Share target model embed and head weights for nextn (#4033)

2025-03-04 05:30:04 +08:00
parent 146ac8df07
commit 9fafa62db7
7 changed files with 47 additions and 45 deletions
--- a/scripts/export_deepseek_nextn.py
+++ b/scripts/export_deepseek_nextn.py
@@ -62,6 +62,8 @@ def export_nextn_layer_parameters(input_dir, output_dir, nextn_layer_id):
                    continue

                for key in matching_keys:
+                    if "embed_tokens" in key or "shared_head.head" in key:
+                        continue
                    new_key = key.replace(prefix, "model.layers.0")
                    params[new_key] = f.get_tensor(key)