Merge branch 'develop' into opt_test_compiler_check

Author: Xreki

graph_net/paddle/test_compiler.py

@@ -318,12 +318,12 @@ def get_cmp_diff_count(expected_out, compiled_out, atol, rtol):
 
 def test_multi_models(args):
     verified_samples = None
-    if args.verified_samples_path is not None:
-        assert os.path.isfile(args.verified_samples_path)
+    if args.verified_samples_list_path is not None:
+        assert os.path.isfile(args.verified_samples_list_path)
         graphnet_root = path_utils.get_graphnet_root()
         print(f"graphnet_root: {graphnet_root}")
         verified_samples = []
-        with open(args.verified_samples_path, "r") as f:
+        with open(args.verified_samples_list_path, "r") as f:
             for line in f.readlines():
                 verified_samples.append(os.path.join(graphnet_root, line.strip()))
 
@@ -406,7 +406,7 @@ def main(args):
         help="Log prompt for performance log filtering.",
     )
     parser.add_argument(
-        "--verified-samples-path",
+        "--verified-samples-list-path",
         type=str,
         required=False,
         default=None,

graph_net/paddle/validate.py

@@ -49,7 +49,7 @@ def _extract_forward_source(model_path, class_name):
 def check_graph_hash(args):
     model_path = args.model_path
     file_path = f"{model_path}/graph_hash.txt"
-    if args.dump_graph_hash_key:
+    if not args.no_dump_graph_hash_key:
         model_str = _extract_forward_source(model_path, class_name="GraphModule")
         assert model_str is not None, f"model_str of {args.model_path} is None."
         new_hash_text = _get_sha_hash(model_str)
@@ -128,9 +128,9 @@ def main(args):
         help="whether check model graph redundancy",
     )
     parser.add_argument(
-        "--dump-graph-hash-key",
+        "--no-dump-graph-hash-key",
         action="store_true",
-        default=True,
+        default=False,
         help="Dump graph hash key",
     )
     parser.add_argument(

graph_net/test/nlp_model_getter.py

@@ -107,3 +107,50 @@ def get_skep_model_and_inputs(model_name, text, dtype):
     tokenizer = TokenizerClass.from_pretrained(model_name)
     inputs = tokenizer(text, return_tensors="pd")
     return model, inputs
+
+
+def get_bart_model_and_inputs(model_name, text, dtype):
+    from paddlenlp.transformers import BartModel, BartTokenizer
+
+    model = BartModel.from_pretrained(model_name)
+    model.eval()
+
+    tokenizer = BartTokenizer.from_pretrained(model_name)
+
+    inputs = tokenizer(
+        text,
+        return_tensors="pd",
+        padding=True,
+        truncation=True,
+        max_length=512,
+    )
+    inputs.pop("token_type_ids", None)
+
+    return model, inputs
+
+
+def get_xlnet_model_and_inputs(model_name, text, dtype):
+    import paddle
+    from paddlenlp.transformers import XLNetModel, XLNetTokenizer, XLNetConfig
+
+    config = XLNetConfig.from_pretrained(model_name)
+    model = XLNetModel(config)
+    if dtype == "float16":
+        model = model.astype(paddle.float16)
+    model.eval()
+
+    tokenizer = XLNetTokenizer.from_pretrained(model_name)
+
+    enc = tokenizer(
+        text,
+        return_tensors="pd",
+        padding=True,
+        truncation=True,
+        # max_length=512,
+    )
+    if "attention_mask" not in enc:
+        input_ids = enc["input_ids"]
+        pad_id = tokenizer.pad_token_id
+        enc["attention_mask"] = (input_ids != pad_id).astype("int64")
+
+    return model, enc

graph_net/test_compiler_util.py

@@ -170,7 +170,7 @@ def tolerance_generator(t):
     yield 10 ** (t * 7 / 5), 10 ** (t * 7 / 5)
 
 
-def compute_tolerance_pair(begin, end):
+def calculate_tolerance_pair(begin, end):
     tolerance_pair_list = []
     for t in range(begin, end + 1):
         for rtol, atol in tolerance_generator(t):
@@ -186,7 +186,7 @@ def compute_tolerance_pair(begin, end):
 
 
 def generate_allclose_configs(cmp_all_close_func):
-    tolerance_pair_list = compute_tolerance_pair(-10, 5)
+    tolerance_pair_list = calculate_tolerance_pair(-10, 5)
 
     cmp_configs = []
     for pair in tolerance_pair_list:

paddle_samples/PaddleNLP/albert-base-v1/graph_net.json

@@ -0,0 +1,6 @@
+{
+    "framework": "paddle",
+    "model_name": "albert-base-v1",
+    "num_devices_required": 1,
+    "num_nodes_required": 1
+}

paddle_samples/PaddleNLP/albert-base-v1/input_meta.py

@@ -0,0 +1,41 @@
+class Program_weight_tensor_data_0:
+    name = "data_0"
+    shape = [1, 21]
+    dtype = "int64"
+    data = [
+        2,
+        10975,
+        15,
+        51,
+        204,
+        25,
+        1909,
+        9,
+        31,
+        589,
+        2477,
+        88,
+        370,
+        816,
+        2761,
+        17,
+        66,
+        2607,
+        18,
+        9,
+        3,
+    ]
+
+
+class Program_weight_tensor_data_1:
+    name = "data_1"
+    shape = [1, 21]
+    dtype = "int64"
+    data = [1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1]
+
+
+class Program_weight_tensor_data_2:
+    name = "data_2"
+    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]