[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"skill:01M71JB4FW0GNYSCT5F3CXY2HT":3,"skill-same:01M71JB4FW0GNYSCT5F3CXY2HT":96},{"skillNo":4,"name":5,"nameZh":6,"description":7,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":6,"repositoryNo":12,"starsCount":13,"licenseSpdxId":14,"licenseFileUrl":6,"sourceHtmlUrl":15,"categories":16,"keywords":22,"publishedAt":43,"linkOnly":44,"markdownContent":45,"markdownBody":46,"frontMatter":47,"refName":61,"commitSha":62,"skillDirectory":63,"forksCount":64,"repositoryHtmlUrl":65,"repositoryHomepageUrl":6,"ownerHtmlUrl":66,"ownerBlogUrl":6,"ownerDescription":6,"updatedAt":67,"install":68,"resources":81},"01M71JB4FW0GNYSCT5F3CXY2HT","weights-and-biases",null,"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","xuanxuan1983\u002Fhermes-skills-backup","xuanxuan1983","https:\u002F\u002Favatars.githubusercontent.com\u002Fu\u002F151701653?v=4","01M71HCZNB5H19QJ6YBF98W2WH","01M71HCZP14T0921BD0TC9EPD3",1,"MIT","https:\u002F\u002Fgithub.com\u002Fxuanxuan1983\u002Fhermes-skills-backup\u002Fblob\u002Fmain\u002Fskills\u002Fmlops\u002Fevaluation\u002Fweights-and-biases\u002FSKILL.md",[17],{"categoryNo":18,"categoryName":19,"taxonomyType":20,"taxonomyTypeDesc":21},"01JBSEEDCAT000000000000013","分析与可视化","SCENARIO","使用场景",[23,24,25,26,27,28,29,30,31,32,33,34,35,36,37,38,39,40,41,42],"weights and biases","track","visualize","Hyperparameter Tuning","logging","experiment","automatic","experiments","registry","tuning","optimize","Model Registry","Real-Time Visualization","Experiment Tracking","WandB","pytorch","training","time","model","visualization","2026-11-10 10:12:30",false,"---\nname: weights-and-biases\ndescription: 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\nversion: 1.0.0\nauthor: Orchestra Research\nlicense: MIT\ndependencies: [wandb]\nmetadata:\n  hermes:\n    tags: [MLOps, Weights And Biases, WandB, Experiment Tracking, Hyperparameter Tuning, Model Registry, Collaboration, Real-Time Visualization, PyTorch, TensorFlow, HuggingFace]\n\n---\n\n# Weights & Biases: ML Experiment Tracking & MLOps\n\n## When to Use This Skill\n\nUse Weights & Biases (W&B) when you need to:\n- **Track ML experiments** with automatic metric logging\n- **Visualize training** in real-time dashboards\n- **Compare runs** across hyperparameters and configurations\n- **Optimize hyperparameters** with automated sweeps\n- **Manage model registry** with versioning and lineage\n- **Collaborate on ML projects** with team workspaces\n- **Track artifacts** (datasets, models, code) with lineage\n\n**Users**: 200,000+ ML practitioners | **GitHub Stars**: 10.5k+ | **Integrations**: 100+\n\n## Installation\n\n```bash\n# Install W&B\npip install wandb\n\n# Login (creates API key)\nwandb login\n\n# Or set API key programmatically\nexport WANDB_API_KEY=your_api_key_here\n```\n\n## Quick Start\n\n### Basic Experiment Tracking\n\n```python\nimport wandb\n\n# Initialize a run\nrun = wandb.init(\n    project=\"my-project\",\n    config={\n        \"learning_rate\": 0.001,\n        \"epochs\": 10,\n        \"batch_size\": 32,\n        \"architecture\": \"ResNet50\"\n    }\n)\n\n# Training loop\nfor epoch in range(run.config.epochs):\n    # Your training code\n    train_loss = train_epoch()\n    val_loss = validate()\n\n    # Log metrics\n    wandb.log({\n        \"epoch\": epoch,\n        \"train\u002Floss\": train_loss,\n        \"val\u002Floss\": val_loss,\n        \"train\u002Faccuracy\": train_acc,\n        \"val\u002Faccuracy\": val_acc\n    })\n\n# Finish the run\nwandb.finish()\n```\n\n### With PyTorch\n\n```python\nimport torch\nimport wandb\n\n# Initialize\nwandb.init(project=\"pytorch-demo\", config={\n    \"lr\": 0.001,\n    \"epochs\": 10\n})\n\n# Access config\nconfig = wandb.config\n\n# Training loop\nfor epoch in range(config.epochs):\n    for batch_idx, (data, target) in enumerate(train_loader):\n        # Forward pass\n        output = model(data)\n        loss = criterion(output, target)\n\n        # Backward pass\n        optimizer.zero_grad()\n        loss.backward()\n        optimizer.step()\n\n        # Log every 100 batches\n        if batch_idx % 100 == 0:\n            wandb.log({\n                \"loss\": loss.item(),\n                \"epoch\": epoch,\n                \"batch\": batch_idx\n            })\n\n# Save model\ntorch.save(model.state_dict(), \"model.pth\")\nwandb.save(\"model.pth\")  # Upload to W&B\n\nwandb.finish()\n```\n\n## Core Concepts\n\n### 1. Projects and Runs\n\n**Project**: Collection of related experiments\n**Run**: Single execution of your training script\n\n```python\n# Create\u002Fuse