[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"skill:01M7AWSMM49599FQE8FKCWM9TK":3,"skill-same:01M7AWSMM49599FQE8FKCWM9TK":44},{"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":17,"publishedAt":18,"linkOnly":19,"markdownContent":20,"markdownBody":21,"frontMatter":22,"refName":23,"commitSha":24,"skillDirectory":25,"forksCount":26,"repositoryHtmlUrl":27,"repositoryHomepageUrl":6,"ownerHtmlUrl":28,"ownerBlogUrl":6,"ownerDescription":6,"updatedAt":29,"install":30,"resources":43},"01M7AWSMM49599FQE8FKCWM9TK","domain-ml",null,"Use when building ML\u002FAI apps in Rust. Keywords: machine learning, ML, AI, tensor, model, inference, neural network, deep learning, training, prediction, ndarray, tch-rs, burn, candle, 机器学习, 人工智能, 模型推理","wxxb789\u002Fmarkturbo","wxxb789","https:\u002F\u002Favatars.githubusercontent.com\u002Fu\u002F49564443?v=4","01M7AWR3AB19SNE4ATARY9KV3W","01M7AWR3AZ5Y5ADAY47KF3RBRJ",1,"Apache-2.0","https:\u002F\u002Fgithub.com\u002Fwxxb789\u002Fmarkturbo\u002Fblob\u002Fmain\u002F.agents\u002Fskills\u002Fdomain-ml\u002FSKILL.md",[],[],"2026-11-15 14:19:17",false,"---\nname: domain-ml\ndescription: \"Use when building ML\u002FAI apps in Rust. Keywords: machine learning, ML, AI, tensor, model, inference, neural network, deep learning, training, prediction, ndarray, tch-rs, burn, candle, 机器学习, 人工智能, 模型推理\"\nuser-invocable: false\n---\n\n# Machine Learning Domain\n\n> **Layer 3: Domain Constraints**\n\n## Domain Constraints → Design Implications\n\n| Domain Rule | Design Constraint | Rust Implication |\n|-------------|-------------------|------------------|\n| Large data | Efficient memory | Zero-copy, streaming |\n| GPU acceleration | CUDA\u002FMetal support | candle, tch-rs |\n| Model portability | Standard formats | ONNX |\n| Batch processing | Throughput over latency | Batched inference |\n| Numerical precision | Float handling | ndarray, careful f32\u002Ff64 |\n| Reproducibility | Deterministic | Seeded random, versioning |\n\n---\n\n## Critical Constraints\n\n### Memory Efficiency\n\n```\nRULE: Avoid copying large tensors\nWHY: Memory bandwidth is bottleneck\nRUST: References, views, in-place ops\n```\n\n### GPU Utilization\n\n```\nRULE: Batch operations for GPU efficiency\nWHY: GPU overhead per kernel launch\nRUST: Batch sizes, async data loading\n```\n\n### Model Portability\n\n```\nRULE: Use standard model formats\nWHY: Train in Python, deploy in Rust\nRUST: ONNX via tract or candle\n```\n\n---\n\n## Trace Down ↓\n\nFrom constraints to design (Layer 2):\n\n```\n\"Need efficient data pipelines\"\n    ↓ m10-performance: Streaming, batching\n    ↓ polars: Lazy evaluation\n\n\"Need GPU inference\"\n    ↓ m07-concurrency: Async data loading\n    ↓ candle\u002Ftch-rs: CUDA backend\n\n\"Need model loading\"\n    ↓ m12-lifecycle: Lazy init, caching\n    ↓ tract: ONNX runtime\n```\n\n---\n\n## Use Case → Framework\n\n| Use Case | Recommended | Why |\n|----------|-------------|-----|\n| Inference only | tract (ONNX) | Lightweight, portable |\n| Training + inference | candle, burn | Pure Rust, GPU |\n| PyTorch models | tch-rs | Direct bindings |\n| Data pipelines | polars | Fast, lazy eval |\n\n## Key Crates\n\n| Purpose | Crate |\n|---------|-------|\n| Tensors | ndarray |\n| ONNX inference | tract |\n| ML framework | candle, burn |\n| PyTorch bindings | tch-rs |\n| Data processing | polars |\n| Embeddings | fastembed |\n\n## Design Patterns\n\n| Pattern | Purpose | Implementation |\n|---------|---------|----------------|\n| Model loading | Once, reuse | `OnceLock\u003CModel>` |\n| Batching | Throughput | Collect then process |\n| Streaming | Large data | Iterator-based |\n| GPU async | Parallelism | Data loading parallel to compute |\n\n## Code Pattern: Inference Server\n\n```rust\nuse std::sync::OnceLock;\nuse tract_onnx::prelude::*;\n\nstatic MODEL: OnceLock\u003CSimplePlan\u003CTypedFact, Box\u003Cdyn TypedOp>, Graph\u003CTypedFact, Box\u003Cdyn TypedOp>>>> = OnceLock::new();\n\nfn get_model() -> &'static SimplePlan\u003C...