feat: drag-and-drop model import and context size configuration

This commit is contained in:
Evandro Fonseca Junior committed 2026-03-26 21:38:58 -03:00
1 parent 18d8b03074
commit d1d77b83ef
4 files changed
+306 -33

No files matched your search

+49 -7
View File
@@ -6,8 +6,6 @@ use tauri::{Manager, State};
pub struct ServerState(pub Mutex<Option<Child>>);
/// Query GPU info via nvidia-smi.
/// Returns (gpu_name, vram_mib, recommended_layers).
/// recommended_layers is estimated as vram_mib / 80 (rough heuristic for 4-bit quant).
#[tauri::command]
fn query_gpu() -> Result<GpuInfo, String> {
let output = Command::new("nvidia-smi")
@@ -28,7 +26,6 @@ fn query_gpu() -> Result<GpuInfo, String> {
return Err("No GPU found".into());
}
// Format: "GPU Name, 8192"
let parts: Vec<&str> = line.splitn(2, ',').collect();
if parts.len() < 2 {
return Err(format!("Unexpected nvidia-smi output: {line}"));
@@ -36,9 +33,6 @@ fn query_gpu() -> Result<GpuInfo, String> {
let name = parts[0].trim().to_string();
let vram_mib: u32 = parts[1].trim().parse().unwrap_or(0);
// Heuristic: ~80 MiB per layer for Q4/Q6 models (conservative)
// Reserve 1 GiB for OS/other usage
let usable = vram_mib.saturating_sub(1024);
let recommended_layers = (usable / 80).min(200) as i32;
@@ -56,12 +50,14 @@ pub struct GpuInfo {
pub recommended_layers: i32,
}
/// Start the inference server with full configuration.
#[tauri::command]
fn start_server(
state: State<ServerState>,
model_path: String,
addr: String,
gpu_layers: i32,
ctx_size: u32,
) -> Result<String, String> {
let mut guard = state.0.lock().map_err(|e| e.to_string())?;
@@ -79,6 +75,7 @@ fn start_server(
.env("LOCALLLM_MODEL", &model_path)
.env("LOCALLLM_ADDR", &addr)
.env("LOCALLLM_GPU_LAYERS", gpu_layers.to_string())
.env("LOCALLLM_CTX", ctx_size.to_string())
.spawn()
.map_err(|e| format!("Failed to start server: {e}"))?;
@@ -116,7 +113,6 @@ fn server_status(state: State<ServerState>) -> String {
}
/// Scan a directory for .gguf model files.
/// Returns a list of { name, path } objects.
#[tauri::command]
fn scan_models(dir: String) -> Result<Vec<ModelFile>, String> {
let path = Path::new(&dir);
@@ -154,6 +150,51 @@ fn scan_models(dir: String) -> Result<Vec<ModelFile>, String> {
Ok(models)
}
/// Copy a .gguf file into the models directory.
/// Returns the new ModelFile on success.
