Sharding fails in TF when absolute scope was modified if . in layer name (#19124)

* simplify loop

* fix layer map split

* update

* update for special variables

* add rag test

* fixup

* revert change : for next PR

src/transformers/modeling_tf_utils.py

@@ -707,8 +707,15 @@ def load_tf_sharded_weights(model, shard_files, ignore_mismatched_sizes=False, s
 
     # Since TF adds the name of the class to its weights, and uses the index and not the name of the layer to load
     # the weight, we have to get rid of the first prefix of the name of the layer.
-    model_keys = set("/".join(k.name.split("/")[1:]) for k in model.weights)
+    model_keys = set()
-    model_layer_map = {"/".join(k.name.split("/")[1:]): i for i, k in enumerate(model.weights)}
+    model_layer_map = dict()
+    for i, k in enumerate(model.weights):
+        if "model." in k.name or len(k.name.split("/")) == 1:
+            layer_name = k.name
+        else:
+            layer_name = "/".join(k.name.split("/")[1:])
+        model_keys.add(layer_name)
+        model_layer_map[layer_name] = i
 
     for shard_file in shard_files:
         state_dict = tf.io.read_file(shard_file)
@@ -2211,17 +2218,19 @@ class TFPreTrainedModel(tf.keras.Model, TFModelUtilsMixin, TFGenerationMixin, Pu
             )
             for shard_file, shard in shards.items():
                 with h5py.File(os.path.join(save_directory, shard_file), mode="w") as shard_file:
-                    save_attributes_to_hdf5_group(
+                    layers = []
-                        shard_file,
-                        "layer_names",
-                        ["/".join(layer.name.split("/")[1:]).encode("utf8") for layer in shard],
-                    )
-
                     for layer in sorted(shard, key=lambda x: x.name):
+                        if "model." in layer.name or len(layer.name.split("/")) == 1:
+                            layer_name = layer.name
+                            print(layer_name)
+                        else:
+                            layer_name = "/".join(layer.name.split("/")[1:])
                         param_dset = shard_file.create_dataset(
-                            "/".join(layer.name.split("/")[1:]), layer.numpy().shape, dtype=layer.numpy().dtype
+                            layer_name, layer.numpy().shape, dtype=layer.numpy().dtype
                         )
                         param_dset[:] = layer.numpy()
+                        layers.append(layer_name.encode("utf8"))
+                    save_attributes_to_hdf5_group(shard_file, "layer_names", layers)
 
         if push_to_hub:
             self._upload_modified_files(

tests/test_modeling_tf_common.py

@@ -77,9 +77,11 @@ if is_tf_available():
         TF_MODEL_FOR_SPEECH_SEQ_2_SEQ_MAPPING,
         TF_MODEL_FOR_TOKEN_CLASSIFICATION_MAPPING,
         BertConfig,
+        RagRetriever,
         TFAutoModel,
         TFAutoModelForSequenceClassification,
         TFBertModel,
+        TFRagModel,
         TFSharedEmbeddings,
     )
     from transformers.generation_tf_utils import (
@@ -2167,6 +2169,18 @@ class UtilsFunctionsTest(unittest.TestCase):
                 },
             )
 
+    @slow
+    def test_special_layer_name_shardind(self):
+        retriever = RagRetriever.from_pretrained("facebook/rag-token-nq", index_name="exact", use_dummy_dataset=True)
+        model = TFRagModel.from_pretrained("facebook/rag-token-nq", retriever=retriever)
+
+        with tempfile.TemporaryDirectory() as tmp_dir:
+            for max_size in ["150kB", "150kiB", "200kB", "200kiB"]:
+                model.save_pretrained(tmp_dir, max_shard_size=max_size)
+                ref_model = TFRagModel.from_pretrained(tmp_dir, retriever=retriever)
+                for p1, p2 in zip(model.weights, ref_model.weights):
+                    assert np.allclose(p1.numpy(), p2.numpy())
+
     def test_checkpoint_sharding_local(self):
         model = TFBertModel.from_pretrained("hf-internal-testing/tiny-random-bert")