> {\n    MODEL.get_or_init(|| {\n        tract_onnx::onnx()\n            .model_for_path(\"model.onnx\")\n            .unwrap()\n            .into_optimized()\n            .unwrap()\n            .into_runnable()\n            .unwrap()\n    })\n}\n\nasync fn predict(input: Vec\u003Cf32>) -> anyhow::Result\u003CVec\u003Cf32>> {\n    let model = get_model();\n    let input = tract_ndarray::arr1(&input).into_shape((1, input.len()))?;\n    let result = model.run(tvec!(input.into()))?;\n    Ok(result[0].to_array_view::\u003Cf32>()?.iter().copied().collect())\n}\n```\n\n## Code Pattern: Batched Inference\n\n```rust\nasync fn batch_predict(inputs: Vec\u003CVec\u003Cf32>>, batch_size: usize) -> Vec\u003CVec\u003Cf32>> {\n    let mut results = Vec::with_capacity(inputs.len());\n\n    for batch in inputs.chunks(batch_size) {\n        \u002F\u002F Stack inputs into batch tensor\n        let batch_tensor = stack_inputs(batch);\n\n        \u002F\u002F Run inference on batch\n        let batch_output = model.run(batch_tensor).await;\n\n        \u002F\u002F Unstack results\n        results.extend(unstack_outputs(batch_output));\n    }\n\n    results\n}\n```\n\n---\n\n## Common Mistakes\n\n| Mistake | Domain Violation | Fix |\n|---------|-----------------|-----|\n| Clone tensors | Memory waste | Use views |\n| Single inference | GPU underutilized | Batch processing |\n| Load model per request | Slow | Singleton pattern |\n| Sync data loading | GPU idle | Async pipeline |\n\n---\n\n## Trace to Layer 1\n\n| Constraint | Layer 2 Pattern | Layer 1 Implementation |\n|------------|-----------------|------------------------|\n| Memory efficiency | Zero-copy | ndarray views |\n| Model singleton | Lazy init | OnceLock\u003CModel> |\n| Batch processing | Chunked iteration | chunks() + parallel |\n| GPU async | Concurrent loading | tokio::spawn + GPU |\n\n---\n\n## Related Skills\n\n| When | See |\n|------|-----|\n| Performance | m10-performance |\n| Lazy initialization | m12-lifecycle |\n| Async patterns | m07-concurrency |\n| Memory efficiency | m01-ownership |\n","# Machine Learning Domain\n\n> **Layer 3: Domain Constraints**\n\n## Domain Constraints → Design Implications\n\n| Domain Rule | Design Constraint | Rust Implication |\n|-------------|-------------------|------------------|\n| Large data | Efficient memory | Zero-copy, streaming |\n| GPU acceleration | CUDA\u002FMetal support | candle, tch-rs |\n| Model portability | Standard formats | ONNX |\n| Batch processing | Throughput over latency | Batched inference |\n| Numerical precision | Float handling | ndarray, careful f32\u002Ff64 |\n| Reproducibility | Deterministic | Seeded random, versioning |\n\n---\n\n## Critical Constraints\n\n### Memory Efficiency\n\n```\nRULE: Avoid copying large tensors\nWHY: Memory bandwidth is bottleneck\nRUST: References, views, in-place ops\n```\n\n### GPU Utilization\n\n```\nRULE: Batch operations for GPU efficiency\nWHY: GPU overhead per kernel launch\nRUST: Batch sizes, async data loading\n```\n\n### Model Portability\n\n```\nRULE: Use standard model formats\nWHY: Train in Python, deploy in Rust\nRUST: ONNX via tract or candle\n```\n\n---\n\n## Trace Down ↓\n\nFrom constraints to design (Layer 2):\n\n```\n\"Need efficient data