MCore dataset compatibility for tokenizers (NVIDIA#8390) (NVIDIA#8397)

* Add unique_identifiers for all tokenizers and eod for SentencePieceTokenizer



* Add generalized token aliases to TokenizerSpec to conform with MegatronTokenizer's interface. Remove now-redundant individual fixes from AutoTokenizer and SentencePieceTokenizer.



---------

Signed-off-by: Valerie Sarge <vsarge@nvidia.com>
Co-authored-by: Valerie Sarge <vsarge@nvidia.com>
Co-authored-by: Pablo Garay <palenq@gmail.com>
Co-authored-by: Eric Harper <complex451@gmail.com>

nemo/collections/common/tokenizers/tokenizer_spec.py

@@ -13,6 +13,7 @@
 # limitations under the License.
 
 from abc import ABC, abstractmethod
+from collections import OrderedDict
 from typing import List
 
 __all__ = ['TokenizerSpec']
@@ -53,3 +54,60 @@ def add_special_tokens(self, special_tokens: List[str]):
     @property
     def name(self):
         return type(self).__name__
+
+    @property
+    def unique_identifiers(self):
+        """Property required for use with megatron-core datasets."""
+        return OrderedDict({"class": f"{type(self).__module__}.{type(self).__qualname__}"})
+
+    @property
+    def cls(self):
+        """Property alias to match MegatronTokenizer; returns cls_id if available."""
+        if hasattr(self, 'cls_id'):
+            return self.cls_id
+        raise AttributeError(f"{type(self).__name__} has no attribute 'cls' or 'cls_id'")
+
+    @property
+    def sep(self):
+        """Property alias to match MegatronTokenizer; returns sep_id if available."""
+        if hasattr(self, 'sep_id'):
+            return self.sep_id
+        raise AttributeError(f"{type(self).__name__} has no attribute 'sep' or 'sep_id'")
+
+    @property
+    def pad(self):
+        """Property alias to match MegatronTokenizer; returns pad_id if available."""
+        if hasattr(self, 'pad_id'):
+            return self.pad_id
+        raise AttributeError(f"{type(self).__name__} has no attribute 'pad' or 'pad_id'")
+
+    @property
+    def eod(self):
+        """Property alias to match MegatronTokenizer; returns eod_id if available."""
+        if hasattr(self, 'eod_id'):
+            return self.eod_id
+        if hasattr(self, 'eos_id'):
+            # Default to end-of-sentence id if end-of-document is not defined.
+            return self.eos_id
+        raise AttributeError(f"{type(self).__name__} has no attribute 'eod', 'eod_id', 'eos', or 'eos_id'")
+
+    @property
+    def bos(self):
+        """Property alias to match MegatronTokenizer; returns bos_id if available."""
+        if hasattr(self, 'bos_id'):
+            return self.bos_id
+        raise AttributeError(f"{type(self).__name__} has no attribute 'bos' or 'bos_id'")
+
+    @property
+    def eos(self):
+        """Property alias to match MegatronTokenizer; returns eos_id if available."""
+        if hasattr(self, 'eos_id'):
+            return self.eos_id
+        raise AttributeError(f"{type(self).__name__} has no attribute 'eos' or 'eos_id'")
+
+    @property
+    def mask(self):
+        """Property alias to match MegatronTokenizer; returns mask_id if available."""
+        if hasattr(self, 'mask_id'):
+            return self.mask_id
+        raise AttributeError(f"{type(self).__name__} has no attribute 'mask' or 'mask_id'")