NVIDIA/TensorRT-Model-Connect is a collection of C++ reference implementations for deploying supported Hugging Face models with NVIDIA TensorRT, a system that runs trained models to produce inference results. It is for developers who want to build and run supported models or evaluate model integrations through TensorRT. The catalogue entries provide skills and instructions for using this model deployment workflow.
Getting it into your agent
One page per mod, every tool's command on it. A separate URL per tool would split the same page into five that compete with each other.
npx skills add NVIDIA/TensorRT-Model-Connect --skill profile-modelgit clone --depth 1 https://github.com/NVIDIA/TensorRT-Model-ConnectWrote this? Show the measurements
A badge with what this costs and how it scanned, read live from this page, so it follows the numbers instead of freezing them. Markdown for a README, HTML for a documentation site or a project page.
[](https://agentmods.dev/skills/nvidia/tensorrt-model-connect/profile-model)<a href="https://agentmods.dev/skills/nvidia/tensorrt-model-connect/profile-model"><img src="https://agentmods.dev/badge/skills/nvidia/tensorrt-model-connect/profile-model/github.svg" alt="Measured on agentmods" height="20"></a>Or the 80×15 button, for a site that already has a row of RSS and ATOM ones. Only the verdict fits; the numbers stay here.
<a href="https://agentmods.dev/skills/nvidia/tensorrt-model-connect/profile-model"><img src="https://agentmods.dev/badge/skills/nvidia/tensorrt-model-connect/profile-model.svg" alt="Reviewed on agentmods" width="80" height="20"></a>- NVIDIA SkillSpector pass
What it costs to keep this loaded
Counted locally with the o200k_base tokenizer, which is exact for GPT models; Claude uses its own tokenizer and its counts differ. Treat this as one consistent yardstick across the catalogue rather than a bill. Prices are per million input tokens.
| Model | Per session | Once invoked |
|---|---|---|
| Fable 5.1 | $0.00030 | $0.00669 |
| Opus 5.5 | $0.00012 | $0.00268 |
| Sonnet 5.5 | $0.00006 | $0.00134 |
| Haiku 4.5 | $0.00003 | $0.00067 |
Grade A, and why
profile-model scanned grade A with 0 findings against 26 rules in 11 categories — prompt injection, anti-refusal, data exfiltration, privilege escalation, supply chain, agent snooping, system-prompt leakage, SSRF and excessive agency — measured 9d ago.
A static scan of the body, not an audit. Every finding is printed with the line that produced it so you can judge whether it matters here. A mod is markdown that instructs an agent; that is exactly why what it instructs is worth reading.
Nothing flagged
None of the 26 patterns this scan looks for appear in this file: no shell pipes, no recursive deletes, no credential paths, no hidden text, no instruction-override or anti-refusal phrasing, no agent-config snooping. That is not a guarantee, it is the absence of the things that are checkable.
How it starts
The opening of the file, as written. The whole thing — 68 lines — stays where its author put it; the contents beside it link to each section on GitHub.
Profile a Model
Choose the evidence level before running:
- For one model or testcase, use
trtmc-bench. - For a release comparison against the native model, use
trtmc-aiperf-qual(apps/aiperf_qual).
The public benchmark boundary is public_task_call_wall. It excludes bundle
building, process startup, warmup, report generation, and bundle loading.
Current public tooling does not provide layer attribution; do not infer a
kernel or layer bottleneck from Task API wall time alone.
Before timing
Run the owning family correctness testcase on the exact bundle. Record the repository and checkpoint revisions, GPU and driver, runtime root, family DSO, operation, input, seed, warmups, iterations, and synchronization boundary.
For a focused run:
trtmc-bench run --model <model> --case <case> \
--bundle <bundle> --no-build \
--warmup 3 --iterations 10 --runtime-root <runtime-root> \
--output <result-dir>
Use the same bundle kind, hardware, input, seed, warmups, iterations, and runtime policy for before/after comparisons. If any differ, report the confounder instead of one causal speedup percentage.
For release evidence, run the exact profile with
trtmc-aiperf-qual run --profile <profile> --environment <environment> --out <result-dir>
(see apps/aiperf_qual/README.md).