project\nrun = wandb.init(\n    project=\"image-classification\",\n    name=\"resnet50-experiment-1\",  # Optional run name\n    tags=[\"baseline\", \"resnet\"],    # Organize with tags\n    notes=\"First baseline run\"      # Add notes\n)\n\n# Each run has unique ID\nprint(f\"Run ID: {run.id}\")\nprint(f\"Run URL: {run.url}\")\n```\n\n### 2. Configuration Tracking\n\nTrack hyperparameters automatically:\n\n```python\nconfig = {\n    # Model architecture\n    \"model\": \"ResNet50\",\n    \"pretrained\": True,\n\n    # Training params\n    \"learning_rate\": 0.001,\n    \"batch_size\": 32,\n    \"epochs\": 50,\n    \"optimizer\": \"Adam\",\n\n    # Data params\n    \"dataset\": \"ImageNet\",\n    \"augmentation\": \"standard\"\n}\n\nwandb.init(project=\"my-project\", config=config)\n\n# Access config during training\nlr = wandb.config.learning_rate\nbatch_size = wandb.config.batch_size\n```\n\n### 3. Metric Logging\n\n```python\n# Log scalars\nwandb.log({\"loss\": 0.5, \"accuracy\": 0.92})\n\n# Log multiple metrics\nwandb.log({\n    \"train\u002Floss\": train_loss,\n    \"train\u002Faccuracy\": train_acc,\n    \"val\u002Floss\": val_loss,\n    \"val\u002Faccuracy\": val_acc,\n    \"learning_rate\": current_lr,\n    \"epoch\": epoch\n})\n\n# Log with custom x-axis\nwandb.log({\"loss\": loss}, step=global_step)\n\n# Log media (images, audio, video)\nwandb.log({\"examples\": [wandb.Image(img) for img in images]})\n\n# Log histograms\nwandb.log({\"gradients\": wandb.Histogram(gradients)})\n\n# Log tables\ntable = wandb.Table(columns=[\"id\", \"prediction\", \"ground_truth\"])\nwandb.log({\"predictions\": table})\n```\n\n### 4. Model Checkpointing\n\n```python\nimport torch\nimport wandb\n\n# Save model checkpoint\ncheckpoint = {\n    'epoch': epoch,\n    'model_state_dict': model.state_dict(),\n    'optimizer_state_dict': optimizer.state_dict(),\n    'loss': loss,\n}\n\ntorch.save(checkpoint, 'checkpoint.pth')\n\n# Upload to W&B\nwandb.save('checkpoint.pth')\n\n# Or use Artifacts (recommended)\nartifact = wandb.Artifact('model', type='model')\nartifact.add_file('checkpoint.pth')\nwandb.log_artifact(artifact)\n```\n\n## Hyperparameter Sweeps\n\nAutomatically search for optimal hyperparameters.\n\n### Define Sweep Configuration\n\n```python\nsweep_config = {\n    'method': 'bayes',  # or 'grid', 'random'\n    'metric': {\n        'name': 'val\u002Faccuracy',\n        'goal': 'maximize'\n    },\n    'parameters': {\n        'learning_rate': {\n            'distribution': 'log_uniform',\n            'min': 1e-5,\n            'max': 1e-1\n        },\n        'batch_size': {\n            'values': [16, 32, 64, 128]\n        },\n        'optimizer': {\n            'values': ['adam', 'sgd', 'rmsprop']\n        },\n        'dropout': {\n            'distribution': 'uniform',\n            'min': 0.1,\n            'max': 0.5\n        }\n    }\n}\n\n# Initialize sweep\nsweep_id = wandb.sweep(sweep_config, project=\"my-project\")\n```\n\n### Define Training Function\n\n```python\ndef train():\n    # Initialize run\n    run = wandb.init()\n\n    # Access sweep parameters\n    lr = wandb.config.learning_rate\n    batch_size = wandb.config.batch_size\n    optimizer_name = wandb.config.optimizer\n\n    # Build model with sweep config\n    model = build_model(wandb.config)\n    optimizer = get_optimizer(optimizer_name, lr)\n\n    # Training loop\n    for epoch in range(NUM_EPOCHS):\n        train_loss = train_epoch(model, optimizer, batch_size)\n        val_acc = validate(model)\n\n        # Log metrics\n        wandb.log({\n            \"train\u002Floss\": train_loss,\n            \"val\u002Faccuracy\": val_acc\n        })\n\n# Run sweep\nwandb.agent(sweep_id, function=train, count=50)  # Run 50 trials\n```\n\n### Sweep Strategies\n\n```python\n# Grid search - exhaustive\nsweep_config = {\n    'method': 'grid',\n    'parameters': {\n        'lr': {'values': [0.001, 0.01, 0.1]},\n        'batch_size': {'values': [16, 32, 64]}\n    }\n}\n\n# Random search\nsweep_config = {\n    'method': 'random',\n    'parameters': {\n        'lr': {'distribution': 'uniform', 'min': 0.0001, 'max': 0.1},\n        'dropout': {'distribution': 'uniform', 'min': 0.1, 'max': 0.5}\n    }\n}\n\n# Bayesian optimization (recommended)\nsweep_config = {\n    'method': 'bayes',\n    'metric': {'name': 'val\u002Floss', 'goal': 'minimize'},\n    'parameters': {\n        'lr': {'distribution': 'log_uniform', 'min': 1e-5, 'max': 1e-1}\n    }\n}\n```\n\n## Artifacts\n\nTrack datasets, models, and other files with lineage.