Add INC modeling position_ids generation (huggingface#456)

* add position_ids in forward

* check if jit model need position_ids

* use MODEL_TYPES_REQUIRING_POSITION_IDS

* fix has_position_ids

* fix position_ids length

* rm useless params

* check model inputs by input names

* fix format

* check input names in graph model

* fix style

* consider eager model in input_names

* add input names

* add text input names

* fix styl;e

* Update optimum/intel/generation/modeling.py

* fix format

* Update optimum/intel/generation/modeling.py

---------

Co-authored-by: Ella Charlaix <ella@huggingface.co>
Co-authored-by: Ella Charlaix <80481427+echarlaix@users.noreply.github.com>

Author: 2 people

optimum/intel/generation/modeling.py

@@ -21,26 +21,20 @@
 
 import torch
 from huggingface_hub import hf_hub_download
-from transformers import AutoConfig, AutoModelForCausalLM, PretrainedConfig, PreTrainedModel
+from transformers import AutoConfig, AutoModelForCausalLM, GenerationConfig, PretrainedConfig, PreTrainedModel
+from transformers.generation import GenerationMixin
 from transformers.modeling_outputs import CausalLMOutputWithPast
 from transformers.utils import WEIGHTS_NAME
 
 from optimum.exporters import TasksManager
-from optimum.exporters.onnx import MODEL_TYPES_REQUIRING_POSITION_IDS
 from optimum.modeling_base import OptimizedModel
 from optimum.utils import NormalizedConfigManager
 
 from ..utils.constant import _TASK_ALIASES
-from ..utils.import_utils import is_torch_version, is_transformers_version
+from ..utils.import_utils import is_torch_version
 from ..utils.modeling_utils import MULTI_QUERY_ATTN_MODELS, patch_decoder_attention_mask
 
 
-if is_transformers_version("<", "4.25.0"):
-    from transformers.generation_utils import GenerationMixin
-else:
-    from transformers.generation import GenerationMixin
-
-
 logger = logging.getLogger(__name__)
 
 
@@ -112,12 +106,14 @@ def __init__(
         self.normalized_config = NormalizedConfigManager.get_normalized_config_class(config.model_type)(config)
         self.model_dtype = kwargs.get("model_dtype", None)
 
-        if is_transformers_version("<=", "4.25.1"):
-            self.generation_config = None
+        if isinstance(model, torch.jit.ScriptModule):
+            self.input_names = {
+                inputs.debugName().split(".")[0] for inputs in model.graph.inputs() if inputs.debugName() != "self"
+            }
         else:
-            from transformers import GenerationConfig
+            self.input_names = set()
 
-            self.generation_config = GenerationConfig.from_model_config(config)
+        self.generation_config = GenerationConfig.from_model_config(config)
 
         # Avoid warnings when creating a transformers pipeline
         AutoConfig.register(self.base_model_prefix, AutoConfig)
@@ -267,6 +263,7 @@ def forward(
         position_ids: Optional[torch.FloatTensor] = None,
         **kwargs,
     ) -> CausalLMOutputWithPast:
+        # 1. Prepare model inputs
         if attention_mask is None:
             attention_mask = torch.ones_like(input_ids)
 
@@ -275,6 +272,15 @@ def forward(
             "attention_mask": attention_mask,
         }
 
+        if "position_ids" in self.input_names and position_ids is None:
+            position_ids = attention_mask.long().cumsum(-1) - 1
+            position_ids.masked_fill_(attention_mask == 0, 1)
+            if past_key_values:
+                position_ids = position_ids[:, -1].unsqueeze(-1)
+
+        if "position_ids" in self.input_names or not self.input_names:
+            inputs["position_ids"] = position_ids
+
         model_type = self.config.model_type.replace("_", "-")
 
         if self.use_cache:
@@ -308,17 +314,17 @@ def forward(
 
             inputs["past_key_values"] = past_key_values
 
-        if position_ids is not None and model_type in MODEL_TYPES_REQUIRING_POSITION_IDS:
-            inputs["position_ids"] = position_ids
-
+        # 2. Model forward
         outputs = self.model(**inputs)
 
+        # 3. Process model outputs
         if isinstance(outputs, (list, tuple)):
             logits = outputs[0]
             past_key_values = outputs[1] if self.use_cache else None
         else:
             logits = outputs["logits"]
             past_key_values = outputs["past_key_values"] if self.use_cache else None
+
         return CausalLMOutputWithPast(logits=logits, past_key_values=past_key_values)
 
 

tests/generation/test_modeling.py

@@ -160,3 +160,15 @@ def test_compare_with_and_without_past_key_values(self):
             f"With pkv latency: {with_pkv_timer.elapsed:.3f} ms, without pkv latency: {without_pkv_timer.elapsed:.3f} ms,"
             f" speedup: {without_pkv_timer.elapsed / with_pkv_timer.elapsed:.3f}",
         )
+
+    @parameterized.expand(SUPPORTED_ARCHITECTURES)
+    def test_input_names(self, model_arch):
+        model_id = MODEL_NAMES[model_arch]
+        model = TSModelForCausalLM.from_pretrained(model_id, export=True)
+        self.assertTrue(isinstance(model.input_names, set))
+        self.assertTrue("input_ids" in model.input_names)
+        self.assertTrue("attention_mask" in model.input_names)
+        if model.use_cache:
+            self.assertTrue("past_key_values" in model.input_names)
+        else:
+            self.assertTrue("past_key_values" not in model.input_names)