debug-trt-mismatch

debug-trt-mismatch is a skill for Claude Code, Codex from NVIDIA/TensorRT-Model-Connect. It costs 61 tokens per session (1,579 once invoked), scanned A, original, Apache-2.0.

A troubleshooting workflow for cases where TensorRT model output differs from a reference model. TensorRT is a system for building and running optimized machine-learning models, and the workflow narrows the problem to the first operation that produces a different result.

In plain words
What is it for?
Use it to reproduce a mismatch, compare reference and TensorRT boundaries, inspect model-specific operations, and identify the smallest responsible change.
Why use it?
It preserves the exact model, inputs, runtime, and environment while investigating, so a mismatch is not hidden by loosening correctness checks. It also provides a path for working when the required GPU or TensorRT setup is elsewhere.

Skill for Claude CodeCodex

Install

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.

agentmods
npx agentmods add skills/nvidia/tensorrt-model-connect/debug-trt-mismatch
Any agent
npx skills add NVIDIA/TensorRT-Model-Connect --skill debug-trt-mismatch
Clone the repo
git clone --depth 1 https://github.com/NVIDIA/TensorRT-Model-Connect

Made for: Claude Code, Codex.

Wrote 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.

agentmods badge for debug-trt-mismatch

README.md
[![agentmods](https://agentmods.dev/badge/skills/nvidia/tensorrt-model-connect/debug-trt-mismatch.svg)](https://agentmods.dev/skills/nvidia/tensorrt-model-connect/debug-trt-mismatch)
Your own site
<a href="https://agentmods.dev/skills/nvidia/tensorrt-model-connect/debug-trt-mismatch"><img src="https://agentmods.dev/badge/skills/nvidia/tensorrt-model-connect/debug-trt-mismatch.svg" alt="Measured on agentmods" height="20"></a>
Per session 61 Skills are progressive disclosure: only the name and description are preloaded; the body loads when the skill is used.
When invoked 1,579 The whole file, excluding the scripts and references it only reads on demand.
Security scan A 0 findings. Scan, not verified.
Origin original No closer match found in the catalogue.
Token cost

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.

ModelPer sessionOnce invoked
Fable 5 $0.00061 $0.01579
Opus 5 $0.00030 $0.00790
Sonnet 5 $0.00012 $0.00316
Haiku 4.5 $0.00006 $0.00158

Measured 3d ago against content hash 2a27694da85e, method: parsed. Prices are Anthropic first-party input rates as of 2026-08-30, from the pricing page.

Security

Grade A, and why

debug-trt-mismatch 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 3d 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.

plugins/trtmc-agent-skills/skills/debug-trt-mismatch/SKILL.md · 204 lines

How it starts

The opening of the file, as written. The whole thing — 204 lines — stays where its author put it; the contents beside it link to each section on GitHub.

Debug TRT Mismatch

Goal

Find the first boundary where TensorRT and the declared reference disagree. Preserve the failing workload, sampling settings, model revision, bundle, and runtime strategy while narrowing the problem. Do not hide a mismatch by loosening validation thresholds.

Preflight

Record the exact revision and environment:

git rev-parse HEAD
nvidia-smi --query-gpu=name,driver_version --format=csv,noheader
python3 -c "import tensorrt as trt; print(trt.__version__)"
PYTHONPATH=python:. python3 -c "import tensorrt_model_connect; print('package: OK')"
test -x ./build/trtmc

If the local checkout does not have the required GPU or TensorRT environment, use the existing team container rather than changing the investigation:

docker ps -a --filter "name=trtmc-dev-gb300" --format "{{.Names}} {{.Status}}"
./scripts/bootstrap_workspace.sh --id <team-id> --branch "$(git branch --show-current)" --detach

Run later commands inside trtmc-dev-gb300-<team-id> when needed.

Reproduce Through The Owned Validation Path

List model-first workloads and dry-run the failing one:

PYTHONPATH=python:. python3 tools/trtmc_validate.py --list
PYTHONPATH=python:. python3 tools/trtmc_validate.py \
  <model> <workload> \
  --dry-run \
  --output /tmp/trtmc-validation

The model binding in tests/validation/model_workloads.yaml, workload contract in tests/validation/workloads.yaml, family manifests, and their sidecars are the source of truth for inputs, sampling, and comparison gates. Read tools/validation/README.md before changing the engine: the persisted task_eval artifact key remains intentionally stable even though executable code moved to tools/validation/. Keep reference generation and comparison separate in the report; an execution failure is not a numerical mismatch.

Route By Model Capability

Inspect the model's Python, C++, and E2E MODEL.toml entries before selecting a debugger:

Model path First focused tool
Decoder family with decoder_debug validation profile tools/diff_logits.py
Decoder layer localization tools/diff_layers.py
Vision-language tools/diff_vl.py
Audio/Bark tools/diff_audio.py
Diffusion tools/debug_diffusion_pipeline.py
Python runner versus C++ tools/test_runner_parity.py

Read the full file on GitHub · 204 lines

Files

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.

Changes

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.

  1. 3d ago First seen · 204 lines · 61 tokens per session scan A 2a27694da85e

Subscribe to this mod's changes

debug-trt-mismatch is a skill published in the GitHub repository NVIDIA/TensorRT-Model-Connect (188 stars, last pushed 3d ago), licensed Apache-2.0. It adds 61 tokens to every session and 1,579 once invoked, about $0.0003 per session on Opus 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.

Related

Other skills, from other repositories

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.

graphsignal/graphsignal-profiler · 74 tokens

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.

Orchestra-Research/AI-Research-SKILLs · 39 tokens

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.

Orchestra-Research/AI-Research-SKILLs · 69 tokens

mlflow

Track ML experiments, manage model registry with versioning, deploy models to production, and reproduce experiments with MLflow - framework-agnostic ML lifecycle platform.

Orchestra-Research/AI-Research-SKILLs · 33 tokens

huggingface-hub

Hugging Face Hub — model discovery, download, inference, and upload.

furkangonel/cowrangler · 22 tokens

ml-training

Machine learning model training authority — classifier and regressor training with scikit-learn and PyTorch, feature engineering, cross-validation, hyperparameter tuning, fine-tuning transformer models with HuggingFace, dataset splits, loss functions, learning rate schedules, and reproducible training runs.

LuuOW/meridian-mcp · 58 tokens