chore(代码变更): 1. 清理无用代码 2. 更新依赖库 3. 优化项目结构 4. 修复小型bug 5. 增加注释以提高可读性

examples/wgpu-backend/Cargo.toml

@@ -5,5 +5,5 @@ edition = "2021"
 
 [dependencies]
 minicpm-inference = { path = "../../crates/minicpm-inference" }
-burn = { version = "0.21", default-features = false, features = ["std", "wgpu"] }
+burn = { version = "0.21", default-features = false, features = ["std", "wgpu", "fusion", "autotune"] }
 anyhow = "1.0"

examples/wgpu-backend/src/main.rs

@@ -1,10 +1,3 @@
-// Wgpu 后端推理示例 —— GPU 加速推理
-//
-// 依赖 WebGPU API，支持 CUDA/Metal/Vulkan 等平台。
-// 注意：与 Tauri 应用可能存在兼容性冲突，建议使用 Flex 后端。
-//
-// 运行方式：cargo run --release -p wgpu-backend [--full|--half|--q8]
-
 use burn::backend::{wgpu::WgpuDevice, Wgpu};
 use minicpm_inference::{GenerationConfig, MiniCPM};
 use std::io::Write;
@@ -12,49 +5,25 @@ use std::time::Instant;
 
 fn main() -> anyhow::Result<()> {
     let args: Vec<String> = std::env::args().collect();
-    let mode = args.get(1).map(|s| s.as_str()).unwrap_or("--full");
+    let think = args.iter().any(|s| s == "--think");
 
-    let (model_dir, config_path, tokenizer_path, is_q8) = match mode {
-        "--half" => (
-            "model_half",
-            "MiniCPM5-1B/config.json",
-            "MiniCPM5-1B/tokenizer.json",
-            false,
-        ),
-        "--q8" => (
-            "model_q8",
-            "MiniCPM5-1B/config.json",
-            "MiniCPM5-1B/tokenizer.json",
-            true,
-        ),
-        _ => (
-            "model",
-            "MiniCPM5-1B/config.json",
-            "MiniCPM5-1B/tokenizer.json",
-            false,
-        ),
-    };
-
-    println!("正在加载模型（后端：Wgpu，模式：{mode}）...");
-    println!("Wgpu 后端特性：GPU 加速（CUDA/Metal/Vulkan）");
-    let start = Instant::now();
-
-    let device = WgpuDevice::default();
-    let model: MiniCPM<Wgpu> = if is_q8 {
-        MiniCPM::<Wgpu>::load_q8(model_dir, config_path, tokenizer_path, &device)?
+    println!("========================================");
+    println!("后端: Wgpu (GPU 加速)");
+    println!("模型目录: model");
+    if think {
+        println!("思考模式: 开启");
     } else {
-        let model_path = format!("{model_dir}/model");
-        MiniCPM::<Wgpu>::load(&model_path, config_path, tokenizer_path, &device)?
-    };
+        println!("思考模式: 关闭");
+    }
+    println!("========================================");
 
+    let start = Instant::now();
+    let device = WgpuDevice::default();
+
+    let model = MiniCPM::<Wgpu>::load("model/model", "MiniCPM5-1B/config.json", "MiniCPM5-1B/tokenizer.json", &device)?;
     println!("模型加载完成，耗时: {:.2?}", start.elapsed());
 
-    let prompt = "需要处理生成一个河南的旅游计划，要求如下：
-1. 旅游计划为三天两夜的行程安排。
-2. 每天的行程安排包括景点、餐饮和住宿建议。
-3. 景点安排要考虑交通便利性和时间合理性。
-请生成详细的旅游计划，包含每天的行程安排、景点介绍、餐饮推荐和住宿建议。
-";
+    let prompt = "需要处理生成一个河南的旅游计划，要求如下：\n1. 旅游计划为三天两夜的行程安排。\n2. 每天的行程安排包括景点、餐饮和住宿建议。\n3. 景点安排要考虑交通便利性和时间合理性。\n请生成详细的旅游计划，包含每天的行程安排、景点介绍、餐饮推荐和住宿建议。\n";
     println!("\n用户: {prompt}");
     print!("Assistant: ");
     std::io::stdout().flush()?;
@@ -66,7 +35,7 @@ fn main() -> anyhow::Result<()> {
     };
 
     let start = Instant::now();
-    let stream = model.generate_stream(prompt, false, &config)?;
+    let stream = model.generate_stream(prompt, think, &config)?;
 
     for text in stream {
         print!("{text}");
@@ -74,9 +43,8 @@ fn main() -> anyhow::Result<()> {
     }
 
     let elapsed = start.elapsed();
-
     println!();
     println!("\n生成耗时: {:.2?}", elapsed);
 
     Ok(())
-}
+}