Upload 3 files

1EQMnbNPf33xQgguCJcINnytltBVBlcYN

@@ -0,0 +1,26 @@
+{
+  "model_type": "bert",
+  "alibi_starting_size": 512,
+  "architectures": [
+    "BertForSequenceClassification"
+  ],
+  "attention_probs_dropout_prob": 0.0,
+  "classifier_dropout": null,
+  "dtype": "float32",
+  "gradient_checkpointing": false,
+  "hidden_act": "gelu",
+  "hidden_dropout_prob": 0.1,
+  "hidden_size": 768,
+  "initializer_range": 0.02,
+  "intermediate_size": 3072,
+  "layer_norm_eps": 1e-12,
+  "max_position_embeddings": 512,
+  "num_attention_heads": 12,
+  "num_hidden_layers": 12,
+  "position_embedding_type": "absolute",
+  "problem_type": "single_label_classification",
+  "transformers_version": "4.57.6",
+  "type_vocab_size": 2,
+  "use_cache": true,
+  "vocab_size": 4096
+}

1d-KKrHiU0v_5NaevSkDUWPhv7W77yt9T

@@ -0,0 +1,26 @@
+# Copyright 2022 MosaicML Examples authors
+# SPDX-License-Identifier: Apache-2.0
+
+from transformers.configuration_utils import PretrainedConfig
+
+
+class BertConfig(PretrainedConfig):
+
+    def __init__(
+        self,
+        alibi_starting_size: int = 512,
+        attention_probs_dropout_prob: float = 0.0,
+        **kwargs,
+    ):
+        """Configuration class for MosaicBert.
+
+        Args:
+            alibi_starting_size (int): Use `alibi_starting_size` to determine how large of an alibi tensor to
+                create when initializing the model. You should be able to ignore this parameter in most cases.
+                Defaults to 512.
+            attention_probs_dropout_prob (float): By default, turn off attention dropout in Mosaic BERT
+                (otherwise, Flash Attention will be off by default). Defaults to 0.0.
+        """
+        super().__init__(
+            attention_probs_dropout_prob=attention_probs_dropout_prob, **kwargs)
+        self.alibi_starting_size = alibi_starting_size