[Gemma] Fix eager attention (#29187)

* fix modelling code

* add tests

* fix tests

* add some logit tests

* style

* fix fix

src/transformers/models/gemma/modeling_gemma.py

@@ -276,7 +276,7 @@ class GemmaAttention(nn.Module):
 
         attn_output = attn_output.transpose(1, 2).contiguous()
 
-        attn_output = attn_output.reshape(bsz, q_len, self.hidden_size)
+        attn_output = attn_output.view(bsz, q_len, -1)
         attn_output = self.o_proj(attn_output)
 
         if not output_attentions:

tests/models/gemma/test_modeling_gemma.py

@@ -26,6 +26,7 @@ from transformers.testing_utils import (
     require_flash_attn,
     require_torch,
     require_torch_gpu,
+    require_torch_sdpa,
     slow,
     torch_device,
 )
@@ -460,6 +461,71 @@ class GemmaModelTest(ModelTesterMixin, GenerationTesterMixin, PipelineTesterMixi
     def test_flash_attn_2_inference_padding_right(self):
         self.skipTest("Gemma flash attention does not support right padding")
 
+    @require_torch_sdpa
+    @require_torch_gpu
+    @slow
+    def test_sdpa_equivalence(self):
+        for model_class in self.all_model_classes:
+            if not model_class._supports_sdpa:
+                return
+
+            config, inputs_dict = self.model_tester.prepare_config_and_inputs_for_common()
+            model = model_class(config)
+
+            with tempfile.TemporaryDirectory() as tmpdirname:
+                model.save_pretrained(tmpdirname)
+                model_sdpa = model_class.from_pretrained(
+                    tmpdirname, torch_dtype=torch.float16, attn_implementation="sdpa"
+                )
+                model_sdpa.to(torch_device)
+
+                model = model_class.from_pretrained(tmpdirname, torch_dtype=torch.float16, attn_implementation="eager")
+                model.to(torch_device)
+
+                dummy_input = inputs_dict[model_class.main_input_name]
+                dummy_input = dummy_input.to(torch_device)
+                outputs = model(dummy_input, output_hidden_states=True)
+                outputs_sdpa = model_sdpa(dummy_input, output_hidden_states=True)
+
+                logits = outputs.hidden_states[-1]
+                logits_sdpa = outputs_sdpa.hidden_states[-1]
+
+                # gemma sdpa needs a high tolerance
+                assert torch.allclose(logits_sdpa, logits, atol=3e-3)
+
+    @require_flash_attn
+    @require_torch_gpu
+    @pytest.mark.flash_attn_test
+    @slow
+    def test_flash_attn_2_equivalence(self):
+        for model_class in self.all_model_classes:
+            if not model_class._supports_flash_attn_2:
+                return
+
+            config, inputs_dict = self.model_tester.prepare_config_and_inputs_for_common()
+            model = model_class(config)
+
+            with tempfile.TemporaryDirectory() as tmpdirname:
+                model.save_pretrained(tmpdirname)
+                model_fa = model_class.from_pretrained(
+                    tmpdirname, torch_dtype=torch.float16, attn_implementation="flash_attention_2"
+                )
+                model_fa.to(torch_device)
+
+                model = model_class.from_pretrained(tmpdirname, torch_dtype=torch.float16, attn_implementation="eager")
+                model.to(torch_device)
+
+                dummy_input = inputs_dict[model_class.main_input_name]
+                dummy_input = dummy_input.to(torch_device)
+                outputs = model(dummy_input, output_hidden_states=True)
+                outputs_fa = model_fa(dummy_input, output_hidden_states=True)
+
+                logits = outputs.hidden_states[-1]
+                logits_fa = outputs_fa.hidden_states[-1]
+
+                # gemma flash attention 2 needs a high tolerance
+                assert torch.allclose(logits_fa, logits, atol=3e-3)
+
 
 @require_torch_gpu
 @slow
@@ -542,6 +608,69 @@ class GemmaIntegrationTest(unittest.TestCase):
 
         self.assertEqual(output_text, EXPECTED_TEXTS)
 
+    def test_model_2b_eager(self):
+        model_id = "google/gemma-2b"
+        EXPECTED_TEXTS = [
+            "Hello I am doing a project on the 1990s and I am looking for some information on the ",
+            "Hi today I am going to share with you a very easy and simple recipe of <strong><em>Kaju Kat",
+        ]
+
+        model = AutoModelForCausalLM.from_pretrained(
+            model_id, low_cpu_mem_usage=True, torch_dtype=torch.bfloat16, attn_implementation="eager"
+        )
+        model.to(torch_device)
+
+        tokenizer = AutoTokenizer.from_pretrained(model_id)
+        inputs = tokenizer(self.input_text, return_tensors="pt", padding=True).to(torch_device)
+
+        output = model.generate(**inputs, max_new_tokens=20, do_sample=False)
+        output_text = tokenizer.batch_decode(output, skip_special_tokens=True)
+
+        self.assertEqual(output_text, EXPECTED_TEXTS)
+
+    @require_torch_sdpa
+    def test_model_2b_sdpa(self):
+        model_id = "google/gemma-2b"
+        EXPECTED_TEXTS = [
+            "Hello I am doing a project on the 1990s and I need to know what the most popular music",
+            "Hi today I am going to share with you a very easy and simple recipe of <strong><em>Khichdi",
+        ]
+
+        model = AutoModelForCausalLM.from_pretrained(
+            model_id, low_cpu_mem_usage=True, torch_dtype=torch.bfloat16, attn_implementation="sdpa"
+        )
+        model.to(torch_device)
+
+        tokenizer = AutoTokenizer.from_pretrained(model_id)
+        inputs = tokenizer(self.input_text, return_tensors="pt", padding=True).to(torch_device)
+
+        output = model.generate(**inputs, max_new_tokens=20, do_sample=False)
+        output_text = tokenizer.batch_decode(output, skip_special_tokens=True)
+
+        self.assertEqual(output_text, EXPECTED_TEXTS)
+
+    @pytest.mark.flash_attn_test
+    @require_flash_attn
+    def test_model_2b_flash_attn(self):
+        model_id = "google/gemma-2b"
+        EXPECTED_TEXTS = [
+            "Hello I am doing a project on the 1990s and I need to know what the most popular music",
+            "Hi today I am going to share with you a very easy and simple recipe of <strong><em>Kaju Kat",
+        ]
+
+        model = AutoModelForCausalLM.from_pretrained(
+            model_id, low_cpu_mem_usage=True, torch_dtype=torch.bfloat16, attn_implementation="flash_attention_2"
+        )
+        model.to(torch_device)
+
+        tokenizer = AutoTokenizer.from_pretrained(model_id)
+        inputs = tokenizer(self.input_text, return_tensors="pt", padding=True).to(torch_device)
+
+        output = model.generate(**inputs, max_new_tokens=20, do_sample=False)
+        output_text = tokenizer.batch_decode(output, skip_special_tokens=True)
+
+        self.assertEqual(output_text, EXPECTED_TEXTS)
+
     @require_bitsandbytes
     def test_model_2b_4bit(self):
         model_id = "google/gemma-2b"