skills.rest
DeepLink-org
DeepLink-orgOFFICIAL
Explore all skills by this author →
WebsiteGitHub

nvidia-cv-pretrain

Benchmarks image classification model training performance on NVIDIA GPUs using mmpretrain.

7|1|Updated Apr 17, 2026
Search Tags:#docker#benchmark#performance evaluation#distributed training#image classification#nvidia gpu#mmpretrain
One-click install
npx skills add https://github.com/DeepLink-org/DeepEval-Skills --skill nvidia-cv-pretrain-deeplink-org
Or copy as Structured Prompt for Agent▼
Please help me install this Agent Skill.
Skill: nvidia-cv-pretrain
Source: https://github.com/DeepLink-org/DeepEval-Skills/tree/main/skills/NVIDIA/cv/nvidia-cv-pretrain
Command: npx skills add https://github.com/DeepLink-org/DeepEval-Skills --skill nvidia-cv-pretrain-deeplink-org

SYSTEM DOCUMENTATION & REQUIREMENTS

💡 This Skill requires torch, mmengine, onedl-mmcv, onedl-mmpretrain, and includes scripts (resource) components.

What problem does it solve? It automates the end-to-end performance evaluation of CV pretraining and classification models on NVIDIA GPUs, eliminating the manual work of launching containers, running distributed training, and parsing iteration-time metrics. ## Core Features & Use Cases - Automated Benchmark Workflow: Starts a Docker container, runs batch_pretrain.sh with torch.distributed, and collects training logs. - Multi-Model and Multi-Precision Testing: Supports resnet50, inception_v3, seresnet50, mobilenet_v2, shufflenet_v2, densenet121, swin_large, and efficientnet_b2 with fixed fp16 and fp32 runs. - Structured Metrics Output: Aggregates avg_iter_time, data_time, and op_time per model/GPU/precision combination into eval_result.json. - Use Case: Ask the agent to test resnet50 training performance on 8 GPUs, and it launches the container, runs both precisions, and returns the AVG_ITER_TIME results. ## Quick Start Ask the agent to test resnet50 classification training performance on NVIDIA GPUs with 8 cards and collect the AVG_ITER_TIME metrics.

Frequently Asked Questions about nvidia-cv-pretrain

High-intent search queries and answers about installing and using this skill.

FAQPage Schema
How do I benchmark ResNet50 training performance on NVIDIA GPUs?▼

Run the batch_pretrain.sh script inside the nvidia-cv Docker container with CV_PRE_MODELS=resnet50 and CV_PRE_NGPU set to your GPU count. It trains with both fp16 and fp32 and writes avg_iter_time results to /workspace/logs/eval_result.json.

Which classification models does this mmpretrain benchmark support?▼

It supports resnet50, inception_v3, seresnet50, mobilenet_v2, shufflenet_v2, densenet121, swin_large, and efficientnet_b2. Each model maps to a predefined mmpretrain config, and multiple models can be passed as a comma-separated list.

Can I run the training benchmark with only fp16 precision?▼

No, the precision is fixed to run both fp16 and fp32 for every model. The script does not accept a user-provided precision parameter or environment variable to select a single precision.

Why is AVG_ITER_TIME missing from my training logs?▼

AVG_ITER_TIME is missing when the project config does not enable the metric output or when you read an old log file. Set CV_PRE_RUN_MARKER before running so only logs created during the current run are parsed.

What environment setup is required for the mmpretrain benchmark?▼

You need the nvidia-cv Docker image, a project root containing onedl-mmpretrain source with configs and tools/train.py, and an ImageNet dataset directory with train and val splits. These are mounted to /workspace/code and /workspace/datasets/imagenet in the container.

Related Skills

Explore More Software Engineering Skills→
mnazaalmnazaal

dev-ml-infra

Configure reproducible ML experiment infrastructure with hydra-zen configs, tracker seams, and smoke tests.

Community
Advanced
InverterNetworkInverterNetwork

llama-cpp

Discovers GGUF models from Hugging Face Hub and configures llama.cpp inference servers.

Official
Advanced
OpenLAIROpenLAIR
1.1k

aris-system-profile

Profiles scripts, processes, GPUs, memory, and interconnects to produce structured performance reports.

Official
Advanced
doanchienthangdevdoanchienthangdev
4

ml-serving-optimization

Optimize ML model inference with dynamic batching, TorchScript, ONNX Runtime, and TensorRT.

Community
Advanced