fix compliation issue

- Update codes according to transformers version

Signed-off-by: jiseong.oh <jiseong.oh@samsung.com>

Author: Jiseong-oh

.github/workflows/lint.yml

@@ -58,7 +58,7 @@ jobs:
           pip install -r requirements-lintrunner.txt
           USE_CPP=0 pip install --no-build-isolation third-party/ao
           # Match requirements-examples.txt for the model APIs checked by mypy.
-          pip install pytest numpy parameterized huggingface_hub "transformers==5.0.0rc1" timm expecttest types-requests tosa-tools
+          pip install pytest numpy parameterized huggingface_hub "transformers==5.11.0" timm expecttest types-requests tosa-tools
 
       - name: Generate mypy stubs for C++ bindings
         run: |

backends/arm/test/models/Gemma3n/test_gemma3nModel.py

@@ -218,6 +218,7 @@ def forward(
         attn_output, _ = self.attn(
             hidden_states=hidden_states,
             position_embeddings=(cos, sin),
+            attention_mask=None,
         )
         return attn_output
 

backends/arm/test/models/Qwen3_VL/test_qwen3_vl_layers.py

@@ -156,16 +156,16 @@ def __init__(self, config) -> None:
         head_dim = config.vision_config.hidden_size // config.vision_config.num_heads
         self.rotary = Qwen3VLVisionRotaryEmbedding(head_dim // 2)
 
-    def forward(self, max_hw: int) -> torch.Tensor:
-        return self.rotary(max_hw)
+    def forward(self, position_ids: torch.Tensor) -> torch.Tensor:
+        return self.rotary(position_ids)
 
     @classmethod
     def prepare_model_and_inputs(cls):
         config = _make_qwen3_vl_2b_instruct_layer_config()
         grid_thw = _make_image_grid_thw(torch.device("cpu"))
         max_hw = int(grid_thw[:, 1:].max().item())
         model = cls(config).eval()
-        return model, (max_hw,)
+        return model, (torch.arange(max_hw),)
 
 
 class VisionRotaryApplyModel(Qwen3VLTestModule):

backends/native/examples/audio/export_hf_whisper.py

@@ -195,6 +195,8 @@ def forward(
         cache_position: torch.Tensor,
         *cross_kv: torch.Tensor,
     ) -> torch.Tensor:
+        for layer in self.self_cache.layers:
+            layer.cumulative_length = cache_position[:1].clone()
         num_layers = len(self.cross_cache.layers)
         for layer_idx, layer in enumerate(self.cross_cache.layers):
             layer.keys = cross_kv[layer_idx]

backends/native/examples/hf_llm/export_hf_llm.py

@@ -44,7 +44,6 @@
     PreTrainedModel,
 )
 from transformers.cache_utils import StaticLayer
-from transformers.integrations.accelerate import init_empty_weights
 from transformers.integrations.executorch import TorchExportableModuleForDecoderOnlyLM
 from transformers.integrations.sdpa_attention import sdpa_attention_forward
 from transformers.masking_utils import (
@@ -64,6 +63,9 @@ def _load(
             text_config.layer_types = text_config.layer_types[:layers]
     kwargs = {"torch_dtype": torch.float32, "attn_implementation": "sdpa"}
     if empty_weights:
+        # transformers 5.11 no longer ships init_empty_weights; only --gguf needs it.
+        from accelerate import init_empty_weights
+
         with init_empty_weights():
             model = AutoModelForCausalLM.from_config(config, **kwargs)
     elif random_weights:
@@ -161,7 +163,9 @@ def _store_kv_as(self, model: PreTrainedModel, dtype: torch.dtype) -> None:
             if type(old) is not StaticLayer:
                 raise ValueError(f"cannot change the KV dtype of {type(old).__name__}")
             layer = _CastingStaticLayer(old.max_cache_len)
-            layer.lazy_initialization(old.keys[:, :, :1].to(dtype))
+            layer.lazy_initialization(
+                old.keys[:, :, :1].to(dtype), old.values[:, :, :1].to(dtype)
+            )
             self.cache.layers[i] = layer
             self.cached.register_buffer(f"key_cache_{i}", layer.keys, persistent=False)
             self.cached.register_buffer(
@@ -176,6 +180,9 @@ def _store_kv_as(self, model: PreTrainedModel, dtype: torch.dtype) -> None:
     def forward(
         self, input_ids: torch.Tensor, cache_position: torch.Tensor
     ) -> torch.Tensor:
+        # StaticLayer writes at cumulative_length and ignores cache_position.
+        for layer in self.cache.layers:
+            layer.cumulative_length = cache_position[:1].clone()
         return self.cached.model(
             input_ids=input_ids,
             cache_position=cache_position,

examples/models/llama/install_requirements.sh

@@ -20,7 +20,7 @@ pip install \
   sentencepiece \
   tokenizers \
   snakeviz \
-  lm_eval==0.4.5 \
+  "lm_eval[hf]==0.4.13" \
   blobfile \
   safetensors
 

examples/models/mobilebert/model.py

@@ -21,7 +21,7 @@ def get_eager_model(self) -> torch.nn.Module:
         logging.info("loading mobilebert model")
         # pyre-ignore
         model = MobileBertModel.from_pretrained(
-            "google/mobilebert-uncased", return_dict=False
+            "google/mobilebert-uncased", return_dict=False, use_safetensors=True
         )
         logging.info("loaded mobilebert model")
         return model

examples/qualcomm/oss_scripts/albert.py

@@ -49,7 +49,7 @@ def main(args):
         random_ids = torch.randint(low=0, high=100, size=(1, 100), dtype=torch.int32)
         attention_mask = create_bidirectional_mask(
             config=module.config,
-            input_embeds=module.albert.embeddings(random_ids),
+            inputs_embeds=module.albert.embeddings(random_ids),
             attention_mask=torch.zeros((1, 100), dtype=torch.float32),
         )
         inputs = [
@@ -70,7 +70,7 @@ def main(args):
                 input_ids,
                 create_bidirectional_mask(
                     config=module.config,
-                    input_embeds=module.albert.embeddings(input_ids),
+                    inputs_embeds=module.albert.embeddings(input_ids),
                     attention_mask=attention_mask,
                 ),
             )

examples/qualcomm/oss_scripts/bert.py

@@ -47,7 +47,7 @@ def main(args):
         random_ids = torch.randint(low=0, high=100, size=(1, 100), dtype=torch.int32)
         attention_mask = create_bidirectional_mask(
             config=module.config,
-            input_embeds=module.bert.embeddings(random_ids),
+            inputs_embeds=module.bert.embeddings(random_ids),
             attention_mask=torch.zeros((1, 100), dtype=torch.float32),
         )
         inputs = [
@@ -68,7 +68,7 @@ def main(args):
                 input_ids,
                 create_bidirectional_mask(
                     config=module.config,
-                    input_embeds=module.bert.embeddings(input_ids),
+                    inputs_embeds=module.bert.embeddings(input_ids),
                     attention_mask=attention_mask,
                 ),
             )

examples/qualcomm/oss_scripts/deit.py

@@ -34,7 +34,7 @@
 def get_instance():
     module = (
         AutoModelForImageClassification.from_pretrained(
-            "facebook/deit-base-distilled-patch16-224"
+            "facebook/deit-base-distilled-patch16-224", use_safetensors=True
         )
         .eval()
         .to("cpu")