llmtop 0.1.0

Realtime TUI monitor for local LLM servers (ollama, llama.cpp). The only GPU monitor that knows what model is running and how much each token costs you in energy and dollar-equivalent.
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
44
45
46
47
48
49

//! Collector layer. Day 1 = stub data so TUI renders.
//! Real implementations land in nvidia.rs / apple.rs / ollama.rs in following days.

pub mod nvidia;
pub mod apple;
pub mod ollama;

#[derive(Debug, Clone, Default)]
pub struct HardwareSnapshot {
    /// Aggregated label. Single GPU: model name. Homogeneous cluster: "Nx <model>".
    /// Mixed cluster: "Nx GPU (mixed)".
    pub gpu_name: String,
    pub gpu_count: u32,
    /// Average across GPUs (0..=100).
    pub gpu_util_pct: u32,
    /// Sum across GPUs.
    pub vram_total_mb: u64,
    pub vram_used_mb: u64,
    /// Sum across GPUs.
    pub power_w: f64,
    pub power_limit_w: f64,
    /// Max (hottest) across GPUs.
    pub temp_c: u32,
}

#[derive(Debug, Clone)]
pub struct ModelInfo {
    pub name: String,
    pub vram_mb: u64,
    pub tokens_per_sec: f64,
    /// Monotonic counter — used to derive per-tick deltas for session totals.
    pub total_tokens: u64,
}

/// Hardware poll — tries NVIDIA, then Apple. Returns default when no backend present.
pub async fn poll_hardware() -> HardwareSnapshot {
    if let Some(s) = nvidia::poll() {
        return s;
    }
    #[cfg(target_os = "macos")]
    if let Some(s) = apple::poll() {
        return s;
    }
    HardwareSnapshot::default()
}

pub async fn poll_ollama(base_url: &str, sink: Option<&crate::proxy::Sink>) -> Vec<ModelInfo> {
    ollama::poll(base_url, sink).await
}