Repeat-penalty in the falcon example. (#634)

candle-examples/examples/falcon/main.rs

@@ -22,6 +22,8 @@ struct TextGeneration {
     device: Device,
     tokenizer: Tokenizer,
     logits_processor: LogitsProcessor,
+    repeat_penalty: f32,
+    repeat_last_n: usize,
 }
 
 impl TextGeneration {
@@ -31,6 +33,8 @@ impl TextGeneration {
         seed: u64,
         temp: Option<f64>,
         device: &Device,
+        repeat_penalty: f32,
+        repeat_last_n: usize,
     ) -> Self {
         let logits_processor = LogitsProcessor::new(seed, temp);
         Self {
@@ -38,6 +42,8 @@ impl TextGeneration {
             tokenizer,
             logits_processor,
             device: device.clone(),
+            repeat_penalty,
+            repeat_last_n,
         }
     }
 
@@ -63,6 +69,16 @@ impl TextGeneration {
             let input = Tensor::new(ctxt, &self.device)?.unsqueeze(0)?;
             let logits = self.model.forward(&input)?;
             let logits = logits.squeeze(0)?.to_dtype(DType::F32)?;
+            let logits = if self.repeat_penalty == 1. {
+                logits
+            } else {
+                let start_at = tokens.len().saturating_sub(self.repeat_last_n);
+                candle_transformers::utils::apply_repeat_penalty(
+                    &logits,
+                    self.repeat_penalty,
+                    &tokens[start_at..],
+                )?
+            };
 
             let next_token = self.logits_processor.sample(&logits)?;
             tokens.push(next_token);
@@ -116,6 +132,14 @@ struct Args {
 
     #[arg(long, default_value = "refs/pr/43")]
     revision: String,
+
+    /// Penalty to be applied for repeating tokens, 1. means no penalty.
+    #[arg(long, default_value_t = 1.0)]
+    repeat_penalty: f32,
+
+    /// The context size to consider for the repeat penalty.
+    #[arg(long, default_value_t = 64)]
+    repeat_last_n: usize,
 }
 
 fn main() -> Result<()> {
@@ -162,7 +186,15 @@ fn main() -> Result<()> {
     let model = Falcon::load(vb, config)?;
     println!("loaded the model in {:?}", start.elapsed());
 
-    let mut pipeline = TextGeneration::new(model, tokenizer, args.seed, args.temperature, &device);
+    let mut pipeline = TextGeneration::new(
+        model,
+        tokenizer,
+        args.seed,
+        args.temperature,
+        &device,
+        args.repeat_penalty,
+        args.repeat_last_n,
+    );
     pipeline.run(&args.prompt, args.sample_len)?;
     Ok(())
 }