skills.rest
MLT-OSS
MLT-OSSOFFICIAL
Explore all skills by this author →
WebsiteGitHub

weights-and-biases

Track ML experiments, run hyperparameter sweeps, and manage model artifacts with Weights & Biases.

14|5|Updated Apr 9, 2026
Search Tags:#mlops#machine learning#pytorch#hyperparameter tuning#experiment tracking#wandb#model registry
One-click install
npx skills add https://github.com/MLT-OSS/hermes-agent-go --skill weights-and-biases-mlt-oss
Or copy as Structured Prompt for Agent▼
Please help me install this Agent Skill.
Skill: weights-and-biases
Source: https://github.com/MLT-OSS/hermes-agent-go/tree/main/skills/mlops/evaluation/weights-and-biases
Command: npx skills add https://github.com/MLT-OSS/hermes-agent-go --skill weights-and-biases-mlt-oss

SYSTEM DOCUMENTATION & REQUIREMENTS

💡 This Skill requires wandb, and includes references (resource) components.

What problem does it solve? Machine learning teams lose track of experiments, hyperparameters, and model versions when training runs are scattered across notebooks and scripts. This Skill provides complete guidance for using Weights & Biases (W&B) to log metrics automatically, visualize training in real time, optimize hyperparameters with sweeps, and version datasets and models with full lineage. ## Core Features & Use Cases - Experiment Tracking: Initialize runs with wandb.init, log metrics, configs, media, and system stats, and compare runs in real-time dashboards. - Hyperparameter Sweeps: Run grid, random, or Bayesian optimization searches with early termination (Hyperband) and parallel agents across GPUs. - Artifacts & Model Registry: Version datasets and models with automatic lineage tracking, aliases for deployment stages, and a central model registry. - Framework Integrations: Native support for PyTorch, PyTorch Lightning, HuggingFace Transformers, Keras/TensorFlow, Fast.ai, XGBoost, and LightGBM. - Use Case: A data scientist fine-tuning a BERT model can launch a Bayesian sweep over learning rate and batch size, let W&B terminate weak runs early, and promote the best checkpoint to a production model registry. ## Quick Start Set up a W&B tracked training run for my PyTorch script that logs loss and accuracy each epoch and saves the final model as an artifact.

Frequently Asked Questions about weights-and-biases

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

FAQPage Schema
How do I track PyTorch training experiments with Weights & Biases?▼

Call wandb.init with your project name and config, then use wandb.log inside your training loop to record loss, accuracy, and other metrics. Use wandb.watch to log gradients and save the final model with wandb.save or as an artifact.

How to run hyperparameter sweeps with wandb?▼

Define a sweep config with a search method (grid, random, or bayes), a target metric, and parameter distributions, then create it with wandb.sweep. Launch agents with wandb.agent pointing to your training function, optionally running multiple agents in parallel across GPUs.

Does W&B integrate with HuggingFace Transformers?▼

Yes, set report_to="wandb" in TrainingArguments and the Trainer automatically logs metrics, evaluation results, and checkpoints to W&B. You can also add custom WandbCallback subclasses for additional logging.

What is the difference between W&B artifacts and the model registry?▼

Artifacts are versioned files such as datasets or model checkpoints with automatic lineage tracking. The model registry is a curated collection where you link model artifacts and manage deployment stages using aliases like staging and production.

Can I use wandb offline without an internet connection?▼

Yes, set the WANDB_MODE environment variable to offline and runs are saved locally. Sync them later to the W&B server using the wandb sync command on the run directory.

Which sweep search method should I use for expensive training runs?▼

Bayesian optimization is recommended for expensive runs because it learns from previous trials and focuses on promising parameter regions. Combine it with Hyperband early termination to stop underperforming runs and save compute.

Related Skills

Explore More Data & Analytics Skills→
CHENyiru3CHENyiru3

datasets

Load and preprocess datasets from Hugging Face Hub or local sources.

Community
Advanced
KentoShimizuKentoShimizu
7

mlops-monitoring-drift

Detect data drift, concept drift, and quality degradation in production ML pipelines.

Community
Advanced
NVIDIANVIDIA
14.5k

perf-analysis

Coordinates GPU performance profiling delegation, bottleneck classification, and structured analysis report generation.

Official
Intermediate
hhhi21ghhhi21g

experiment-tracking-swanlab

Track and visualize machine learning experiments with local or self-hosted dashboards.

Community
Advanced