\n\n### Log Artifacts\n\n```python\n# Create artifact\nartifact = wandb.Artifact(\n    name='training-dataset',\n    type='dataset',\n    description='ImageNet training split',\n    metadata={'size': '1.2M images', 'split': 'train'}\n)\n\n# Add files\nartifact.add_file('data\u002Ftrain.csv')\nartifact.add_dir('data\u002Fimages\u002F')\n\n# Log artifact\nwandb.log_artifact(artifact)\n```\n\n### Use Artifacts\n\n```python\n# Download and use artifact\nrun = wandb.init(project=\"my-project\")\n\n# Download artifact\nartifact = run.use_artifact('training-dataset:latest')\nartifact_dir = artifact.download()\n\n# Use the data\ndata = load_data(f\"{artifact_dir}\u002Ftrain.csv\")\n```\n\n### Model Registry\n\n```python\n# Log model as artifact\nmodel_artifact = wandb.Artifact(\n    name='resnet50-model',\n    type='model',\n    metadata={'architecture': 'ResNet50', 'accuracy': 0.95}\n)\n\nmodel_artifact.add_file('model.pth')\nwandb.log_artifact(model_artifact, aliases=['best', 'production'])\n\n# Link to model registry\nrun.link_artifact(model_artifact, 'model-registry\u002Fproduction-models')\n```\n\n## Integration Examples\n\n### HuggingFace Transformers\n\n```python\nfrom transformers import Trainer, TrainingArguments\nimport wandb\n\n# Initialize W&B\nwandb.init(project=\"hf-transformers\")\n\n# Training arguments with W&B\ntraining_args = TrainingArguments(\n    output_dir=\".\u002Fresults\",\n    report_to=\"wandb\",  # Enable W&B logging\n    run_name=\"bert-finetuning\",\n    logging_steps=100,\n    save_steps=500\n)\n\n# Trainer automatically logs to W&B\ntrainer = Trainer(\n    model=model,\n    args=training_args,\n    train_dataset=train_dataset,\n    eval_dataset=eval_dataset\n)\n\ntrainer.train()\n```\n\n### PyTorch Lightning\n\n```python\nfrom pytorch_lightning import Trainer\nfrom pytorch_lightning.loggers import WandbLogger\nimport wandb\n\n# Create W&B logger\nwandb_logger = WandbLogger(\n    project=\"lightning-demo\",\n    log_model=True  # Log model checkpoints\n)\n\n# Use with Trainer\ntrainer = Trainer(\n    logger=wandb_logger,\n    max_epochs=10\n)\n\ntrainer.fit(model, datamodule=dm)\n```\n\n### Keras\u002FTensorFlow\n\n```python\nimport wandb\nfrom wandb.keras import WandbCallback\n\n# Initialize\nwandb.init(project=\"keras-demo\")\n\n# Add callback\nmodel.fit(\n    x_train, y_train,\n    validation_data=(x_val, y_val),\n    epochs=10,\n    callbacks=[WandbCallback()]  # Auto-logs metrics\n)\n```\n\n## Visualization & Analysis\n\n### Custom Charts\n\n```python\n# Log custom visualizations\nimport matplotlib.pyplot as plt\n\nfig, ax = plt.subplots()\nax.plot(x, y)\nwandb.log({\"custom_plot\": wandb.Image(fig)})\n\n# Log confusion matrix\nwandb.log({\"conf_mat\": wandb.plot.confusion_matrix(\n    probs=None,\n    y_true=ground_truth,\n    preds=predictions,\n    class_names=class_names\n)})\n```\n\n### Reports\n\nCreate shareable reports in W&B UI:\n- Combine runs, charts, and text\n- Markdown support\n- Embeddable visualizations\n- Team collaboration\n\n## Best Practices\n\n### 1. Organize with Tags and Groups\n\n```python\nwandb.init(\n    project=\"my-project\",\n    tags=[\"baseline\", \"resnet50\", \"imagenet\"],\n    group=\"resnet-experiments\",  # Group related runs\n    job_type=\"train\"             # Type of job\n)\n```\n\n### 2. Log Everything Relevant\n\n```python\n# Log system metrics\nwandb.log({\n    \"gpu\u002Futil\": gpu_utilization,\n    \"gpu\u002Fmemory\": gpu_memory_used,\n    \"cpu\u002Futil\": cpu_utilization\n})\n\n# Log code version\nwandb.log({\"git_commit\": git_commit_hash})\n\n# Log data splits\nwandb.log({\n    \"data\u002Ftrain_size\": len(train_dataset),\n    \"data\u002Fval_size\": len(val_dataset)\n})\n```\n\n### 3. Use Descriptive Names\n\n```python\n# ✅ Good: Descriptive run names\nwandb.init(\n    project=\"nlp-classification\",\n    name=\"bert-base-lr0.001-bs32-epoch10\"\n)\n\n# ❌ Bad: Generic names\nwandb.init(project=\"nlp\", name=\"run1\")\n```\n\n### 4. Save Important Artifacts\n\n```python\n# Save final model\nartifact = wandb.Artifact('final-model', type='model')\nartifact.add_file('model.pth')\nwandb.log_artifact(artifact)\n\n# Save predictions for analysis\npredictions_table = wandb.Table(\n    columns=[\"id\", \"input\", \"prediction\", \"ground_truth\"],\n    data=predictions_data\n)\nwandb.log({\"predictions\": predictions_table})\n```\n\n### 5. Use Offline Mode for Unstable Connections\n\n```python\nimport os\n\n# Enable offline mode\nos.environ[\"WANDB_MODE\"] = \"offline\"\n\nwandb.init(project=\"my-project\")\n# ... your code ...