agentsclimarketplace

Ml deployment

Skill pluginagentmarketplace/custom-plugin-machine-learning/skills/ml-deployment

Deploy ML models to production - APIs, containerization, monitoring, and MLOpsFrom its SKILL.md

Install
npx -y skills add pluginagentmarketplace/custom-plugin-machine-learning --skill ml-deployment

Assembled from the repository path, not quoted from the project. Check it against their README if it does not work.

2 things to look at

  • no licenseNo license file was found in the repository. Code published without one is not open source by default, so using it at work is a question for whoever answers licensing questions where you are.
  • 1 stars1 stars. Stars are a popularity signal and not a quality one, but at this level it is likely that nobody has read this closely except its author, and you would be relying on your own review.

SKILL.md

5.8 KB, as published. Nobody here has run it

ML Deployment Skill

Take models from development to production.

Quick Start

from fastapi import FastAPI
from pydantic import BaseModel
import numpy as np
import joblib

app = FastAPI(title="ML Model API")
model = joblib.load('model.pkl')

class PredictRequest(BaseModel):
    features: list[float]

class PredictResponse(BaseModel):
    prediction: float

@app.post("/predict", response_model=PredictResponse)
async def predict(request: PredictRequest):
    X = np.array([request.features])
    prediction = model.predict(X)[0]
    return PredictResponse(prediction=float(prediction))

@app.get("/health")
async def health():
    return {"status": "healthy"}

Key Topics

1. Model Export

import torch
import torch.onnx

# Export PyTorch to ONNX
def export_to_onnx(model, sample_input, path='model.onnx'):
    model.eval()
    torch.onnx.export(
        model,
        sample_input,
        path,
        export_params=True,
        opset_version=14,
        input_names=['input'],
        output_names=['output'],
        dynamic_axes={'input': {0: 'batch'}, 'output': {0: 'batch'}}
    )

# ONNX inference
import onnxruntime as ort

session = ort.InferenceSession('model.onnx')
input_name = session.get_inputs()[0].name
output = session.run(None, {input_name: input_data})[0]

2. Docker Containerization

# Dockerfile
FROM python:3.10-slim

WORKDIR /app

COPY requirements.txt .
RUN pip install --no-cache-dir -r requirements.txt

COPY . .

EXPOSE 8000

CMD ["uvicorn", "main:app", "--host", "0.0.0.0", "--port", "8000"]
# docker-compose.yml
version: '3.8'
services:
  api:
    build: .
    ports:
      - "8000:8000"
    environment:
      - MODEL_PATH=/models/model.onnx
    volumes:
      - ./models:/models:ro
    restart: unless-stopped

3. Monitoring with Prometheus

from prometheus_client import Counter, Histogram, start_http_server

# Define metrics
REQUESTS = Counter('model_requests_total', 'Total requests', ['status'])
LATENCY = Histogram('model_latency_seconds', 'Latency in seconds')

@app.post("/predict")
async def predict(request: PredictRequest):
    import time
    start = time.time()

    try:
        prediction = model.predict(request.features)
        REQUESTS.labels(status='success').inc()
        return {"prediction": prediction}
    except Exception as e:
        REQUESTS.labels(status='error').inc()
        raise
    finally:
        LATENCY.observe(time.time() - start)

4. Model Versioning

import mlflow

# Log model
with mlflow.start_run():
    mlflow.log_params({"n_estimators": 100, "max_depth": 10})
    mlflow.log_metrics({"accuracy": 0.95, "f1": 0.93})
    mlflow.sklearn.log_model(model, "model")

# Load model
model_uri = "runs:/abc123/model"
model = mlflow.sklearn.load_model(model_uri)

5. A/B Testing

import random

class ABTest:
    def __init__(self, variants: dict[str, float]):
        self.variants = variants  # {"A": 0.5, "B": 0.5}
        self.results = {v: {"count": 0, "success": 0} for v in variants}

    def get_variant(self, user_id: str) -> str:
        random.seed(hash(user_id))
        r = random.random()
        cumulative = 0
        for variant, weight in self.variants.items():
            cumulative += weight
            if r <= cumulative:
                return variant
        return list(self.variants.keys())[-1]

    def record(self, variant: str, success: bool):
        self.results[variant]["count"] += 1
        if success:
            self.results[variant]["success"] += 1

Best Practices

DO

  • Version your models
  • Implement health checks
  • Use async logging
  • Set up monitoring day one
  • Use canary deployments

DON'T

  • Don't deploy without validation
  • Don't skip latency testing
  • Don't ignore drift
  • Don't hard-code configs

Exercises

Exercise 1: FastAPI Service

# TODO: Create a FastAPI service that:
# 1. Loads a model on startup
# 2. Has /predict and /health endpoints
# 3. Validates input with Pydantic

Exercise 2: Docker Deployment

# TODO: Containerize your ML service
# Create Dockerfile and docker-compose.yml

Unit Test Template

import pytest
from fastapi.testclient import TestClient

def test_health_endpoint():
    """Test health check."""
    client = TestClient(app)
    response = client.get("/health")

    assert response.status_code == 200
    assert response.json()["status"] == "healthy"

def test_predict_endpoint():
    """Test prediction."""
    client = TestClient(app)
    response = client.post("/predict", json={"features": [1.0, 2.0, 3.0]})

    assert response.status_code == 200
    assert "prediction" in response.json()

Troubleshooting

ProblemCauseSolution
High latencyModel too largeQuantize, use ONNX
Memory leaksPoor cleanupImplement proper lifecycle
API errorsInput validationAdd Pydantic schemas
Scaling issuesBlocking I/OUse async, add workers

Related Resources

  • Agent: 07-model-deployment
  • Previous: computer-vision
  • Docs: FastAPI

Version: 1.4.0 | Status: Production Ready

Keep looking

Skills are one crate of 608,475. Ordering is by how many stacks a row turns up in, so the top of any crate is what has actually been picked rather than what has the most stars.