hetki ennen webgpu inferenssiä

network-poc/node/src/smollm.rs

@@ -196,7 +196,7 @@ pub async fn run_smollm_inference(prompt: String, ws: Rc<RefCell<WebSocket>>) {
     } else {
         logits
     };
-    let mut next_token = logits.argmax(0).unwrap().to_vec0::<u32>().unwrap();
+    let mut next_token = crate::sampling::sample_top_k(&logits, 10, 5.0);
     console_log!("[SmolLM] Ensimmäinen generoitu token: {}", next_token);
     pos = input_len;
 
@@ -229,7 +229,7 @@ pub async fn run_smollm_inference(prompt: String, ws: Rc<RefCell<WebSocket>>) {
         } else {
             logits
         };
-        next_token = logits.argmax(0).unwrap().to_vec0::<u32>().unwrap();
+        next_token = crate::sampling::sample_top_k(&logits, 10, 5.0);
         pos += 1;
 
         if next_token == 2 { break; }