\n\n# Sync later\n# wandb sync \u003Crun_directory>\n```\n\n## Team Collaboration\n\n### Share Runs\n\n```python\n# Runs are automatically shareable via URL\nrun = wandb.init(project=\"team-project\")\nprint(f\"Share this URL: {run.url}\")\n```\n\n### Team Projects\n\n- Create team account at wandb.ai\n- Add team members\n- Set project visibility (private\u002Fpublic)\n- Use team-level artifacts and model registry\n\n## Pricing\n\n- **Free**: Unlimited public projects, 100GB storage\n- **Academic**: Free for students\u002Fresearchers\n- **Teams**: $50\u002Fseat\u002Fmonth, private projects, unlimited storage\n- **Enterprise**: Custom pricing, on-prem options\n\n## Resources\n\n- **Documentation**: https:\u002F\u002Fdocs.wandb.ai\n- **GitHub**: https:\u002F\u002Fgithub.com\u002Fwandb\u002Fwandb (10.5k+ stars)\n- **Examples**: https:\u002F\u002Fgithub.com\u002Fwandb\u002Fexamples\n- **Community**: https:\u002F\u002Fwandb.ai\u002Fcommunity\n- **Discord**: https:\u002F\u002Fwandb.me\u002Fdiscord\n\n## See Also\n\n- `references\u002Fsweeps.md` - Comprehensive hyperparameter optimization guide\n- `references\u002Fartifacts.md` - Data and model versioning patterns\n- `references\u002Fintegrations.md` - Framework-specific examples\n\n\n","# Weights & Biases: ML Experiment Tracking & MLOps\n\n## When to Use This Skill\n\nUse Weights & Biases (W&B) when you need to:\n- **Track ML experiments** with automatic metric logging\n- **Visualize training** in real-time dashboards\n- **Compare runs** across hyperparameters and configurations\n- **Optimize hyperparameters** with automated sweeps\n- **Manage model registry** with versioning and lineage\n- **Collaborate on ML projects** with team workspaces\n- **Track artifacts** (datasets, models, code) with lineage\n\n**Users**: 200,000+ ML practitioners | **GitHub Stars**: 10.5k+ | **Integrations**: 100+\n\n## Installation\n\n```bash\n# Install W&B\npip install wandb\n\n# Login (creates API key)\nwandb login\n\n# Or set API key programmatically\nexport WANDB_API_KEY=your_api_key_here\n```\n\n## Quick Start\n\n### Basic Experiment Tracking\n\n```python\nimport wandb\n\n# Initialize a run\nrun = wandb.init(\n    project=\"my-project\",\n    config={\n        \"learning_rate\": 0.001,\n        \"epochs\": 10,\n        \"batch_size\": 32,\n        \"architecture\": \"ResNet50\"\n    }\n)\n\n# Training loop\nfor epoch in range(run.config.epochs):\n    # Your training code\n    train_loss = train_epoch()\n    val_loss = validate()\n\n    # Log metrics\n    wandb.log({\n        \"epoch\": epoch,\n        \"train\u002Floss\": train_loss,\n        \"val\u002Floss\": val_loss,\n        \"train\u002Faccuracy\": train_acc,\n        \"val\u002Faccuracy\": val_acc\n    })\n\n# Finish the run\nwandb.finish()\n```\n\n### With PyTorch\n\n```python\nimport torch\nimport wandb\n\n# Initialize\nwandb.init(project=\"pytorch-demo\", config={\n    \"lr\": 0.001,\n    \"epochs\": 10\n})\n\n# Access config\nconfig = wandb.config\n\n# Training loop\nfor epoch in range(config.epochs):\n    for batch_idx, (data, target) in enumerate(train_loader):\n        # Forward pass\n        output = model(data)\n        loss = criterion(output, target)\n\n        # Backward pass\n        optimizer.zero_grad()\n        loss.backward()\n        optimizer.step()\n\n        # Log every 100 batches\n        if batch_idx % 100 == 0:\n            wandb.log({\n                \"loss\": loss.item(),\n                \"epoch\": epoch,\n                \"batch\": batch_idx\n            })\n\n# Save model\ntorch.save(model.state_dict(), \"model.pth\")\nwandb.save(\"model.pth\")  # Upload to W&B\n\nwandb.finish()\n```\n\n## Core Concepts\n\n### 1. Projects and Runs\n\n**Project**: Collection of related experiments\n**Run**: Single execution of your training script\n\n```python\n# Create\u002Fuse project\nrun = wandb.init(\n    project=\"image-classification\",\n    name=\"resnet50-experiment-1\",  # Optional run name\n    tags=[\"baseline\", \"resnet\"],    # Organize with tags\n    notes=\"First baseline run\"      # Add notes\n)\n\n# Each run has unique ID\nprint(f\"Run ID: {run.id}\")\nprint(f\"Run URL: {run.url}\")\n```\n\n### 2. Configuration Tracking\n\nTrack hyperparameters automatically:\n\n```python\nconfig = {\n    # Model architecture\n    \"model\": \"ResNet50\",\n    \"pretrained\": True,\n\n    # Training params\n    \"learning_rate\": 0.001,\n    \"batch_size\": 32,\n    \"epochs\": 50,\n    \"optimizer\": \"Adam\",\n\n    # Data params\n    \"dataset\": \"ImageNet\",\n    \"augmentation\": \"standard\"\n}\n\nwandb.init(project=\"my-project\", config=config)\n\n# Access config during training\nlr = wandb.config.learning_rate\nbatch_size = wandb.config.batch_size\n```\n\n### 3. Metric Logging\n\n```python\n# Log scalars\nwandb.log({\"loss\": 0.5, \"accuracy\": 0.92})\n\n# Log multiple metrics\nwandb.log({\n    \"train\u002Floss\": train_loss,\n    \"train\u002Faccuracy\": train_acc,\n    \"val\u002Floss\": val_loss,\n    \"val\u002Faccuracy\": val_acc,\n    \"learning_rate\": current_lr,\n    \"epoch\": epoch\n})\n\n# Log with custom x-axis\nwandb.log({\"loss\": loss}, step=global_step)\n\n# Log media (images, audio, video)\nwandb.log({\"examples\": [wandb.Image(img) for img in images]})\n\n# Log histograms\nwandb.log({\"gradients\": wandb.Histogram(gradients)})\n\n# Log tables\ntable = wandb.Table(columns=[\"id\", \"prediction\", \"ground_truth\"])\nwandb.log({\"predictions\": table})\n```\n\n### 4. Model Checkpointing\n\n```python\nimport torch\nimport wandb\n\n# Save model checkpoint\ncheckpoint = {\n    'epoch': epoch,\n    'model_state_dict': model.state_dict(),\n    'optimizer_state_dict': optimizer.state_dict(),\n    'loss': loss,\n}\n\ntorch.save(checkpoint, 'checkpoint.pth')\n\n# Upload to W&B\nwandb.save('checkpoint.pth')\n\n# Or use Artifacts (recommended)\nartifact = wandb.Artifact('model', type='model')\nartifact.add_file('checkpoint.pth')\nwandb.log_artifact(artifact)\n```\n\n## Hyperparameter Sweeps\n\nAutomatically search for optimal hyperparameters.\n\n### Define Sweep Configuration\n\n```python\nsweep_config = {\n    'method': 'bayes',  # or 'grid', 'random'\n    'metric': {\n        'name': 'val\u002Faccuracy',\n        'goal': 'maximize'\n    },\n    'parameters': {\n        'learning_rate': {\n            'distribution': 'log_uniform',\n            'min': 1e-5,\n            'max': 1e-1\n        },\n        'batch_size': {\n            'values': [16, 32, 64, 128]\n        },\n        'optimizer': {\n            'values': ['adam', 'sgd', 'rmsprop']\n        },\n        'dropout': {\n            'distribution': 'uniform',\n            'min': 0.1,\n            'max': 0.5\n        }\n    }\n}\n\n# Initialize sweep\nsweep_id = wandb.sweep(sweep_config, project=\"my-project\")\n```\n\n### Define Training Function\n\n```python\ndef train():\n    # Initialize run\n    run = wandb.init()\n\n    # Access sweep parameters\n    lr = wandb.config.learning_rate\n    batch_size = wandb.config.batch_size\n    optimizer_name = wandb.config.optimizer\n\n    # Build model with sweep config\n    model = build_model(wandb.config)\n    optimizer = get_optimizer(optimizer_name, lr)\n\n    # Training loop\n    for epoch in range(NUM_EPOCHS):\n        train_loss = train_epoch(model, optimizer, batch_size)\n        val_acc = validate(model)\n\n        # Log metrics\n        wandb.log({\n            \"train\u002Floss\": train_loss,\n            \"val\u002Faccuracy\": val_acc\n        })\n\n# Run sweep\nwandb.agent(sweep_id, function=train, count=50)  # Run 50 trials\n```\n\n### Sweep Strategies\n\n```python\n# Grid search - exhaustive\nsweep_config = {\n    'method': 'grid',\n    'parameters': {\n        'lr': {'values': [0.001, 0.01, 0.1]},\n        'batch_size': {'values': [16, 32, 64]}\n    }\n}\n\n# Random search\nsweep_config = {\n    'method': 'random',\n    'parameters': {\n        'lr': {'distribution': 'uniform', 'min': 0.0001, 'max': 0.1},\n        'dropout': {'distribution': 'uniform', 'min': 0.1, 'max': 0.5}\n    }\n}\n\n# Bayesian optimization (recommended)\nsweep_config = {\n    'method': 'bayes',\n    'metric': {'name': 'val\u002Floss', 'goal': 'minimize'},\n    'parameters': {\n        'lr': {'distribution': 'log_uniform', 'min': 1e-5, 'max': 1e-1}\n    }\n}\n```\n\n## Artifacts\n\nTrack datasets, models, and other files with lineage.\n\n### Log Artifacts\n\n```python\n# Create artifact\nartifact = wandb.Artifact(\n    name='training-dataset',\n    type='dataset',\n    description='ImageNet training split',\n    metadata={'size': '1.2M images', 'split': 'train'}\n)\n\n# Add files\nartifact.add_file('data\u002Ftrain.csv')\nartifact.add_dir('data\u002Fimages\u002F')\n\n# Log artifact\nwandb.log_artifact(artifact)\n```\n\n### Use Artifacts\n\n```python\n# Download and use artifact\nrun = wandb.init(project=\"my-project\")\n\n# Download artifact\nartifact = run.use_artifact('training-dataset:latest')\nartifact_dir = artifact.download()\n\n# Use the data\ndata = load_data(f\"{artifact_dir}\u002Ftrain.csv\")\n```\n\n### Model Registry\n\n```python\n# Log model as artifact\nmodel_artifact = wandb.Artifact(\n    name='resnet50-model',\n    type='model',\n    metadata={'architecture': 'ResNet50', 'accuracy': 0.95}\n)\n\nmodel_artifact.add_file('model.pth')\nwandb.log_artifact(model_artifact, aliases=['best', 'production'])\n\n# Link to model registry\nrun.link_artifact(model_artifact, 'model-registry\u002Fproduction-models')\n```\n\n## Integration Examples\n\n### HuggingFace Transformers\n\n```python\nfrom transformers import Trainer, TrainingArguments\nimport wandb\n\n# Initialize W&B\nwandb.init(project=\"hf-transformers\")\n\n# Training arguments with W&B\ntraining_args = TrainingArguments(\n    output_dir=\".