#[tauri::command]
fn import_model(src_path: String, dest_dir: String) -> Result<ModelFile, String> {
let src = Path::new(&src_path);
// Validate source
if !src.exists() {
return Err(format!("Source file not found: {src_path}"));
}
if !src.is_file() {
return Err("Source is not a file".into());
}
let ext = src.extension().unwrap_or_default();
if !ext.eq_ignore_ascii_case("gguf") {
return Err("Only .gguf files are supported".into());
}
// Ensure destination directory exists
let dest_dir_path = Path::new(&dest_dir);
if !dest_dir_path.exists() {
std::fs::create_dir_all(dest_dir_path)
.map_err(|e| format!("Failed to create models directory: {e}"))?;
}
let file_name = src.file_name().unwrap_or_default();
let dest_path = dest_dir_path.join(file_name);
// If already in the same directory, just return it
if src == dest_path {
return Ok(ModelFile {
name: file_name.to_string_lossy().to_string(),
path: dest_path.to_string_lossy().to_string(),
});
}
// Copy the file
std::fs::copy(src, &dest_path).map_err(|e| format!("Failed to copy model: {e}"))?;
Ok(ModelFile {
name: file_name.to_string_lossy().to_string(),
path: dest_path.to_string_lossy().to_string(),
})
}
#[derive(serde::Serialize, Clone)]
pub struct ModelFile {
pub name: String,
@@ -168,6 +209,7 @@ pub fn run() {
.invoke_handler(tauri::generate_handler![
query_gpu,
scan_models,
import_model,
start_server,
stop_server,
server_status
+93
View File
@@ -388,3 +388,96 @@
background: rgba(240,64,64,0.07);
border-color: rgba(240,64,64,0.2);
}
/* ── Drop zone ── */
.sp-drop-zone {
position: relative;
border-radius: var(--radius-sm);
transition: border-color 0.15s, background 0.15s;
}
.sp-drop-zone--active {
outline: 2px dashed var(--accent);
outline-offset: 2px;
background: rgba(91,106,240,0.06);
border-radius: var(--radius-sm);
}
.sp-drop-zone--disabled {
pointer-events: none;
}
.sp-drop-overlay {
position: absolute;
inset: 0;
z-index: 10;
display: flex;
flex-direction: column;
align-items: center;
justify-content: center;
gap: 6px;
background: rgba(91,106,240,0.18);
border-radius: var(--radius-sm);
backdrop-filter: blur(2px);
font-size: 12px;
font-weight: 600;
color: var(--accent);
pointer-events: none;
}
.sp-drop-icon {
font-size: 22px;
}
.sp-drop-hint {
font-size: 10px;
color: var(--text3);
text-align: center;
margin-top: 4px;
padding: 0 2px;
line-height: 1.4;
}
/* ── Context size row ── */
.sp-ctx-row {
display: flex;
flex-direction: column;
gap: 6px;
}
.sp-ctx-presets {
display: flex;
flex-wrap: wrap;
gap: 4px;
}
.sp-ctx-preset {
background: var(--bg4);
border: 1px solid var(--border);
border-radius: 4px;
color: var(--text2);
font-size: 10px;
font-weight: 600;
padding: 3px 7px;
cursor: pointer;
transition: background 0.12s, border-color 0.12s, color 0.12s;
font-family: inherit;
letter-spacing: 0.02em;
}
.sp-ctx-preset:hover:not(:disabled) {
background: rgba(91,106,240,0.15);
border-color: var(--accent);
color: var(--accent);
}
.sp-ctx-preset--active {
background: rgba(91,106,240,0.2);
border-color: var(--accent);
color: var(--accent);
}
.sp-ctx-preset:disabled {
opacity: 0.35;
cursor: not-allowed;
}
+162 -26
View File
@@ -1,4 +1,4 @@
import { useState, useEffect, useCallback } from "react";
import { useState, useEffect, useCallback, useRef } from "react";
import { invoke } from "@tauri-apps/api/core";
import type { ProcessStatus, ServerConfig } from "../hooks/useServerProcess";
import "./ServerPanel.css";
@@ -37,8 +37,8 @@ const STATUS_COLOR: Record<ProcessStatus, string> = {
error: "#ef4444",
};
// Default models directory relative to the repo root
const DEFAULT_MODELS_DIR = "models";
const DEFAULT_CTX = 4096;
export default function ServerPanel({ status, message, onStart, onStop }: Props) {
const [modelsDir, setModelsDir] = useState(
@@ -54,11 +54,18 @@ export default function ServerPanel({ status, message, onStart, onStop }: Props)
const [gpuLayers, setGpuLayers] = useState(
() => parseInt(localStorage.getItem("forgelm_gpu_layers") || "0", 10)
);
