diff --git a/packages/transformers/src/generation/logits_process.js b/packages/transformers/src/generation/logits_process.js
index 647a30806..473c4f798 100644
--- a/packages/transformers/src/generation/logits_process.js
+++ b/packages/transformers/src/generation/logits_process.js
@@ -304,7 +304,6 @@ export class WhisperTimeStampLogitsProcessor extends LogitsProcessor {
 
             if (input_ids[i].length === this.begin_index) {
                 batch_logits_data.subarray(0, this.timestamp_begin).fill(-Infinity);
-                continue;
             }
 
             // timestamps have to appear in pairs, except directly before eos_token; mask logits accordingly
diff --git a/packages/transformers/src/models/whisper/modeling_whisper.js b/packages/transformers/src/models/whisper/modeling_whisper.js
index d0d6f0b5a..9b4f9689a 100644
--- a/packages/transformers/src/models/whisper/modeling_whisper.js
+++ b/packages/transformers/src/models/whisper/modeling_whisper.js
@@ -201,7 +201,9 @@ export class WhisperForConditionalGeneration extends WhisperPreTrainedModel {
 
         // input_features shape: [batch=1, n_mels, total_frames]
         const input_features = inputs;
-        const total_frames = input_features.dims[2];
+        const total_frames = generation_config.num_frames
+            ? Math.min(input_features.dims[2], generation_config.num_frames)
+            : input_features.dims[2];
 
         // The encoder downsamples by input_stride (=2 for whisper), so:
         //   num_segment_frames = input_stride * max_source_positions = 3000 mel frames per segment
@@ -343,6 +345,7 @@ export class WhisperForConditionalGeneration extends WhisperPreTrainedModel {
             if (seek_token_timestamps) {
                 allTokenTimestamps.push(...seek_token_timestamps.slice(0, tokens_to_keep));
             }
+            if (segment_offset <= 0) break;
             seek += segment_offset;
         }