\u002Fresults\",\n    report_to=\"wandb\",  # Enable W&B logging\n    run_name=\"bert-finetuning\",\n    logging_steps=100,\n    save_steps=500\n)\n\n# Trainer automatically logs to W&B\ntrainer = Trainer(\n    model=model,\n    args=training_args,\n    train_dataset=train_dataset,\n    eval_dataset=eval_dataset\n)\n\ntrainer.train()\n```\n\n### PyTorch Lightning\n\n```python\nfrom pytorch_lightning import Trainer\nfrom pytorch_lightning.loggers import WandbLogger\nimport wandb\n\n# Create W&B logger\nwandb_logger = WandbLogger(\n    project=\"lightning-demo\",\n    log_model=True  # Log model checkpoints\n)\n\n# Use with Trainer\ntrainer = Trainer(\n    logger=wandb_logger,\n    max_epochs=10\n)\n\ntrainer.fit(model, datamodule=dm)\n```\n\n### Keras\u002FTensorFlow\n\n```python\nimport wandb\nfrom wandb.keras import WandbCallback\n\n# Initialize\nwandb.init(project=\"keras-demo\")\n\n# Add callback\nmodel.fit(\n    x_train, y_train,\n    validation_data=(x_val, y_val),\n    epochs=10,\n    callbacks=[WandbCallback()]  # Auto-logs metrics\n)\n```\n\n## Visualization & Analysis\n\n### Custom Charts\n\n```python\n# Log custom visualizations\nimport matplotlib.pyplot as plt\n\nfig, ax = plt.subplots()\nax.plot(x, y)\nwandb.log({\"custom_plot\": wandb.Image(fig)})\n\n# Log confusion matrix\nwandb.log({\"conf_mat\": wandb.plot.confusion_matrix(\n    probs=None,\n    y_true=ground_truth,\n    preds=predictions,\n    class_names=class_names\n)})\n```\n\n### Reports\n\nCreate shareable reports in W&B UI:\n- Combine runs, charts, and text\n- Markdown support\n- Embeddable visualizations\n- Team collaboration\n\n## Best Practices\n\n### 1. Organize with Tags and Groups\n\n```python\nwandb.init(\n    project=\"my-project\",\n    tags=[\"baseline\", \"resnet50\", \"imagenet\"],\n    group=\"resnet-experiments\",  # Group related runs\n    job_type=\"train\"             # Type of job\n)\n```\n\n### 2. Log Everything Relevant\n\n```python\n# Log system metrics\nwandb.log({\n    \"gpu\u002Futil\": gpu_utilization,\n    \"gpu\u002Fmemory\": gpu_memory_used,\n    \"cpu\u002Futil\": cpu_utilization\n})\n\n# Log code version\nwandb.log({\"git_commit\": git_commit_hash})\n\n# Log data splits\nwandb.log({\n    \"data\u002Ftrain_size\": len(train_dataset),\n    \"data\u002Fval_size\": len(val_dataset)\n})\n```\n\n### 3. Use Descriptive Names\n\n```python\n# ✅ Good: Descriptive run names\nwandb.init(\n    project=\"nlp-classification\",\n    name=\"bert-base-lr0.001-bs32-epoch10\"\n)\n\n# ❌ Bad: Generic names\nwandb.init(project=\"nlp\", name=\"run1\")\n```\n\n### 4. Save Important Artifacts\n\n```python\n# Save final model\nartifact = wandb.Artifact('final-model', type='model')\nartifact.add_file('model.pth')\nwandb.log_artifact(artifact)\n\n# Save predictions for analysis\npredictions_table = wandb.Table(\n    columns=[\"id\", \"input\", \"prediction\", \"ground_truth\"],\n    data=predictions_data\n)\nwandb.log({\"predictions\": predictions_table})\n```\n\n### 5. Use Offline Mode for Unstable Connections\n\n```python\nimport os\n\n# Enable offline mode\nos.environ[\"WANDB_MODE\"] = \"offline\"\n\nwandb.init(project=\"my-project\")\n# ... your code ...\n\n# Sync later\n# wandb sync \u003Crun_directory>\n```\n\n## Team Collaboration\n\n### Share Runs\n\n```python\n# Runs are automatically shareable via URL\nrun = wandb.init(project=\"team-project\")\nprint(f\"Share this URL: {run.url}\")\n```\n\n### Team Projects\n\n- Create team account at wandb.ai\n- Add team members\n- Set project visibility (private\u002Fpublic)\n- Use team-level artifacts and model registry\n\n## Pricing\n\n- **Free**: Unlimited public projects, 100GB storage\n- **Academic**: Free for students\u002Fresearchers\n- **Teams**: $50\u002Fseat\u002Fmonth, private projects, unlimited storage\n- **Enterprise**: Custom pricing, on-prem options\n\n## Resources\n\n- **Documentation**: https:\u002F\u002Fdocs.wandb.ai\n- **GitHub**: https:\u002F\u002Fgithub.com\u002Fwandb\u002Fwandb (10.5k+ stars)\n- **Examples**: https:\u002F\u002Fgithub.com\u002Fwandb\u002Fexamples\n- **Community**: https:\u002F\u002Fwandb.ai\u002Fcommunity\n- **Discord**: https:\u002F\u002Fwandb.me\u002Fdiscord\n\n## See