actionbook/rust-skills

domain-ml

Use when building ML/AI apps in Rust.

查看源码

使用方法

  1. 复制安装命令,在自己的智能体环境安装前先检查源码包。
  2. 从原始文档核对所需工具、凭据和支持的宿主。
  3. 说明输入材料与期望交付物,并对照来源检查输出。
npx skills add https://github.com/actionbook/rust-skills --skill domain-ml

开始前需了解

  • 安装不包含模型订阅、API 额度或访问文件的权限。
  • 仓库星数和安装量反映活跃情况,不等于独立质量或安全测试。
仓库原始内容

按源仓库内容呈现,保留标题、案例、代码、表格、链接以及原文引用的演示图片。

Machine Learning Domain

Layer 3: Domain Constraints

Domain Constraints → Design Implications

Domain RuleDesign ConstraintRust Implication
Large dataEfficient memoryZero-copy, streaming
GPU accelerationCUDA/Metal supportcandle, tch-rs
Model portabilityStandard formatsONNX
Batch processingThroughput over latencyBatched inference
Numerical precisionFloat handlingndarray, careful f32/f64
ReproducibilityDeterministicSeeded random, versioning

Critical Constraints

Memory Efficiency

RULE: Avoid copying large tensors
WHY: Memory bandwidth is bottleneck
RUST: References, views, in-place ops

GPU Utilization

RULE: Batch operations for GPU efficiency
WHY: GPU overhead per kernel launch
RUST: Batch sizes, async data loading

Model Portability

RULE: Use standard model formats
WHY: Train in Python, deploy in Rust
RUST: ONNX via tract or candle

Trace Down ↓

From constraints to design (Layer 2):

"Need efficient data pipelines"
    ↓ m10-performance: Streaming, batching
    ↓ polars: Lazy evaluation

"Need GPU inference"
    ↓ m07-concurrency: Async data loading
    ↓ candle/tch-rs: CUDA backend

"Need model loading"
    ↓ m12-lifecycle: Lazy init, caching
    ↓ tract: ONNX runtime

Use Case → Framework

Use CaseRecommendedWhy
Inference onlytract (ONNX)Lightweight, portable
Training + inferencecandle, burnPure Rust, GPU
PyTorch modelstch-rsDirect bindings
Data pipelinespolarsFast, lazy eval

Key Crates

PurposeCrate
Tensorsndarray
ONNX inferencetract
ML frameworkcandle, burn
PyTorch bindingstch-rs
Data processingpolars
Embeddingsfastembed

Design Patterns

PatternPurposeImplementation
Model loadingOnce, reuseOnceLock<Model>
BatchingThroughputCollect then process
StreamingLarge dataIterator-based
GPU asyncParallelismData loading parallel to compute

Code Pattern: Inference Server

rust
use std::sync::OnceLock;
use tract_onnx::prelude::*;

static MODEL: OnceLock<SimplePlan<TypedFact, Box<dyn TypedOp>, Graph<TypedFact, Box<dyn TypedOp>>>> = OnceLock::new();

fn get_model() -> &'static SimplePlan<...> {
    MODEL.get_or_init(|| {
        tract_onnx::onnx()
            .model_for_path("model.onnx")
            .unwrap()
            .into_optimized()
            .unwrap()
            .into_runnable()
            .unwrap()
    })
}

async fn predict(input: Vec<f32>) -> anyhow::Result<Vec<f32>> {
    let model = get_model();
    let input = tract_ndarray::arr1(&input).into_shape((1, input.len()))?;
    let result = model.run(tvec!(input.into()))?;
    Ok(result[0].to_array_view::<f32>()?.iter().copied().collect())
}

Code Pattern: Batched Inference

rust
async fn batch_predict(inputs: Vec<Vec<f32>>, batch_size: usize) -> Vec<Vec<f32>> {
    let mut results = Vec::with_capacity(inputs.len());

    for batch in inputs.chunks(batch_size) {
        // Stack inputs into batch tensor
        let batch_tensor = stack_inputs(batch);

        // Run inference on batch
        let batch_output = model.run(batch_tensor).await;

        // Unstack results
        results.extend(unstack_outputs(batch_output));
    }

    results
}

Common Mistakes

MistakeDomain ViolationFix
Clone tensorsMemory wasteUse views
Single inferenceGPU underutilizedBatch processing
Load model per requestSlowSingleton pattern
Sync data loadingGPU idleAsync pipeline

Trace to Layer 1

ConstraintLayer 2 PatternLayer 1 Implementation
Memory efficiencyZero-copyndarray views
Model singletonLazy initOnceLock<Model>
Batch processingChunked iterationchunks() + parallel
GPU asyncConcurrent loadingtokio::spawn + GPU

Related Skills

WhenSee
Performancem10-performance
Lazy initializationm12-lifecycle
Async patternsm07-concurrency
Memory efficiencym01-ownership

按任务相关推荐

来自同一仓库

更多 Skills

全部 Skills
actionbook
社区

domain-embedded

Use when developing embedded/nostd Rust. Keywords: embedded, nostd, microcontroller, MCU, ARM, RISC-V, bare metal, firmware, HAL, PAC, RTIC, embassy, interrupt, DMA, peripheral, GPIO, SPI, I2C, UART, embedded-hal, cortex-m, esp32, stm32, nrf, 嵌入式, 单片机, 固件, 裸机

1小时安装量
1
GitHub Stars
1595
最近更新
8月23日
actionbook
社区

m15-anti-pattern

Use when reviewing code for anti-patterns. Keywords: anti-pattern, common mistake, pitfall, code smell, bad practice, code review, is this an anti-pattern, better way to do this, common mistake to avoid, why is this bad, idiomatic way, beginner mistake, fighting borrow checker, clone everywhere, unwrap in production, should I refactor, 反模式, 常见错误, 代码异味, 最佳实践, 地道写法

1小时安装量
14
GitHub Stars
1554
最近更新
8月23日
actionbook
社区

coding-guidelines

Use when asking about Rust code style or best practices. Keywords: naming, formatting, comment, clippy, rustfmt, lint, code style, best practice, P.NAM, G.FMT, code review, naming convention, variable naming, function naming, type naming, 命名规范, 代码风格, 格式化, 最佳实践, 代码审查, 怎么命名

1小时安装量
2
GitHub Stars
1538
最近更新
8月23日
actionbook
社区

m03-mutability

CRITICAL: Use for mutability issues. Triggers: E0596, E0499, E0502, cannot borrow as mutable, already borrowed as immutable, mut, &mut, interior mutability, Cell, RefCell, Mutex, RwLock, 可变性, 内部可变性, 借用冲突

1小时安装量
2
GitHub Stars
1538
最近更新
8月23日