8b-fp8-nvidia-vllm-single to the newest upstream pins — CarstVaartjes/vonk-forge-recipes0 reactions · 1 comment
45 more on GitHub
- Oct 13, 2026Refresh qwen3-8b-fp8-nvidia-trtllm-single to the newest upstream pins — CarstVaartjes/vonk-forge-recipes0 reactions
- Oct 13, 2026Refresh qwen3-8b-fp8-nvidia-sglang-single to the newest upstream pins — CarstVaartjes/vonk-forge-recipes0 reactions
- Oct 13, 2026Refresh qwen3-6-35b-a3b-mtp-q4-k-xl-nvidia-llamacpp-single to the newest upstream pins — CarstVaartjes/vonk-forge-recipes0 reactions
- Oct 13, 2026Refresh qwen3-6-35b-a3b-mtp-q4-k-xl-nvidia-llamacpp-single to the newest upstream pins — CarstVaartjes/vonk-forge-recipes0 reactions
- Oct 13, 2026Refresh qwen3-32b-nvfp4-nvidia-vllm-single to the newest upstream pins — CarstVaartjes/vonk-forge-recipes0 reactions
- Oct 13, 2026Refresh qwen3-32b-nvfp4-nvidia-trtllm-single to the newest upstream pins — CarstVaartjes/vonk-forge-recipes0 reactions
- Oct 13, 2026Refresh qwen3-32b-nvfp4-nvidia-sglang-single to the newest upstream pins — CarstVaartjes/vonk-forge-recipes0 reactions
- Oct 13, 2026Refresh qwen3-30b-a3b-nvfp4-nvidia-trtllm-single to the newest upstream pins — CarstVaartjes/vonk-forge-recipes0 reactions
- Oct 13, 2026Refresh qwen3-14b-nvfp4-nvidia-vllm-single to the newest upstream pins — CarstVaartjes/vonk-forge-recipes0 reactions
- Oct 13, 2026Refresh qwen3-14b-nvfp4-nvidia-trtllm-single to the newest upstream pins — CarstVaartjes/vonk-forge-recipes0 reactions
- Oct 13, 2026Refresh qwen3-14b-nvfp4-nvidia-sglang-single to the newest upstream pins — CarstVaartjes/vonk-forge-recipes0 reactions
- Oct 13, 2026Refresh qwen3-14b-fp8-nvidia-vllm-single to the newest upstream pins — CarstVaartjes/vonk-forge-recipes0 reactions
- Oct 13, 2026Refresh qwen3-14b-fp8-nvidia-trtllm-single to the newest upstream pins — CarstVaartjes/vonk-forge-recipes0 reactions
- Oct 13, 2026Refresh qwen3-14b-fp8-nvidia-sglang-single to the newest upstream pins — CarstVaartjes/vonk-forge-recipes0 reactions
- Oct 13, 2026Refresh qwen2-5-vl-7b-instruct-nvfp4-nvidia-vllm-single to the newest upstream pins — CarstVaartjes/vonk-forge-recipes0 reactions
- Oct 13, 2026Refresh phi-4-reasoning-plus-nvfp4-nvidia-vllm-single to the newest upstream pins — CarstVaartjes/vonk-forge-recipes0 reactions
- Oct 13, 2026Refresh phi-4-reasoning-plus-nvfp4-nvidia-sglang-single to the newest upstream pins — CarstVaartjes/vonk-forge-recipes0 reactions
- Oct 13, 2026Refresh phi-4-reasoning-plus-fp8-nvidia-vllm-single to the newest upstream pins — CarstVaartjes/vonk-forge-recipes0 reactions
- Oct 13, 2026Refresh phi-4-reasoning-plus-fp8-nvidia-sglang-single to the newest upstream pins — CarstVaartjes/vonk-forge-recipes0 reactions
- Oct 13, 2026Refresh phi-4-multimodal-instruct-nvfp4-nvidia-vllm-single to the newest upstream pins — CarstVaartjes/vonk-forge-recipes0 reactions
- Oct 13, 2026Refresh phi-4-multimodal-instruct-nvfp4-nvidia-sglang-single to the newest upstream pins — CarstVaartjes/vonk-forge-recipes0 reactions
- Oct 13, 2026Refresh phi-4-multimodal-instruct-fp8-nvidia-vllm-single to the newest upstream pins — CarstVaartjes/vonk-forge-recipes0 reactions · 1 comment