Also\n\n- `references\u002Fsweeps.md` - Comprehensive hyperparameter optimization guide\n- `references\u002Fartifacts.md` - Data and model versioning patterns\n- `references\u002Fintegrations.md` - Framework-specific examples\n\n\n",{"name":5,"author":48,"license":14,"version":49,"metadata":50,"description":7,"dependencies":59},"Orchestra Research","1.0.0",{"hermes":51},{"tags":52},[53,54,37,36,26,34,55,35,56,57,58],"MLOps","Weights And Biases","Collaboration","PyTorch","TensorFlow","HuggingFace",[60],"wandb","main","17ee50c38aefa9426a89d7e8ee167fdc4c184fb7","skills\u002Fmlops\u002Fevaluation\u002Fweights-and-biases",0,"https:\u002F\u002Fgithub.com\u002Fxuanxuan1983\u002Fhermes-skills-backup","https:\u002F\u002Fgithub.com\u002Fxuanxuan1983","2026-05-24 00:14:48",{"source":8,"skillName":5,"commands":69,"prompt":79,"zipAvailable":80},[70,73,76],{"tool":71,"command":72},"npx","npx skills add xuanxuan1983\u002Fhermes-skills-backup --skill weights-and-biases",{"tool":74,"command":75},"bunx","bunx skills add xuanxuan1983\u002Fhermes-skills-backup --skill weights-and-biases",{"tool":77,"command":78},"pnpm","pnpm dlx skills add xuanxuan1983\u002Fhermes-skills-backup --skill weights-and-biases","安装技能，技能地址：https:\u002F\u002Fgithub.com\u002Fxuanxuan1983\u002Fhermes-skills-backup\u002Fblob\u002Fmain\u002Fskills\u002Fmlops\u002Fevaluation\u002Fweights-and-biases\u002FSKILL.md",true,[82,88,92],{"path":83,"resourceType":84,"resourceTypeDesc":85,"sizeBytes":86,"sourceHtmlUrl":87},"references\u002Fartifacts.md","REFERENCE","参考文档",13425,"https:\u002F\u002Fgithub.com\u002Fxuanxuan1983\u002Fhermes-skills-backup\u002Fblob\u002Fmain\u002Fskills\u002Fmlops\u002Fevaluation\u002Fweights-and-biases\u002Freferences\u002Fartifacts.md",{"path":89,"resourceType":84,"resourceTypeDesc":85,"sizeBytes":90,"sourceHtmlUrl":91},"references\u002Fintegrations.md",16219,"https:\u002F\u002Fgithub.com\u002Fxuanxuan1983\u002Fhermes-skills-backup\u002Fblob\u002Fmain\u002Fskills\u002Fmlops\u002Fevaluation\u002Fweights-and-biases\u002Freferences\u002Fintegrations.md",{"path":93,"resourceType":84,"resourceTypeDesc":85,"sizeBytes":94,"sourceHtmlUrl":95},"references\u002Fsweeps.md",17675,"https:\u002F\u002Fgithub.com\u002Fxuanxuan1983\u002Fhermes-skills-backup\u002Fblob\u002Fmain\u002Fskills\u002Fmlops\u002Fevaluation\u002Fweights-and-biases\u002Freferences\u002Fsweeps.md",[97,138,166,182,204,225],{"skillNo":98,"name":99,"nameZh":6,"description":100,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":6,"repositoryNo":12,"starsCount":13,"licenseSpdxId":14,"licenseFileUrl":6,"sourceHtmlUrl":101,"categories":102,"keywords":117,"publishedAt":43},"01M71JFHVW1W4XE013TY9ZYN9P","writing-plans","Use when you have a spec or requirements for a multi-step task. Creates comprehensive implementation plans with bite-sized tasks, exact file paths, and complete code examples.","https:\u002F\u002Fgithub.com\u002Fxuanxuan1983\u002Fhermes-skills-backup\u002Fblob\u002Fmain\u002Fskills\u002Fsoftware-development\u002Fwriting-plans\u002FSKILL.md",[103,106,109,114],{"categoryNo":104,"categoryName":105,"taxonomyType":20,"taxonomyTypeDesc":21},"01JBSEEDCAT000000000000015","视觉设计",{"categoryNo":107,"categoryName":108,"taxonomyType":20,"taxonomyTypeDesc":21},"01JBSEEDCAT000000000000018","写作与文案",{"categoryNo":110,"categoryName":111,"taxonomyType":112,"taxonomyTypeDesc":113},"01JBSEEDCAT000000000000030","开发工程师","PROFESSION","适用职业",{"categoryNo":115,"categoryName":116,"taxonomyType":112,"taxonomyTypeDesc":113},"01JBSEEDCAT000000000000036","内容创作者",[118,119,120,121,122,123,124,125,126,127,128,129,130,131,132,133,134,135,136,137],"writing","writing plans","plans","test-driven","test-driven-development","superpowers","design","when","workflow","you","code","tasks","subagent","code-review","review","documentation","paths","planning","development","for",{"skillNo":139,"name":122,"nameZh":6,"description":140,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":6,"repositoryNo":12,"starsCount":13,"licenseSpdxId":14,"licenseFileUrl":6,"sourceHtmlUrl":141,"categories":142,"keywords":151,"publishedAt":43},"01M71JFF859RV0XGWS206X1TMB","Use when implementing any feature or bugfix, before writing implementation code. Enforces RED-GREEN-REFACTOR cycle with test-first approach.","https:\u002F\u002Fgithub.com\u002Fxuanxuan1983\u002Fhermes-skills-backup\u002Fblob\u002Fmain\u002Fskills\u002Fsoftware-development\u002Ftest-driven-development\u002FSKILL.md",[143,146,149,150],{"categoryNo":144,"categoryName":145,"taxonomyType":20,"taxonomyTypeDesc":21},"01JBSEEDCAT000000000000007","代码审查",{"categoryNo":147,"categoryName":148,"taxonomyType":20,"taxonomyTypeDesc":21},"01JBSEEDCAT000000000000008","测试与质量",{"categoryNo":107,"categoryName":108,"taxonomyType":20,"taxonomyTypeDesc":21},{"categoryNo":115,"categoryName":116,"taxonomyType":112,"taxonomyTypeDesc":113},[152,136,153,154,155,156,157,158,159,160,123,118,161,125,128,130,162,163,164,165],"test