Summary

Add ability to publish trained ML models to Hugging Face Hub directly from aprender, with automatic model card generation.

Motivation

Users of aprender (including renacer) need to:

1. Share trained models - Push .apr files to HF Hub for team collaboration
2. Version models - Track model iterations with HF Hub's versioning
3. Auto-generate model cards - Include training metadata, hyperparameters, metrics

Requested API

use aprender::hf_hub::{HfHubClient, ModelCard, PushOptions};

// Push model to HF Hub
let client = HfHubClient::new()?; // Uses HF_TOKEN env var
client.push_to_hub(
    "org/model-name",
    &model,
    PushOptions::default()
        .with_model_card(ModelCard::auto_generate(&metadata))
        .with_commit_message("Add v1.0.0 trained model")
)?;

// Pull model from HF Hub  
let model: KMeansModel = client.pull_from_hub("org/model-name")?;

Model Card Auto-Generation

Should include:

• Model type (KMeans, IsolationForest, etc.)
• Training timestamp
• Hyperparameters used
• Training sample count
• Silhouette score / other metrics
• aprender version

Example Model Card Output

---
library_name: aprender
tags:
  - anomaly-detection
  - syscall-analysis
  - rust
---

# Model Card: syscall-anomaly-detector

## Model Details
- **Type:** KMeans Clustering
- **Clusters:** 5
- **Features:** 3 (count, total_time, avg_time)
- **Trained:** 2024-01-15T10:30:00Z
- **aprender version:** 0.10.0

## Training Data
- **Samples:** 15,234 syscall records
- **Source:** renacer trace of `cargo build`

## Metrics
- **Silhouette Score:** 0.847
- **Inertia:** 1234.56

## Usage

\`\`\`rust
use aprender::hf_hub::HfHubClient;

let model = HfHubClient::new()?.pull_from_hub("paiml/syscall-anomaly")?;
\`\`\`

Downstream Consumer

renacer will add CLI flags:

• --push-model hub:org/repo - Push trained model to HF Hub
• --load-model hub:org/repo - Load model from HF Hub

References

• hf-hub crate - Rust HF Hub client
• huggingface_hub Python - Reference API

🤖 Generated with Claude Code