Lead the report with correctness and evidence level. Include exact commands, p50 and other suite-owned statistics, measurement scope, limitations, and any target not run.
HTTP text load evidence
For text serving measurements, reuse the persistent trtmc-server through
python3 -m trtmc_aiperf_qual text-profile --environment <environment> --config <text-config> --out <result-dir> with PYTHONPATH=apps/aiperf_qual. Consult
website/docs/user-guides/profile-text-with-aiperf.md and the checked-in
apps/aiperf_qual/config/text/ examples. Keep the owning family's exact-bundle
correctness checks before timing; HTTP success does not replace them.
Require incremental capability before interpreting TTFT or inter-token
latency. Nonstreaming native records use public_task_call_wall; streaming
records include relay/backpressure and have a different scope. Preserve warmup
and failures when joining AIPerf exports to server records with X-Request-ID.
Native prompt counts are unavailable; client tokenizer counts are estimates.
AIPerf request latency ends at the last content response. Inspect the client
request lifecycle and server handler timing as well: terminal delivery or
cleanup can delay the next request without appearing in content latency.
The optional sequential reference comparison requires matching payloads,
outputs, token counts, precision and Task boundaries. It does not change the
qualification criteria or establish speedup from HTTP latency alone.
What ships with it
1 file beside SKILL.md in the same directory: the scripts, references and assets a skill reads on demand. Not counted in the per-session cost; read them before you install if any of them is executable.
What this file has done since we first saw it
Hashed on every crawl. A supply-chain change to an agent config is a question of when, not whether, so the history is kept rather than the latest state alone.
- 9d ago Changed · -5 lines 8e871b462850
- 17d ago Changed · +22 lines 54a52ed883e2
- 1mo ago Changed 14327040f8f4
- 1mo ago Changed · -145 lines · -17 tokens per session ffc6492c6d1c
- 1mo ago First seen · 196 lines · 47 tokens per session scan A c84b25ff46b6
profile-model is a skill published in the GitHub repository NVIDIA/TensorRT-Model-Connect (280 stars, last pushed yesterday), licensed Apache-2.0. It adds 30 tokens to every session and 669 once invoked, about $0.0001 per session on Opus 5.5. A static security scan graded it A with 0 findings. No closer match exists in the catalogue, so it is treated as the original; first seen 2026-08-30.
Other skills, from other repositories
huggingface-transformers
Use when running or fine-tuning Hugging Face Transformers models in Python — pipeline() inference for text classification and generation, loading weights with AutoModel/AutoTokenizer, devicemap and dtype memory management, tokenizer usage, Trainer fine-tuning, and BGE embeddings for Chinese RAG. Covers transformers…
graphsignal-profiler
Set up the Graphsignal Profiler for inference workloads — vLLM, SGLang, PyTorch, and dstack services. Use when the user wants GPU profiling, tracing, or monitoring for inference, asks about graphsignal-run or graphsignal.watch(), or asks about CUPTI / Prometheus / OTLP setup.
rocketride-debugging-pipelines
Use when a RocketRide pipeline run failed, errored, or produced wrong/empty output, and you need to find the failing node and fix it. Reads run status and execution traces, diagnoses the cause, and routes back to design or configuration. Also use directly when asked to debug a run.
cutedsl_megamoe
The explicit GenPhase prototype is independently owned by ../nextcutedslmegamoe. This legacy package lazily re-exports its six public GenPhase names. Follow that package's update workflow for GenPhase changes; keep this older snapshot separate. The default backend heuristics do not select GenPhase automatically.
huggingface-accelerate
Simplest distributed training API. 4 lines to add distributed support to any PyTorch script. Unified API for DeepSpeed/FSDP/Megatron/DDP. Automatic device placement, mixed precision (FP16/BF16/FP8). Interactive config, single launch command. HuggingFace ecosystem standard.
weights-and-biases
Track ML experiments with automatic logging, visualize training in real-time, optimize hyperparameters with sweeps, and manage model registry with W&B - collaborative MLOps platform.