generate code refactoring progress 1/3

tests/test_minicpm5.rs

@@ -1,10 +1,11 @@
-use std::time::Instant;
+use std::{pin::pin, time::Instant};
 
 use aha::{
     models::{GenerateModel, minicpm5::generate::MiniCPM5GenerateModel},
     params::chat::ChatCompletionParameters,
 };
 use anyhow::Result;
+use rocket::futures::StreamExt;
 #[test]
 fn minicpm5_generate() -> Result<()> {
     // test with cuda: RUST_BACKTRACE=1 cargo test -F cuda --test test_minicpm5 minicpm5_generate -r -- --nocapture
@@ -16,7 +17,7 @@ fn minicpm5_generate() -> Result<()> {
     {
         "temperature": 0.3,
         "top_p": 0.8,
-        "model": "minicpm4",
+        "model": "minicpm5",
         "messages": [
             {
                 "role": "user",
@@ -39,3 +40,41 @@ fn minicpm5_generate() -> Result<()> {
     }
     Ok(())
 }
+
+#[tokio::test]
+async fn minicpm5_stream() -> Result<()> {
+    // test with cuda: RUST_BACKTRACE=1 cargo test -F cuda --test test_minicpm5 minicpm5_stream -r -- --nocapture
+
+    let save_dir =
+        aha::utils::get_default_save_dir().ok_or(anyhow::anyhow!("Failed to get save dir"))?;
+    let model_path = format!("{}/OpenBMB/MiniCPM5-1B/", save_dir);
+
+    let message = r#"
+    {
+        "model": "minicpm5",
+        "messages": [
+            {
+                "role": "user",
+                "content": "什么是AI"
+            }
+        ],
+        "enable_thinking": true
+    }
+    "#;
+    let mes: ChatCompletionParameters = serde_json::from_str(message)?;
+    let i_start = Instant::now();
+    let mut model = MiniCPM5GenerateModel::init(&model_path, None, None)?;
+    let i_duration = i_start.elapsed();
+    println!("Time elapsed in load model is: {:?}", i_duration);
+
+    let i_start = Instant::now();
+    let mut stream = pin!(model.generate_stream(mes)?);
+    while let Some(item) = stream.next().await {
+        println!("generate: \n {:?}", item);
+    }
+
+    let i_duration = i_start.elapsed();
+    println!("Time elapsed in generate is: {:?}", i_duration);
+
+    Ok(())
+}

tests/test_qwen3.rs

@@ -42,7 +42,7 @@ fn qwen3_0_6b_generate() -> Result<()> {
 
 #[tokio::test]
 async fn qwen3_0_6b_stream() -> Result<()> {
-    // test with cuda: RUST_BACKTRACE=1 cargo test -F cuda qwen3_0_6b_stream -r -- --nocapture
+    // test with cuda: RUST_BACKTRACE=1 cargo test -F cuda --test test_qwen3 qwen3_0_6b_stream -r -- --nocapture
 
     let save_dir =
         aha::utils::get_default_save_dir().ok_or(anyhow::anyhow!("Failed to get save dir"))?;