diff --git a/rust/src/embeddings/local/qwen3.rs b/rust/src/embeddings/local/qwen3.rs
index 5bb17e35..c903ee01 100644
--- a/rust/src/embeddings/local/qwen3.rs
+++ b/rust/src/embeddings/local/qwen3.rs
@@ -126,12 +126,20 @@ impl Qwen3Embed for Qwen3Embedder {
         for mini_text_batch in text_batch.chunks(batch_size) {
             let (token_ids, attention_mask) =
                 tokenize_batch(&self.tokenizer, mini_text_batch, &self.device)?;
+
             let embeddings: Tensor = {
-                let mut model = self.model.write().map_err(|e| anyhow::anyhow!("Lock poisoned: {}", e))?;
-                model.forward(&token_ids, &attention_mask, 0)?.to_dtype(DType::F32)?
+                let mut model = self
+                    .model
+                    .write()
+                    .map_err(|e| anyhow::anyhow!("Lock poisoned: {}", e))?;
+                let result = model
+                    .forward(&token_ids, &attention_mask, 0)?
+                    .to_dtype(DType::F32)?;
+                model.clear_kv_cache();
+
+                result
             };
 
-            self.model.write().map_err(|e| anyhow::anyhow!("Lock poisoned: {}", e))?.clear_kv_cache();
             let attention_mask = PooledOutputType::from(attention_mask);
             let attention_mask = Some(&attention_mask);
             let model_output = ModelOutput::Tensor(embeddings.clone());