dflash: refactor draft model conversion (#25110)

* dflash: refactor draft model conversion

* apply fix for eagle3 convert

gguf-py/gguf/constants.py

@@ -156,6 +156,7 @@ class Keys:
         DENSE_FEAT_OUT_SIZE               = "{arch}.{dense}_feat_out"
         TARGET_LAYERS                     = "{arch}.target_layers"
         TARGET_HIDDEN_SIZE                = "{arch}.target_hidden_size"
+        BLOCK_SIZE                        = "{arch}.block_size"
         NORM_BEFORE_RESIDUAL              = "{arch}.norm_before_residual"
 
     class Attention:

gguf-py/gguf/gguf_writer.py

@@ -940,6 +940,18 @@ class GGUFWriter:
     def add_sliding_window(self, value: int) -> None:
         self.add_uint32(Keys.Attention.SLIDING_WINDOW.format(arch=self.arch), value)
 
+    def add_block_size(self, value: int) -> None:
+        self.add_uint32(Keys.LLM.BLOCK_SIZE.format(arch=self.arch), value)
+
+    def add_target_layers(self, value: Sequence[int]) -> None:
+        self.add_array(Keys.LLM.TARGET_LAYERS.format(arch=self.arch), value)
+
+    def add_target_hidden_size(self, value: int) -> None:
+        self.add_uint32(Keys.LLM.TARGET_HIDDEN_SIZE.format(arch=self.arch), value)
+
+    def add_norm_before_residual(self, value: bool) -> None:
+        self.add_bool(Keys.LLM.NORM_BEFORE_RESIDUAL.format(arch=self.arch), value)
+
     def add_attention_scale(self, value: float) -> None:
         self.add_float32(Keys.Attention.SCALE.format(arch=self.arch), value)
 

gguf-py/gguf/tensor_mapping.py

@@ -1283,6 +1283,11 @@ class TensorNameMap:
         MODEL_TENSOR.ENC_OUTPUT_NORM: (
             "encoder.final_layer_norm", # t5
             "layer_norm",               # neobert
+            "model.hidden_norm",        # dflash
+        ),
+
+        MODEL_TENSOR.FC: (
+            "model.fc", # dflash
         ),
 
         MODEL_TENSOR.CLS: (