pipelines\"\n    ↓ m10-performance: Streaming, batching\n    ↓ polars: Lazy evaluation\n\n\"Need GPU inference\"\n    ↓ m07-concurrency: Async data loading\n    ↓ candle\u002Ftch-rs: CUDA backend\n\n\"Need model loading\"\n    ↓ m12-lifecycle: Lazy init, caching\n    ↓ tract: ONNX runtime\n```\n\n---\n\n## Use Case → Framework\n\n| Use Case | Recommended | Why |\n|----------|-------------|-----|\n| Inference only | tract (ONNX) | Lightweight, portable |\n| Training + inference | candle, burn | Pure Rust, GPU |\n| PyTorch models | tch-rs | Direct bindings |\n| Data pipelines | polars | Fast, lazy eval |\n\n## Key Crates\n\n| Purpose | Crate |\n|---------|-------|\n| Tensors | ndarray |\n| ONNX inference | tract |\n| ML framework | candle, burn |\n| PyTorch bindings | tch-rs |\n| Data processing | polars |\n| Embeddings | fastembed |\n\n## Design Patterns\n\n| Pattern | Purpose | Implementation |\n|---------|---------|----------------|\n| Model loading | Once, reuse | `OnceLock\u003CModel>` |\n| Batching | Throughput | Collect then process |\n| Streaming | Large data | Iterator-based |\n| GPU async | Parallelism | Data loading parallel to compute |\n\n## Code Pattern: Inference Server\n\n```rust\nuse std::sync::OnceLock;\nuse tract_onnx::prelude::*;\n\nstatic MODEL: OnceLock\u003CSimplePlan\u003CTypedFact, Box\u003Cdyn TypedOp>, Graph\u003CTypedFact, Box\u003Cdyn TypedOp>>>> = OnceLock::new();\n\nfn get_model() -> &'static SimplePlan\u003C...> {\n    MODEL.get_or_init(|| {\n        tract_onnx::onnx()\n            .model_for_path(\"model.onnx\")\n            .unwrap()\n            .into_optimized()\n            .unwrap()\n            .into_runnable()\n            .unwrap()\n    })\n}\n\nasync fn predict(input: Vec\u003Cf32>) -> anyhow::Result\u003CVec\u003Cf32>> {\n    let model = get_model();\n    let input = tract_ndarray::arr1(&input).into_shape((1, input.len()))?;\n    let result = model.run(tvec!(input.into()))?;\n    Ok(result[0].to_array_view::\u003Cf32>()?.iter().copied().collect())\n}\n```\n\n## Code Pattern: Batched Inference\n\n```rust\nasync fn batch_predict(inputs: Vec\u003CVec\u003Cf32>>, batch_size: usize) -> Vec\u003CVec\u003Cf32>> {\n    let mut results = Vec::with_capacity(inputs.len());\n\n    for batch in inputs.chunks(batch_size) {\n        \u002F\u002F Stack inputs into batch tensor\n        let batch_tensor = stack_inputs(batch);\n\n        \u002F\u002F Run inference on batch\n        let batch_output = model.run(batch_tensor).await;\n\n        \u002F\u002F Unstack results\n        results.extend(unstack_outputs(batch_output));\n    }\n\n    results\n}\n```\n\n---\n\n## Common Mistakes\n\n| Mistake | Domain Violation | Fix |\n|---------|-----------------|-----|\n| Clone tensors | Memory waste | Use views |\n| Single inference | GPU underutilized | Batch processing |\n| Load model per request | Slow | Singleton pattern |\n| Sync data loading | GPU idle | Async pipeline |\n\n---\n\n## Trace to Layer 1\n\n| Constraint | Layer 2 Pattern | Layer 1 Implementation |\n|------------|-----------------|------------------------|\n| Memory efficiency | Zero-copy | ndarray views |\n| Model singleton | Lazy init | OnceLock\u003CModel> |\n| Batch processing | Chunked iteration | chunks() + parallel |\n| GPU async | Concurrent loading | tokio::spawn + GPU |\n\n---\n\n## Related Skills\n\n| When | See |\n|------|-----|\n| Performance | m10-performance |\n| Lazy initialization | m12-lifecycle |\n| Async patterns | m07-concurrency |\n| Memory efficiency | m01-ownership |\n",{"name":5,"description":7,"user-invocable":19},"main","0f3f41ed51c357d4f79561a132514089c1aaa97d",".agents\u002Fskills\u002Fdomain-ml",0,"https:\u002F\u002Fgithub.com\u002Fwxxb789\u002Fmarkturbo","https:\u002F\u002Fgithub.com\u002Fwxxb789","2026-11-15 04:06:53",{"source":8,"skillName":5,"commands":31,"prompt":41,"zipAvailable":42},[32,35,38],{"tool":33,"command":34},"npx","npx skills add wxxb789\u002Fmarkturbo --skill domain-ml",{"tool":36,"command":37},"bunx","bunx skills add wxxb789\u002Fmarkturbo --skill domain-ml",{"tool":39,"command":40},"pnpm","pnpm dlx skills add wxxb789\u002Fmarkturbo --skill domain-ml","安装技能，技能地址：https:\u002F\u002Fgithub.com\u002Fwxxb789\u002Fmarkturbo\u002Fblob\u002Fmain\u002F.agents\u002Fskills\u002Fdomain-ml\u002FSKILL.md",true,[],[45,52,58,65,72,78],{"skillNo":46,"name":47,"nameZh":6,"description":48,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":6,"repositoryNo":12,"starsCount":13,"licenseSpdxId":14,"licenseFileUrl":6,"sourceHtmlUrl":49,"categories":50,"keywords":51,"publishedAt":18},"01M7AWXVVZ6JEJVFGGT2WBSEN0","hello-diagrams","A valid, fully-populated Agent Skill, shown here so the Skills inspector has something correct to display.","https:\u002F\u002Fgithub.com\u002Fwxxb789\u002Fmarkturbo\u002Fblob\u002Fmain\u002Fsample\u002F.claude\u002Fskills\u002Fhello-diagrams\u002FSKILL.md",[],[],{"skillNo":53,"name":54,"nameZh":6,"description":6,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":6,"repositoryNo":12,"starsCount":13,"licenseSpdxId":14,"licenseFileUrl":6,"sourceHtmlUrl":55,"categories":56,"keywords":57,"publishedAt":18},"01M7AWXTNG6ZV6WW5CHJZXJYEC","Broken-Example","https:\u002F\u002Fgithub.com\u002Fwxxb789\u002Fmarkturbo\u002Fblob\u002Fmain\u002Fsample\u002F.claude\u002Fskills\u002Fbroken-example\u002FSKILL.md",[],[],{"skillNo":59,"name":60,"nameZh":6,"description":61,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":6,"repositoryNo":12,"starsCount":13,"licenseSpdxId":14,"licenseFileUrl":6,"sourceHtmlUrl":62,"categories":63,"keywords":64,"publishedAt":18},"01M7AWXS5WN306ENF2MYJYP7VM","valid-skill","A fully conformant skill with every optional field populated.","https:\u002F\u002Fgithub.com\u002Fwxxb789\u002Fmarkturbo\u002Fblob\u002Fmain\u002Ffixtures\u002Fskills\u002Fskills\u002Fvalid-skill\u002FSKILL.md",[],[],{"skillNo":66,"name":67,"nameZh":6,"description":68,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":6,"repositoryNo":12,"starsCount":13,"licenseSpdxId":14,"licenseFileUrl":6,"sourceHtmlUrl":69,"categories":70,"keywords":71,"publishedAt":18},"01M7AWXR0NY08YVHE0G8VTJJ97","a-different-name","The name does not match the directory, which is a conformance warning.","https:\u002F\u002Fgithub.com\u002Fwxxb789\u002Fmarkturbo\u002Fblob\u002Fmain\u002Ffixtures\u002Fskills\u002Fskills\u002Fname-mismatch\u002FSKILL.md",[],[],{"skillNo":73,"name":74,"nameZh":6,"description":6,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":6,"repositoryNo":12,"starsCount":13,"licenseSpdxId":14,"licenseFileUrl":6,"sourceHtmlUrl":75,"categories":76,"keywords":77,"publishedAt":18},"01M7AWXQ5EV9JCA0WG9Q1WB13Y","missing-metadata","https:\u002F\u002Fgithub.com\u002Fwxxb789\u002Fmarkturbo\u002Fblob\u002Fmain\u002Ffixtures\u002Fskills\u002Fskills\u002Fmissing-metadata\u002FSKILL.md",[],[],{"skillNo":79,"name":80,"nameZh":6,"description":6,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":6,"repositoryNo":12,"starsCount":13,"licenseSpdxId":14,"licenseFileUrl":6,"sourceHtmlUrl":81,"categories":82,"keywords":83,"publishedAt":18},"01M7AWXP8GT7FATM53B914ZN50","malformed-yaml","https:\u002F\u002Fgithub.com\u002Fwxxb789\u002Fmarkturbo\u002Fblob\u002Fmain\u002Ffixtures\u002Fskills\u002Fskills\u002Fmalformed-yaml\u002FSKILL.md",[],[]]