Avoid re-encoding the input in the T5 example. (#875)

1 files changed, 2 insertions, 1 deletions

diff --git a/candle-examples/examples/t5/main.rs b/candle-examples/examples/t5/main.rs
index 72be23bc..36cbee7c 100644
--- a/candle-examples/examples/t5/main.rs
+++ b/candle-examples/examples/t5/main.rs
@@ -171,6 +171,7 @@ fn main() -> Result<()> {
                     Some(args.temperature)
                 };
                 let mut logits_processor = LogitsProcessor::new(299792458, temperature, args.top_p);
+                let encoder_output = model.encode(&input_token_ids)?;
                 let start = std::time::Instant::now();
 
                 for index in 0.. {
@@ -184,7 +185,7 @@ fn main() -> Result<()> {
                         Tensor::new(&[last_token], device)?.unsqueeze(0)?
                     };
                     let logits = model
-                        .forward(&input_token_ids, &decoder_token_ids)?
+                        .decode(&decoder_token_ids, &encoder_output)?
                         .squeeze(0)?;
                     let logits = if args.repeat_penalty == 1. {
                         logits