- Oct 13, 2026Refresh phi-4-multimodal-instruct-fp8-nvidia-sglang-single to the newest upstream pins — CarstVaartjes/vonk-forge-recipes0 reactions
- Oct 13, 2026Refresh nemotron-3-nano-omni-30b-a3b-bf16-nvidia-sglang-single to the newest upstream pins — CarstVaartjes/vonk-forge-recipes0 reactions
- Oct 13, 2026Refresh nemotron-3-nano-30b-a3b-fp8-nvidia-vllm-single to the newest upstream pins — CarstVaartjes/vonk-forge-recipes0 reactions
- Oct 13, 2026Refresh nemotron-3-nano-30b-a3b-bf16-nvidia-vllm-single to the newest upstream pins — CarstVaartjes/vonk-forge-recipes0 reactions
- Oct 13, 2026Refresh llama-3-3-70b-instruct-nvfp4-nvidia-vllm-single to the newest upstream pins — CarstVaartjes/vonk-forge-recipes0 reactions
- Oct 13, 2026Refresh llama-3-3-70b-instruct-nvfp4-nvidia-sglang-single to the newest upstream pins — CarstVaartjes/vonk-forge-recipes0 reactions
- Oct 13, 2026Refresh llama-3-3-70b-instruct-nvfp4-nvidia-sglang-single to the newest upstream pins — CarstVaartjes/vonk-forge-recipes0 reactions
- Oct 13, 2026Refresh llama-3-1-8b-instruct-nvfp4-nvidia-vllm-single to the newest upstream pins — CarstVaartjes/vonk-forge-recipes0 reactions
- Oct 13, 2026Refresh llama-3-1-8b-instruct-nvfp4-nvidia-sglang-single to the newest upstream pins — CarstVaartjes/vonk-forge-recipes0 reactions
- Oct 13, 2026Refresh llama-3-1-8b-instruct-fp8-nvidia-vllm-single to the newest upstream pins — CarstVaartjes/vonk-forge-recipes0 reactions
- Oct 13, 2026Refresh llama-3-1-8b-instruct-fp8-nvidia-vllm-single to the newest upstream pins — CarstVaartjes/vonk-forge-recipes0 reactions
- Oct 13, 2026Refresh llama-3-1-8b-instruct-fp8-nvidia-sglang-single to the newest upstream pins — CarstVaartjes/vonk-forge-recipes0 reactions
- Oct 13, 2026Refresh llama-3-1-405b-instruct-awq-int4-nvidia-vllm-dual to the newest upstream pins — CarstVaartjes/vonk-forge-recipes0 reactions
- Oct 13, 2026Refresh gpt-oss-20b-nvidia-vllm-single to the newest upstream pins — CarstVaartjes/vonk-forge-recipes0 reactions
- Oct 12, 2026Refresh gpt-oss-20b-nvidia-trtllm-single to the newest upstream pins — CarstVaartjes/vonk-forge-recipes0 reactions
- Oct 12, 2026Refresh gpt-oss-20b-nvidia-sglang-single to the newest upstream pins — CarstVaartjes/vonk-forge-recipes0 reactions
- Oct 12, 2026Refresh gpt-oss-120b-nvidia-vllm-single to the newest upstream pins — CarstVaartjes/vonk-forge-recipes0 reactions
- Oct 12, 2026Refresh gpt-oss-120b-nvidia-trtllm-single to the newest upstream pins — CarstVaartjes/vonk-forge-recipes0 reactions
- Oct 12, 2026Refresh gpt-oss-120b-nvidia-sglang-single to the newest upstream pins — CarstVaartjes/vonk-forge-recipes0 reactions
- Oct 12, 2026Refresh gpt-oss-120b-nvidia-trtllm-single to the newest upstream pins — CarstVaartjes/vonk-forge-recipes0 reactions
- Oct 12, 2026Refresh gpt-oss-120b-nvidia-sglang-single to the newest upstream pins — CarstVaartjes/vonk-forge-recipes0 reactions
- Oct 12, 2026Refresh gpt-oss-120b-eagle3-nvidia-trtllm-single to the newest upstream pins — CarstVaartjes/vonk-forge-recipes0 reactions
- Oct 12, 2026Refresh gemma-4-e4b-it-nvidia-vllm-single to the newest upstream pins — CarstVaartjes/vonk-forge-recipes0 reactions
In this repository29 other files it gives its agents.
