Skip to main content

models/
models.rs

1//! Model download and management - self-contained example.
2//!
3//! Run: `cargo run --example models`
4
5use anyhow::Result;
6use qmd::{
7    EmbeddingEngine, GenerationEngine, RerankEngine,
8    config::get_model_cache_dir,
9    llm::{DEFAULT_EMBED_MODEL, DEFAULT_EMBED_MODEL_URI, list_cached_models, model_exists},
10    pull_model, resolve_model,
11};
12
13fn main() -> Result<()> {
14    // Cache directory
15    println!("Cache: {}\n", get_model_cache_dir().display());
16
17    // Cached models
18    println!("Cached models:");
19    let models = list_cached_models();
20    if models.is_empty() {
21        println!("  (none - will download)");
22    } else {
23        for m in &models {
24            println!("  {}", m);
25        }
26    }
27
28    // Model availability check
29    println!("\nAvailability:");
30    let check = |name: &str, ok: bool| println!("  {}: {}", name, if ok { "yes" } else { "no" });
31    check("embed model", model_exists(DEFAULT_EMBED_MODEL));
32    check("EmbeddingEngine", EmbeddingEngine::load_default().is_ok());
33    check("GenerationEngine", GenerationEngine::is_available());
34    check("RerankEngine", RerankEngine::is_available());
35
36    // Download model (demo)
37    println!("\nDownloading embedding model...");
38    let result = pull_model(DEFAULT_EMBED_MODEL_URI, false)?;
39    println!("  Path: {}", result.path.display());
40    println!("  Size: {} bytes", result.size_bytes);
41    println!("  Refreshed: {}", result.refreshed);
42
43    // URI resolution
44    println!("\nURI resolution:");
45    let uri = "hf:ggml-org/embeddinggemma-300M-GGUF/embeddinggemma-300M-Q8_0.gguf";
46    if let Ok(path) = resolve_model(uri) {
47        println!("  {} ->\n  {}", uri, path.display());
48    }
49
50    // Load and test
51    let mut engine = EmbeddingEngine::new(&result.path)?;
52    let emb = engine.embed("test embedding")?;
53    println!("\nModel info:");
54    println!("  Name: {}", emb.model);
55    println!("  Dimensions: {}", emb.embedding.len());
56
57    Ok(())
58}