const [ctxSize, setCtxSize] = useState(
() => parseInt(localStorage.getItem("forgelm_ctx_size") || String(DEFAULT_CTX), 10)
);
const [gpuInfo, setGpuInfo] = useState<GpuInfo | null>(null);
const [gpuError, setGpuError] = useState<string>("");
const [expanded, setExpanded] = useState(true);
const [scanning, setScanning] = useState(false);
const [scanError, setScanError] = useState("");
const [importing, setImporting] = useState(false);
const [importError, setImportError] = useState("");
const [isDragOver, setIsDragOver] = useState(false);
const dropRef = useRef<HTMLDivElement>(null);
const scanModels = useCallback(async (dir: string) => {
if (!dir.trim()) return;
@@ -67,7 +74,6 @@ export default function ServerPanel({ status, message, onStart, onStop }: Props)
try {
const models = await invoke<ModelFile[]>("scan_models", { dir: dir.trim() });
setScannedModels(models);
// Auto-select first model if none selected or current not in list
if (models.length > 0) {
const current = localStorage.getItem("forgelm_model_path") || "";
const stillValid = models.some((m) => m.path === current);
@@ -83,6 +89,24 @@ export default function ServerPanel({ status, message, onStart, onStop }: Props)
}
}, []);
const importModel = useCallback(async (srcPath: string) => {
setImporting(true);
setImportError("");
try {
const model = await invoke<ModelFile>("import_model", {
srcPath,
destDir: modelsDir.trim() || DEFAULT_MODELS_DIR,
});
// Refresh model list and select the imported model
await scanModels(modelsDir.trim() || DEFAULT_MODELS_DIR);
setSelectedModel(model.path);
} catch (e) {
setImportError(String(e));
} finally {
setImporting(false);
}
}, [modelsDir, scanModels]);
// Detect GPU and scan models on mount
useEffect(() => {
invoke<GpuInfo>("query_gpu")
@@ -102,6 +126,54 @@ export default function ServerPanel({ status, message, onStart, onStop }: Props)
scanModels(modelsDir);
}, []); // eslint-disable-line react-hooks/exhaustive-deps
// Drag-and-drop handlers
const handleDragOver = useCallback((e: React.DragEvent) => {
e.preventDefault();
e.stopPropagation();
if (e.dataTransfer.types.includes("Files")) {
setIsDragOver(true);
}
}, []);
const handleDragLeave = useCallback((e: React.DragEvent) => {
e.preventDefault();
e.stopPropagation();
// Only clear if leaving the drop zone itself (not a child)
if (dropRef.current && !dropRef.current.contains(e.relatedTarget as Node)) {
setIsDragOver(false);
}
}, []);
const handleDrop = useCallback(async (e: React.DragEvent) => {
e.preventDefault();
e.stopPropagation();
setIsDragOver(false);
const files = Array.from(e.dataTransfer.files);
const gguf = files.find((f) => f.name.toLowerCase().endsWith(".gguf"));
if (!gguf) {
setImportError("Only .gguf files are supported");
return;
}
// Tauri exposes the real path via the file object's path property
// In Tauri v2, we can read it from the dataTransfer items
const item = e.dataTransfer.items[files.indexOf(gguf)];
const entry = item?.webkitGetAsEntry?.();
// The actual file path is available as a non-standard property in Tauri's webview
// eslint-disable-next-line @typescript-eslint/no-explicit-any
const filePath = (gguf as any).path as string | undefined;
if (filePath) {
await importModel(filePath);
} else if (entry) {
// Fallback: use the file name to construct path (best effort)
setImportError("Could not read file path. Try using the scan button after copying the file manually.");
} else {
setImportError("Could not read file path from drop event.");
}
}, [importModel]);
const isActive = status === "running" || status === "starting" || status === "stopping";
const maxLayers = gpuInfo ? gpuInfo.recommended_layers : 999;
const canStart = !!selectedModel.trim() && !isActive;
@@ -111,7 +183,8 @@ export default function ServerPanel({ status, message, onStart, onStop }: Props)
localStorage.setItem("forgelm_server_addr", addr);
localStorage.setItem("forgelm_gpu_layers", String(gpuLayers));
localStorage.setItem("forgelm_models_dir", modelsDir);