driven development","test","driven","test-first","green","cycle","red-green-refactor","testing","implementing","debugging","systematic","bugfix","refactor","systematic-debugging",{"skillNo":167,"name":165,"nameZh":6,"description":168,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":6,"repositoryNo":12,"starsCount":13,"licenseSpdxId":14,"licenseFileUrl":6,"sourceHtmlUrl":169,"categories":170,"keywords":172,"publishedAt":43},"01M71JFCA5S48PZ5M7DG31R8K5","Use when encountering any bug, test failure, or unexpected behavior. 4-phase root cause investigation — NO fixes without understanding the problem first.","https:\u002F\u002Fgithub.com\u002Fxuanxuan1983\u002Fhermes-skills-backup\u002Fblob\u002Fmain\u002Fskills\u002Fsoftware-development\u002Fsystematic-debugging\u002FSKILL.md",[171],{"categoryNo":147,"categoryName":148,"taxonomyType":20,"taxonomyTypeDesc":21},[173,161,162,121,174,175,122,123,118,125,130,176,177,136,178,179,120,180,181,153],"systematic debugging","problem","problem-solving","phase","bug","fixes","root cause","use","root-cause",{"skillNo":183,"name":184,"nameZh":6,"description":185,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":6,"repositoryNo":12,"starsCount":13,"licenseSpdxId":14,"licenseFileUrl":6,"sourceHtmlUrl":186,"categories":187,"keywords":197,"publishedAt":43},"01M71JF9RQ092KM5ZXCBBT6FS1","subagent-driven-development","Use when executing implementation plans with independent tasks. Dispatches fresh delegate_task per task with two-stage review (spec compliance then code quality).","https:\u002F\u002Fgithub.com\u002Fxuanxuan1983\u002Fhermes-skills-backup\u002Fblob\u002Fmain\u002Fskills\u002Fsoftware-development\u002Fsubagent-driven-development\u002FSKILL.md",[188,189,192,193,194],{"categoryNo":107,"categoryName":108,"taxonomyType":20,"taxonomyTypeDesc":21},{"categoryNo":190,"categoryName":191,"taxonomyType":20,"taxonomyTypeDesc":21},"01JBSEEDCAT000000000000023","Skill与智能体开发",{"categoryNo":110,"categoryName":111,"taxonomyType":112,"taxonomyTypeDesc":113},{"categoryNo":115,"categoryName":116,"taxonomyType":112,"taxonomyTypeDesc":113},{"categoryNo":195,"categoryName":196,"taxonomyType":112,"taxonomyTypeDesc":113},"01JBSEEDCAT000000000000037","律师与法务",[198,130,136,154,121,122,199,200,123,118,125,126,128,129,131,132,201,202,203,120],"subagent driven development","delegate","delegation","fresh","parallel","code quality",{"skillNo":205,"name":206,"nameZh":6,"description":207,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":6,"repositoryNo":12,"starsCount":13,"licenseSpdxId":14,"licenseFileUrl":6,"sourceHtmlUrl":208,"categories":209,"keywords":212,"publishedAt":43},"01M71JF76DA8ARZ4JRYR4VBRWM","requesting-code-review","Pre-commit verification pipeline — static security scan, baseline-aware quality gates, independent reviewer subagent, and auto-fix loop. Use after code changes and before committing, pushing, or opening a PR.","https:\u002F\u002Fgithub.com\u002Fxuanxuan1983\u002Fhermes-skills-backup\u002Fblob\u002Fmain\u002Fskills\u002Fsoftware-development\u002Frequesting-code-review\u002FSKILL.md",[210,211],{"categoryNo":144,"categoryName":145,"taxonomyType":20,"taxonomyTypeDesc":21},{"categoryNo":110,"categoryName":111,"taxonomyType":112,"taxonomyTypeDesc":113},[213,214,128,132,121,215,216,122,217,218,123,219,118,220,221,222,130,131,223,224],"code review","requesting code review","pipeline","static","fix","auto-fix","auto","loop","security scan","commit","reviewer","verification",{"skillNo":226,"name":227,"nameZh":6,"description":228,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":6,"repositoryNo":12,"starsCount":13,"licenseSpdxId":14,"licenseFileUrl":6,"sourceHtmlUrl":229,"categories":230,"keywords":231,"publishedAt":43},"01M71JF52HZT5VQXFVF9K2H76K","plan","Plan mode for Hermes — inspect context, write a markdown plan into the active workspace's `.hermes\u002Fplans\u002F` directory, and do not execute the work.","https:\u002F\u002Fgithub.com\u002Fxuanxuan1983\u002Fhermes-skills-backup\u002Fblob\u002Fmain\u002Fskills\u002Fsoftware-development\u002Fplan\u002FSKILL.md",[],[227,232,233,135,136,137,234,235,120,118,126,130,236,237,238,239,240,241,242,243],"plan-mode","do","inspect","mode","work","markdown","active","directory","workspace","plan mode","write","implementation"]