【PaddleNLP No.22】Fix PIR for ernie-layout (#10484)

Author: hanlintang

slm/model_zoo/ernie-layout/README.md

@@ -408,7 +408,7 @@ python export_model.py --task_type cls --model_path ./ernie-layoutx-base-uncased
   export/
   ├── inference.pdiparams
   ├── inference.pdiparams.info
-  └── inference.pdmodel
+  └── inference.json(or inference.pdmodel if PIR is not enabled)
   ```
 
 <a name="52"></a>

slm/model_zoo/ernie-layout/README_ch.md

@@ -410,7 +410,7 @@ python export_model.py --task_type cls --model_path ./ernie-layoutx-base-uncased
 export/
 ├── inference.pdiparams
 ├── inference.pdiparams.info
-└── inference.pdmodel
+└── inference.json (在未开启PIR时为inference.pdmodel)
 ```
 
 <a name="52"></a>

slm/model_zoo/ernie-layout/deploy/python/predictor.py

@@ -25,6 +25,10 @@
 from seqeval.metrics.sequence_labeling import get_entities
 
 from paddlenlp.transformers import AutoTokenizer
+from paddlenlp.utils.env import (
+    PADDLE_INFERENCE_MODEL_SUFFIX,
+    PADDLE_INFERENCE_WEIGHTS_SUFFIX,
+)
 from paddlenlp.utils.image_utils import ppocr2example
 from paddlenlp.utils.log import logger
 
@@ -33,8 +37,8 @@ class InferBackend(object):
     def __init__(self, model_path_prefix, device="cpu"):
         logger.info(">>> [InferBackend] Creating Engine ...")
         config = paddle.inference.Config(
-            model_path_prefix + ".pdmodel",
-            model_path_prefix + ".pdiparams",
+            model_path_prefix + f"{PADDLE_INFERENCE_MODEL_SUFFIX}",
+            model_path_prefix + f"{PADDLE_INFERENCE_WEIGHTS_SUFFIX}",
         )
         if device == "gpu":
             config.enable_use_gpu(100, 0)