accelerate support for OwlViT (#20411)

* `accelerate` support for `OwlViT`

- added `accelerate` support
- added slow `fp16` tests

* apply suggestions

src/transformers/models/owlvit/modeling_owlvit.py

@@ -434,6 +434,9 @@ class OwlViTAttention(nn.Module):
 
         attn_probs = nn.functional.dropout(attn_weights, p=self.dropout, training=self.training)
 
+        # For int8 compatibility, sometimes the `attn_probs` are in `fp32`
+        attn_probs = attn_probs.to(value_states.dtype)
+
         attn_output = torch.bmm(attn_probs, value_states)
 
         if attn_output.size() != (bsz * self.num_heads, tgt_len, self.head_dim):
@@ -528,6 +531,7 @@ class OwlViTPreTrainedModel(PreTrainedModel):
     base_model_prefix = "owlvit"
     supports_gradient_checkpointing = True
     _keys_to_ignore_on_load_missing = [r"position_ids"]
+    _no_split_modules = ["OwlViTEncoderLayer"]
 
     def _init_weights(self, module):
         """Initialize the weights"""
@@ -836,7 +840,8 @@ class OwlViTTextTransformer(nn.Module):
         # take features from the end of tokens embedding (end of token is the highest number in each sequence)
         # casting to torch.int for onnx compatibility: argmax doesn't support int64 inputs with opset 14
         pooled_output = last_hidden_state[
-            torch.arange(last_hidden_state.shape[0]), input_ids.to(torch.int).argmax(dim=-1)
+            torch.arange(last_hidden_state.shape[0], device=last_hidden_state.device),
+            input_ids.to(torch.int).argmax(dim=-1).to(last_hidden_state.device),
         ]
 
         if not return_dict:
@@ -939,8 +944,13 @@ class OwlViTVisionTransformer(nn.Module):
         )
         return_dict = return_dict if return_dict is not None else self.config.use_return_dict
 
+        # Cast the input to the expected `dtype`
+        expected_input_dtype = self.embeddings.patch_embedding.weight.dtype
+        pixel_values = pixel_values.to(expected_input_dtype)
+
         hidden_states = self.embeddings(pixel_values)
         hidden_states = self.pre_layernorm(hidden_states)
+
         encoder_outputs = self.encoder(
             inputs_embeds=hidden_states,
             output_attentions=output_attentions,
@@ -1193,8 +1203,9 @@ class OwlViTModel(OwlViTPreTrainedModel):
         image_embeds = image_embeds / torch.linalg.norm(image_embeds, ord=2, dim=-1, keepdim=True)
         text_embeds_norm = text_embeds / torch.linalg.norm(text_embeds, ord=2, dim=-1, keepdim=True)
 
-        # cosine similarity as logits
-        logit_scale = self.logit_scale.exp()
+        # cosine similarity as logits and set it on the correct device
+        logit_scale = self.logit_scale.exp().to(image_embeds.device)
+
         logits_per_text = torch.matmul(text_embeds_norm, image_embeds.t()) * logit_scale
         logits_per_image = logits_per_text.t()
 

tests/models/owlvit/test_modeling_owlvit.py

@@ -24,7 +24,7 @@ import numpy as np
 
 import requests
 from transformers import OwlViTConfig, OwlViTTextConfig, OwlViTVisionConfig
-from transformers.testing_utils import require_torch, require_vision, slow, torch_device
+from transformers.testing_utils import require_torch, require_torch_gpu, require_vision, slow, torch_device
 from transformers.utils import is_torch_available, is_vision_available
 
 from ...test_configuration_common import ConfigTester
@@ -778,3 +778,28 @@ class OwlViTModelIntegrationTest(unittest.TestCase):
             [[0.0691, 0.0445, 0.1373], [0.1592, 0.0456, 0.3192], [0.1632, 0.0423, 0.2478]]
         ).to(torch_device)
         self.assertTrue(torch.allclose(outputs.target_pred_boxes[0, :3, :3], expected_slice_boxes, atol=1e-4))
+
+    @slow
+    @require_torch_gpu
+    def test_inference_one_shot_object_detection_fp16(self):
+        model_name = "google/owlvit-base-patch32"
+        model = OwlViTForObjectDetection.from_pretrained(model_name, torch_dtype=torch.float16).to(torch_device)
+
+        processor = OwlViTProcessor.from_pretrained(model_name)
+
+        image = prepare_img()
+        query_image = prepare_img()
+        inputs = processor(
+            images=image,
+            query_images=query_image,
+            max_length=16,
+            padding="max_length",
+            return_tensors="pt",
+        ).to(torch_device)
+
+        with torch.no_grad():
+            outputs = model.image_guided_detection(**inputs)
+
+        # No need to check the logits, we just check inference runs fine.
+        num_queries = int((model.config.vision_config.image_size / model.config.vision_config.patch_size) ** 2)
+        self.assertEqual(outputs.target_pred_boxes.shape, torch.Size((1, num_queries, 4)))