fix(conversion): Simplify tensor name mapping in conversion

Branch: GraniteMoE Co-Authored-By: [email protected] Signed-off-by: Gabe Goodhart <[email protected]>
ggerganov · Sep 23, 2024 · 1349625 · 1349625
1 parent e68431b
commit 1349625
Showing 1 changed file with 2 additions and 2 deletions.
diff --git a/convert_hf_to_gguf.py b/convert_hf_to_gguf.py
@@ -4129,8 +4129,8 @@ def modify_tensors(self, data_torch: Tensor, name: str, bid: int | None) -> Iter
         if name.endswith("block_sparse_moe.input_linear.weight"):
             gate, up = data_torch.chunk(2, dim=-2)
             return [
-                (self.map_tensor_name(f"model.layers.{bid}.block_sparse_moe.input_linear.gate.weight"), gate),
-                (self.map_tensor_name(f"model.layers.{bid}.block_sparse_moe.input_linear.up.weight"), up),
+                (self.format_tensor_name(gguf.MODEL_TENSOR.FFN_GATE_EXP, bid), gate),
+                (self.format_tensor_name(gguf.MODEL_TENSOR.FFN_UP_EXP, bid), up),
             ]
 
         return super().modify_tensors(data_torch, name, bid)