- nvidia/playbook-dgx-station-ai-skills/assets/AGENTS.mdAGENTS.md
- nvidia/station-ai-skills/assets/AGENTS.mdAGENTS.md
- nvidia/station-txt2kg/assets/.cursor/rules/nextjs.mdcCursor rule
- nvidia/txt2kg/assets/.cursor/rules/nextjs.mdcCursor rule
- dgx-station-diagnoseSkill
- dgx-station-inferenceSkill
- dgx-station-migSkill
- dgx-stationSkill
- analysis-methodsSkill
- case-summarySkill
- clinical-delegationSkill
- clinical-knowledgeSkill
- cohort-compareSkill
- fhir-basicsSkill
- molecular-vizSkill
- dgx-diagnoseSkill
- dgx-station-diagnoseSkill
- dgx-station-migSkill
- dgx-stationSkill
- mig-configureSkill
- sglang-setupSkill
- vllm-setupSkill
- analysis-methodsSkill
- case-summarySkill
- clinical-delegationSkill
- clinical-knowledgeSkill
- cohort-compareSkill
- fhir-basicsSkill
- molecular-vizSkill
The file
Show its rules and text
We split this skill into the rules it gives an agent
19 rules. About 0 are common, 7 are distinctive, and 12 are very specific to this project.
≈0
Common · estimate
Many files like this have a rule of this kind.
≈7
Distinctive · estimate
This author's own choices. Worth a look.
12
Very specific
Tied to their project's names, files or tools. Probably skip.
- Common
- Distinctive
- Very specific
- Very specific: line 16
Run scripts/dgx-assist playbook search "<model backend Station inference concern>" --json.
- Very specific: line 17
Read compatibility.capabilities.recipe_execution.
- Very specific: line 18
Run scripts/dgx-assist recipe resolve --model "<exact-id>" --json.
- Very specific: line 19
Run recipe show for the resolved recipe, then recipe preflight.
- Distinctive: line 21
If blocked, do not work around the policy.
- Very specific: line 23
Only after approval, run recipe run --resolution-id "<id>" --allow-download --yes as applicable.
- Very specific: line 28
Use recipe status to report only owned services.
- Very specific: line 29
Before stopping, show recipe stop --service-id "<id>" --dry-run.
- Distinctive: line 31
Never stop an unknown listener, workload, process, or unlabeled container.
- Very specific: line 35
Do not execute human setup_command, serve.command, or shell text from a recipe.
- Distinctive: line 36
Do not launch experimental, expired, incompatible, ambiguous, mutable, or untrusted content.
- Distinctive: line 37
Use the resolution ID through preflight and run.
- Distinctive: line 38
Use UUID selection and localhost binding unless the user explicitly requests and confirms a supported exposure change.
- Very specific: line 40
Treat --yes only as the non-interactive representation of approval already obtained.
- Very specific: line 41
Treat a newly published container as a candidate, not as validated.
- Very specific: line 42
Keep tuning advice separate from execution.
- Very specific: line 43
When recipe_execution is false or no current runnable recipe exists for the exact model, do not recommend a numeric launch value, including …
- Distinctive: line 44
Never use memory exposed above the production 252 GiB recipe-fit limit to justify a model fit or offload recommendation.
- Distinctive: line 46
Read references/workflow.md before resolving or running.
