Add a cuda kernel for upsampling. (#441)

* Add a cuda kernel for upsampling.

* Update for the latest tokenizers version.

1 files changed, 2 insertions, 2 deletions

diff --git a/candle-examples/examples/llama/main.rs b/candle-examples/examples/llama/main.rs
index 9a62eba5..b1e112fd 100644
--- a/candle-examples/examples/llama/main.rs
+++ b/candle-examples/examples/llama/main.rs
@@ -223,7 +223,7 @@ fn main() -> Result<()> {
             "{} token: {} '{}'",
             index + 1,
             next_token,
-            tokenizer.decode(vec![next_token], true).map_err(E::msg)?
+            tokenizer.decode(&[next_token], true).map_err(E::msg)?
         );
     }
     let dt = start_gen.elapsed();
@@ -231,7 +231,7 @@ fn main() -> Result<()> {
         "{} tokens generated ({} token/s)\n----\n{}\n----",
         args.sample_len,
         args.sample_len as f64 / dt.as_secs_f64(),
-        tokenizer.decode(new_tokens, true).map_err(E::msg)?
+        tokenizer.decode(&new_tokens, true).map_err(E::msg)?
     );
     Ok(())
 }