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 optimize-model-precisiongit 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/optimize-model-precision)<a href="https://agentmods.dev/skills/nvidia/tensorrt-model-connect/optimize-model-precision"><img src="https://agentmods.dev/badge/skills/nvidia/tensorrt-model-connect/optimize-model-precision/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/optimize-model-precision"><img src="https://agentmods.dev/badge/skills/nvidia/tensorrt-model-connect/optimize-model-precision.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 | $0.00034 | $0.00464 |
| Opus 5.5 | $0.00014 | $0.00186 |
| Sonnet 5.5 | $0.00007 | $0.00093 |
| Haiku 4.5 | $0.00003 | $0.00046 |
Grade A, and why
optimize-model-precision 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 10d 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.
What it actually says
Optimize Model Precision
Name the objective first: bundle size, device memory, setup time, Task API latency, throughput, or another measured family outcome. The best candidate is the lowest-cost option that still passes the existing family contract.
Establish a baseline
Use one family manifest and testcase. Record the repository and checkpoint revisions, hardware, build request, seed, input, bundle path, correctness result, warmups, iterations, and measured boundary.
Build the current declared configuration and run its E2E before exploring
alternatives. The public build interface exposes direct values for
--precision, --quantization, and repeated --fp32-layer selections.
Attempt only values the owning family implements.
python -m tensorrt_model_connect build <model-id-or-path> \
--revision <revision> --precision <fp32-or-fp16-or-bf16> \
--output <candidate.bundle>
trtmc inspect <candidate.bundle>
Change one effective value per attempt. If a candidate fails, localize the first mismatch in the owner family before changing another value. Do not add cross-family precision policy or weaken a threshold, dataset, seed, or oracle.
Qualify a candidate
Run the exact family E2E testcase with the candidate configuration. Only after it passes, compare Task API performance with matched inputs and protocol:
trtmc-bench run --model <model> --case <case> \
--bundle <candidate.bundle> --no-build \
--warmup 3 --iterations 10 --runtime-root <runtime-root> \
--output <result-dir>
Use trtmc-aiperf-qual (apps/aiperf_qual) when the claim targets a release comparison against the native model.
Report every attempt in a compact table, including effective request values,
correctness, measured objective, confounders, and unrun targets. It is valid
for the original configuration to remain the best qualified result.
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.
- 10d ago Changed 32d1d58cf90f
- 1mo ago Changed 4b2595ba84c7
- 1mo ago Changed · -138 lines · -30 tokens per session 497ac00e320b
- 1mo ago First seen · 190 lines · 64 tokens per session scan A 6a53f3dc414b
optimize-model-precision is a skill published in the GitHub repository NVIDIA/TensorRT-Model-Connect (281 stars, last pushed today), licensed Apache-2.0. It adds 34 tokens to every session and 464 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.
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.
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.