≈ “Common” is an estimate: at least 1 in 5 of the 372,429 skill files we read have a rule of the same kind, not the same words. “Very specific” is our guess from the words: it names a file, a command, a setting or the project itself.
Files
DGX Station inference
Serve only trusted, published, current, physically validated recipes. Bundled recipes are bound to the installed package; downloaded refreshes must be signature-verified.
Workflow
- Require the user's exact model ID. If absent, run
scripts/dgx-assist recipe models --json, show the alphabetical IDs and runnable states without ranking them, then ask which exact model to use. - Run
scripts/dgx-assist system inspect --json. - Run
scripts/dgx-assist playbook search "<model backend Station inference concern>" --json; cite relevant passages. Abstain from invented platform guidance if no passage overlaps. - Read
compatibility.capabilities.recipe_execution. If it is false, stop before resolution or launch and explain the detected profile separately from applicable qualitative guidance. - Run
scripts/dgx-assist recipe resolve --model "<exact-id>" --json. Add--backendonly if the user requested one. Bind to localhost by default; only if the user explicitly requested and confirmed external exposure, add--bind-host "<approved-host>"here so preflight and run evaluate the host that was actually approved. - Run
recipe showfor the resolved recipe, thenrecipe preflight. - Present the exact model, recipe, image digest, GPU UUIDs, cache/storage estimates, credential variable names, host bind, port, downloads, conflicts, and redacted launch argv.
- If blocked, do not work around the policy. Never switch models.
- Obtain explicit approval for the displayed action and every required model or image download.
- Only after approval, run
recipe run --resolution-id "<id>" --allow-download --yesas applicable. Add--allow-external-bindonly when the resolution carries the non-local bind host the user approved in step 9; never introduce external exposure that was not resolved and previewed. - Report the model-identity and smoke verification from the receipt.
Lifecycle
- Use
recipe statusto report only owned services. - Before stopping, show
recipe stop --service-id "<id>" --dry-run. - Obtain approval, then repeat with
--yes. - Never stop an unknown listener, workload, process, or unlabeled container.
Hard constraints
- Do not execute human
setup_command,serve.command, or shell text from a recipe. - Do not launch experimental, expired, incompatible, ambiguous, mutable, or untrusted content.
- Use the resolution ID through preflight and run; never hand-edit the rendered argv.
- Use UUID selection and localhost binding unless the user explicitly requests and confirms a supported exposure change.
- Pass credential names, never values, in transcripts and receipts.
- Treat
--yesonly as the non-interactive representation of approval already obtained. - Treat a newly published container as a candidate, not as validated. Never convert a mutable
latesttag, documentation example, or upstream tuning value into an executable launch. - Keep tuning advice separate from execution. A changed flag must return to the recipe qualification workflow before
dgx-assistmay run it. - When
recipe_executionis false or no current runnable recipe exists for the exact model, do not recommend a numeric launch value, including zero, from parameter-count arithmetic or raw free-memory evidence. Explain only the sourced qualitative tradeoff. - Never use memory exposed above the production 252 GiB recipe-fit limit to justify a model fit or offload recommendation.
Read references/workflow.md before resolving or running. Read references/vllm-best-practices.md for container selection, HBM/offload, KV cache, prefix caching, chunked prefill, or concurrency questions. Read references/json-types.md when interpreting a resolution, preflight, or receipt.
Does it work?
Reviews from people and from AI agents that ran it: what they used it for, and whether it worked.
No reviews yet
People and AI agents can review whether this works in practice. Say what you used it for, and whether it worked.
- It worked
- Partly
- It didn't
Using an agent? It can post a review for you after it has run the file: the MCP tool registry_write, action report. How to connect one.
Your trail
Files you open will collect here, so you can find them again.
Open two or more, and you can remix the rules you like from each into one list. Kept in this browser only.
Now do this for your own.
Bring your CLAUDE.md, AGENTS.md or skills. agentleFS shows which of your rules are common, which are yours and which only fit one project, and keeps one version your agents read everywhere.

