add test

tests/models/whisper/test_modeling_whisper.py

@@ -1820,6 +1820,31 @@ def test_custom_4d_attention_mask(self):
         normalized_1 = torch.nn.functional.softmax(out_shared_prefix_last_tokens)
         torch.testing.assert_close(normalized_0, normalized_1, rtol=1e-3, atol=1e-4)
 
+    def test_generate_output_type(self):
+        for model_class in self.all_generative_model_classes:
+            config, inputs = self.model_tester.prepare_config_and_inputs()
+            model = model_class(config).to(torch_device).eval()
+
+            # short-form generation without fallback
+            pred_ids = model.generate(**inputs)
+            assert isinstance(pred_ids, torch.Tensor)
+
+            # short-form generation with fallback
+            pred_ids = model.generate(**inputs, logprob_threshold=-1.0, temperature=[0.0, 0.1])
+            assert isinstance(pred_ids, torch.Tensor)
+
+            # create artificial long-form inputs
+            inputs["input_features"] = torch.cat([inputs["input_features"], inputs["input_features"]], dim=-1)
+            inputs["attention_mask"] = torch.ones(inputs["input_features"].shape[:2], dtype=torch.int, device=inputs["input_features"].device)
+            model.generation_config.no_timestamps_token_id = model.generation_config.decoder_start_token_id
+
+            # long-form generation without fallback
+            pred_ids = model.generate(**inputs)
+            assert isinstance(pred_ids, torch.Tensor)
+
+            # short-form generation with fallback
+            pred_ids = model.generate(**inputs, logprob_threshold=-1.0, temperature=[0.0, 0.1])
+            assert isinstance(pred_ids, torch.Tensor)
 
 @require_torch
 @require_torchaudio