Transformers v5.10

Author: echarlaix

.github/workflows/test_openvino.yml

@@ -39,7 +39,7 @@ jobs:
             "*quantization*",
             "*transformation*",
           ]
-        transformers-version: ["4.57.6", "5.3.0", "latest"]
+        transformers-version: ["5.8.0", "latest"]
 
     runs-on: ubuntu-22.04
 

.github/workflows/test_openvino_slow.yml

@@ -36,7 +36,7 @@ jobs:
       fail-fast: false
       matrix:
         os: ["ubuntu-22.04", "windows-2022"]
-        transformers-version: ["4.57.6", "latest"]
+        transformers-version: ["5.8.0", "latest"]
         include:
           - transformers-version: "main"
             os: "ubuntu-22.04"

optimum/exporters/openvino/convert.py

@@ -1236,7 +1236,9 @@ def get_sd3_models_for_export(pipeline, exporter, int_dtype, float_dtype):
     text_encoder = getattr(pipeline, "text_encoder", None)
     if text_encoder is not None:
         text_encoder.config.output_hidden_states = True
-        text_encoder.text_model.config.output_hidden_states = True
+        # `CLIPTextTransformer` removed since transformers v5.6
+        if hasattr(text_encoder, "text_model"):
+            text_encoder.text_model.config.output_hidden_states = True
         text_encoder_config_constructor = TasksManager.get_exporter_config_constructor(
             model=text_encoder,
             exporter=exporter,
@@ -1298,7 +1300,9 @@ def get_sd3_models_for_export(pipeline, exporter, int_dtype, float_dtype):
     text_encoder_2 = getattr(pipeline, "text_encoder_2", None)
     if text_encoder_2 is not None:
         text_encoder_2.config.output_hidden_states = True
-        text_encoder_2.text_model.config.output_hidden_states = True
+        # `CLIPTextTransformer` removed since transformers v5.6
+        if hasattr(text_encoder_2, "text_model"):
+            text_encoder_2.text_model.config.output_hidden_states = True
         export_config_constructor = TasksManager.get_exporter_config_constructor(
             model=text_encoder_2,
             exporter=exporter,

optimum/exporters/openvino/model_patcher.py

@@ -4714,6 +4714,10 @@ def gemma4_text_attention_forward(
 ) -> tuple:
     from transformers.models.gemma4.modeling_gemma4 import apply_rotary_pos_emb as apply_rotary_pos_emb_gemma4
 
+    # since transformers >= v5.6 (PR #45788) `shared_kv_states` dict and is passed and `kv_shared_layer_index` removed
+    shared_kv_states = kwargs.pop("shared_kv_states", None)
+    legacy_shared_kv_states = hasattr(self, "kv_shared_layer_index")
+
     input_shape = hidden_states.shape[:-1]
     hidden_shape = (*input_shape, -1, self.head_dim)
 
@@ -4724,8 +4728,12 @@ def gemma4_text_attention_forward(
     query_states = apply_rotary_pos_emb_gemma4(query_states, cos, sin, unsqueeze_dim=2)
     query_states = query_states.transpose(1, 2)
 
-    if self.is_kv_shared_layer and past_key_values is not None:
-        key_states, value_states = past_key_values.shared_layers[self.kv_shared_layer_index]
+    if self.is_kv_shared_layer and (not legacy_shared_kv_states or past_key_values is not None):
+        if legacy_shared_kv_states:
+            key_states, value_states = past_key_values.shared_layers[self.kv_shared_layer_index]
+        else:
+            key_states, value_states = shared_kv_states[self.layer_type]
+
         key_states = key_states.to(query_states.device)
         value_states = value_states.to(query_states.device)
     else:
@@ -4740,18 +4748,26 @@ def gemma4_text_attention_forward(
         value_states = value_states.transpose(1, 2)
 
     if past_key_values is not None:
-        cache_kwargs = {
-            "sin": sin,
-            "cos": cos,
-            "cache_position": cache_position,
-            "sliding_window": self.sliding_window,
-        }
-        if not self.is_kv_shared_layer:
-            key_states, value_states = past_key_values.update(key_states, value_states, self.layer_idx, cache_kwargs)
-        if self.store_full_length_kv:
-            if not hasattr(past_key_values, "shared_layers"):
-                past_key_values.shared_layers = {}
-            past_key_values.shared_layers[self.layer_idx] = key_states, value_states
+        if legacy_shared_kv_states:
+            cache_kwargs = {
+                "sin": sin,
+                "cos": cos,
+                "cache_position": cache_position,
+                "sliding_window": self.sliding_window,
+            }
+            if not self.is_kv_shared_layer:
+                key_states, value_states = past_key_values.update(
+                    key_states, value_states, self.layer_idx, cache_kwargs
+                )
+            if self.store_full_length_kv:
+                if not hasattr(past_key_values, "shared_layers"):
+                    past_key_values.shared_layers = {}
+                past_key_values.shared_layers[self.layer_idx] = key_states, value_states
+        else:
+            if not self.is_kv_shared_layer:
+                key_states, value_states = past_key_values.update(key_states, value_states, self.layer_idx)
+            if self.store_full_length_kv and shared_kv_states is not None:
+                shared_kv_states[self.layer_type] = key_states, value_states
 
     attention_interface = gemma4_eager_attention_forward_patched
 

optimum/intel/openvino/modeling_visual_language.py

@@ -2768,9 +2768,12 @@ def rot_pos_emb(self, grid_thw):
             wpos_ids = wpos_ids.flatten()
             pos_ids.append(torch.stack([hpos_ids, wpos_ids], dim=-1).repeat(t, 1))
         pos_ids = torch.cat(pos_ids, dim=0)
-        max_grid_size = grid_thw[:, 1:].max()
-        rotary_pos_emb_full = self._rotary_pos_emb(max_grid_size)
-        rotary_pos_emb = rotary_pos_emb_full[pos_ids].flatten(1)
+        if is_transformers_version(">=", "5.9"):
+            rotary_pos_emb = self._rotary_pos_emb(pos_ids)
+        else:
+            max_grid_size = grid_thw[:, 1:].max()
+            rotary_pos_emb_full = self._rotary_pos_emb(max_grid_size)
+            rotary_pos_emb = rotary_pos_emb_full[pos_ids].flatten(1)
         return rotary_pos_emb
 
     def get_multimodal_embeddings(

setup.py

@@ -29,8 +29,8 @@
 INSTALL_REQUIRE = [
     "torch>=2.1",
     "safetensors<0.8.0",
-    "optimum~=2.2.0",
-    "transformers>=4.57,<5.6",
+    "optimum@git+https://github.com/huggingface/optimum.git@transformers-v5.10",
+    "transformers>=5.8,<5.11",
     "setuptools",
     "huggingface-hub>=0.23.2,<2.0",
     "nncf>=2.19.0",