Add original_name for more Paddle samples.

.pre-commit-config.yaml

@@ -9,7 +9,8 @@ repos:
   rev: v0.14.4
   hooks:
     - id: ruff-check
-      args: [--fix, --exit-non-zero-on-fix, --no-cache, --exclude=samples]
+      args: [--fix, --exit-non-zero-on-fix, --no-cache]
+      exclude: (^|/)(samples|paddle_samples)/
 
 - repo: https://github.com/Lucas-C/pre-commit-hooks.git
   rev: v1.5.1

graph_net/test/paddle_nlp_model_getter.py

@@ -1,3 +1,6 @@
+# Reference implementation: https://github.com/PaddlePaddle/PaddleTest/tree/develop/framework/e2e/PaddleLT_new/layerNLPcase/transformers
+
+
 def get_auto_model_and_inputs(model_name, text, dtype):
     from paddlenlp.transformers import AutoConfig, AutoModelForCausalLM, AutoTokenizer
 
@@ -139,18 +142,18 @@ def get_xlnet_model_and_inputs(model_name, text, dtype):
 
     tokenizer = XLNetTokenizer.from_pretrained(model_name)
 
-    enc = tokenizer(
+    inputs = tokenizer(
         text,
         return_tensors="pd",
         padding=True,
         truncation=True,
         # max_length=512,
     )
-    if "attention_mask" not in enc:
-        input_ids = enc["input_ids"]
+    if "attention_mask" not in inputs:
+        input_ids = inputs["input_ids"]
         pad_id = tokenizer.pad_token_id
-        enc["attention_mask"] = (input_ids != pad_id).astype("int64")
-    return model, enc
+        inputs["attention_mask"] = (input_ids != pad_id).astype("int64")
+    return model, inputs
 
 
 def get_fnet_model_and_inputs(model_name, text, dtype):
@@ -163,3 +166,30 @@ def get_fnet_model_and_inputs(model_name, text, dtype):
     tokenizer = FNetTokenizer.from_pretrained(model_name)
     inputs = tokenizer(text, return_tensors="pd")
     return model, inputs
+
+
+def get_prophetnet_model_and_inputs(model_name, text, dtype):
+    import paddle
+    from paddlenlp.transformers import ProphetNetModel, ProphetNetConfig
+    from paddlenlp.transformers import ProphetNetTokenizer
+
+    config = ProphetNetConfig.from_pretrained(model_name)
+    model = ProphetNetModel(config)
+
+    tokenizer = ProphetNetTokenizer.from_pretrained(model_name)
+    inputs = tokenizer(text, return_tensors="pd")
+    inputs.pop("token_type_ids", None)
+
+    if "attention_mask" not in inputs:
+        input_ids = inputs["input_ids"]
+        pad_id = tokenizer.pad_token_id
+        inputs["attention_mask"] = (input_ids != pad_id).astype("int64")
+
+    if "decoder_input_ids" not in inputs:
+        batch_size = inputs["input_ids"].shape[0]
+        decoder_input_ids = paddle.full(
+            [batch_size, 1], tokenizer.bos_token_id, dtype="int64"
+        )
+        inputs["decoder_input_ids"] = decoder_input_ids
+
+    return model, inputs

graph_net/test/paddle_nlp_test.py

@@ -275,6 +275,12 @@ def extract_fnet_models(text_en, text_cn):
     process_model(model_name, nlp_model_getter.get_fnet_model_and_inputs, text_en)
 
 
+def extract_prophetnet_models(text_en, text_cn):
+    # prophetnet-large-uncased models
+    model_name = "prophetnet-large-uncased"
+    process_model(model_name, nlp_model_getter.get_prophetnet_model_and_inputs, text_en)
+
+
 def main():
     current_dir = os.path.dirname(os.path.abspath(__file__))
     dump_dir = os.path.join(current_dir, "dump")

paddle_samples/PaddleNLP/bert-base-cased/graph_hash.txt

@@ -1 +1 @@
-517608d4d2699e09c6171648da38a4f924556cf25abd97875599acfdda5807e4
+a354c7a9af04a38be394ae238e9a62e2595ecb8743b22059be566ce6ae3d04e3

paddle_samples/PaddleNLP/bert-base-cased/input_meta.py

@@ -1,5 +1,6 @@
 class Program_weight_tensor_data_0:
     name = "data_0"
+    original_name = "input_ids"
     shape = [1, 21]
     dtype = "int64"
     data = [
@@ -29,6 +30,7 @@ class Program_weight_tensor_data_0:
 
 class Program_weight_tensor_data_1:
     name = "data_1"
+    original_name = "token_type_ids"
     shape = [1, 21]
     dtype = "int64"
     data = [0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0]