onStart({ modelPath: selectedModel, addr, gpuLayers });
localStorage.setItem("forgelm_ctx_size", String(ctxSize));
onStart({ modelPath: selectedModel, addr, gpuLayers, ctxSize });
};
const handleLayersChange = (val: number) => {
@@ -119,6 +192,11 @@ export default function ServerPanel({ status, message, onStart, onStop }: Props)
setGpuLayers(clamped);
};
const handleCtxChange = (val: number) => {
const clamped = Math.max(512, Math.min(131072, val));
setCtxSize(clamped);
};
const handleScanDir = () => {
localStorage.setItem("forgelm_models_dir", modelsDir);
scanModels(modelsDir);
@@ -128,6 +206,8 @@ export default function ServerPanel({ status, message, onStart, onStop }: Props)
? Math.min(100, (gpuLayers / gpuInfo.recommended_layers) * 100)
: 0;
const CTX_PRESETS = [512, 1024, 2048, 4096, 8192, 16384, 32768, 65536, 131072];
return (
<div className="server-panel">
{/* Header */}
@@ -187,37 +267,93 @@ export default function ServerPanel({ status, message, onStart, onStop }: Props)
</button>
</div>
{/* Model selector */}
{/* Model selector + drag-and-drop zone */}
<label className="sp-label">
Model
{scannedModels.length > 0 && (
<span className="sp-label-hint"> ({scannedModels.length} found)</span>
)}
</label>
{scannedModels.length > 0 ? (
<select
className="sp-model-select"
value={selectedModel}
onChange={(e) => setSelectedModel(e.target.value)}
disabled={isActive}
>
{scannedModels.map((m) => (
<option key={m.path} value={m.path} title={m.path}>
{m.name}
</option>
))}
</select>
) : (
{/* Drop zone wrapper */}
<div
ref={dropRef}
className={`sp-drop-zone${isDragOver ? " sp-drop-zone--active" : ""}${isActive ? " sp-drop-zone--disabled" : ""}`}
onDragOver={!isActive ? handleDragOver : undefined}
onDragLeave={!isActive ? handleDragLeave : undefined}
onDrop={!isActive ? handleDrop : undefined}
>
{isDragOver && (
<div className="sp-drop-overlay">
<span className="sp-drop-icon">📦</span>
<span>Drop .gguf to import</span>
</div>
)}
{scannedModels.length > 0 ? (
<select
className="sp-model-select"
value={selectedModel}
onChange={(e) => setSelectedModel(e.target.value)}
disabled={isActive}
>
{scannedModels.map((m) => (
<option key={m.path} value={m.path} title={m.path}>
{m.name}
</option>
))}
</select>
) : (
<input
className="sp-input"
type="text"
value={selectedModel}
onChange={(e) => setSelectedModel(e.target.value)}
placeholder="path/to/model.gguf"
disabled={isActive}
/>
)}
{!isActive && (
<div className="sp-drop-hint">
{importing ? "Importing…" : "Drop a .gguf file here to add it"}
</div>
)}
</div>
{scanError && <div className="sp-scan-error">{scanError}</div>}
{importError && <div className="sp-scan-error">{importError}</div>}
{/* Context size */}
<label className="sp-label">
Context size
<span className="sp-label-hint"> (tokens)</span>
</label>
<div className="sp-ctx-row">
<input
className="sp-input"
type="text"
value={selectedModel}
onChange={(e) => setSelectedModel(e.target.value)}
placeholder="path/to/model.gguf"
className="sp-input sp-input-num"
type="number"
min={512}
max={131072}
step={512}
value={ctxSize}
onChange={(e) => handleCtxChange(parseInt(e.target.value, 10) || DEFAULT_CTX)}
disabled={isActive}
/>
)}
{scanError && <div className="sp-scan-error">{scanError}</div>}
<div className="sp-ctx-presets">
{CTX_PRESETS.map((p) => (
<button
key={p}
className={`sp-ctx-preset${ctxSize === p ? " sp-ctx-preset--active" : ""}`}
onClick={() => setCtxSize(p)}
disabled={isActive}
title={`Set context to ${p.toLocaleString()} tokens`}
>
{p >= 1024 ? `${p / 1024}k` : p}
</button>
))}
</div>
</div>
{/* Address */}
<label className="sp-label">Listen address</label>
+2
View File
@@ -7,6 +7,7 @@ export interface ServerConfig {
modelPath: string;
addr: string;
gpuLayers: number;
ctxSize: number;
}
export function useServerProcess() {
@@ -39,6 +40,7 @@ export function useServerProcess() {
modelPath: cfg.modelPath,
addr: cfg.addr,
gpuLayers: cfg.gpuLayers,
ctxSize: cfg.ctxSize,
});
setStatus("running");
setMessage(msg);