model : full modern bert support (#18330)

* full modern bert support

* added gelu op in rank pooling for modern bert

* still working on stuff, added mean calculation before classifier head

* Update convert_hf_to_gguf.py

Co-authored-by: Sigbjørn Skjæret <sigbjorn.skjaeret@scala.com>

* first layer is dense, as per modern bert research paper

* Update src/llama-graph.cpp

Co-authored-by: Sigbjørn Skjæret <sigbjorn.skjaeret@scala.com>

* fixed set input for mean pooling to check if pooling type is ranking since modern bert does mean & rank

* Update src/llama-graph.cpp

Co-authored-by: Sigbjørn Skjæret <sigbjorn.skjaeret@scala.com>

* Update convert_hf_to_gguf.py

Co-authored-by: Sigbjørn Skjæret <sigbjorn.skjaeret@scala.com>

---------

Co-authored-by: Sigbjørn Skjæret <sigbjorn.skjaeret@scala.com>

gguf-py/gguf/constants.py

@@ -652,6 +652,7 @@ class MODEL_TENSOR(IntEnum):
     ENC_OUTPUT_NORM      = auto()
     CLS                  = auto() # classifier
     CLS_OUT              = auto() # classifier output projection
+    CLS_NORM             = auto()
     CONV1D               = auto()
     CONVNEXT_DW          = auto()
     CONVNEXT_NORM        = auto()
@@ -1088,6 +1089,7 @@ TENSOR_NAMES: dict[MODEL_TENSOR, str] = {
     MODEL_TENSOR.ENC_OUTPUT_NORM:           "enc.output_norm",
     MODEL_TENSOR.CLS:                       "cls",
     MODEL_TENSOR.CLS_OUT:                   "cls.output",
+    MODEL_TENSOR.CLS_NORM:                  "cls.norm",
     MODEL_TENSOR.CONV1D:                    "conv1d",
     MODEL_TENSOR.CONVNEXT_DW:               "convnext.{bid}.dw",
     MODEL_TENSOR.CONVNEXT_NORM:             "convnext.{bid}.norm",
@@ -1507,6 +1509,7 @@ MODEL_TENSORS: dict[MODEL_ARCH, list[MODEL_TENSOR]] = {
         MODEL_TENSOR.FFN_NORM,
         MODEL_TENSOR.CLS,
         MODEL_TENSOR.CLS_OUT,
+        MODEL_TENSOR.CLS_NORM,
     ],
     MODEL_ARCH.NOMIC_BERT: [
         MODEL_TENSOR.TOKEN_EMBD,

gguf-py/gguf/tensor_mapping.py

@@ -1240,6 +1240,10 @@ class TensorNameMap:
         MODEL_TENSOR.CLS_OUT: (
             "classifier.out_proj", # roberta
         ),
+
+        MODEL_TENSOR.CLS_NORM: (
+            "head.norm", # modern-bert
+        ),
         #############################################################################
 
         MODEL_TENSOR.CONVNEXT_DW: (