slm_inference 0.1.0

Backend-agnostic trait layer for running Small Language Model (SLM) inference in Rust.
Documentation 
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43

use crate::SlmModelConfig;
use anyhow::Context as _;
use hf_hub::api::sync::ApiBuilder;
use std::path::PathBuf;
use tracing::debug;

pub struct HfModelInfo {
    pub repo: &'static str,
    pub filename: &'static str,
}

impl HfModelInfo {
    pub fn get_or_download(&self) -> anyhow::Result<PathBuf> {
        get_or_download_model(self.repo, self.filename)
    }

    #[allow(dead_code)]
    pub fn load<Config>(&self, cfg: Config) -> anyhow::Result<Config::Model>
    where
        Config: SlmModelConfig,
    {
        let path = self.get_or_download()?;
        cfg.load_gguf(path).context("Failed to load model")
    }
}

pub fn get_or_download_model(repo: &str, filename: &str) -> anyhow::Result<PathBuf> {
    // This will use default HF cache directory (~/.cache/huggingface/hub)
    let api = ApiBuilder::new()
        .with_progress(true) // Enable progress bars automatically
        .build()?;

    let api_repo = api.model(repo.to_string());

    debug!("Checking for model: {} in repo: {}...", filename, repo);

    // If it's already in the cache, it returns the local path instantly.
    // If not, it starts the download.
    let model_path = api_repo.get(filename)?;

    debug!("Model ready at: {:?}", model_path);
    Ok(model_path)
}