Qwen2.5: fix convert lit/hf checkpoint scripts

Lightning-AI · Nov 24, 2024 · d58eb95 · d58eb95
1 parent f68808d
commit d58eb95
Show file tree

Hide file tree

Showing 2 changed files with 6 additions and 6 deletions.
diff --git a/litgpt/scripts/convert_hf_checkpoint.py b/litgpt/scripts/convert_hf_checkpoint.py
@@ -562,14 +562,14 @@ def convert_hf_checkpoint(
         # holder to reconstitute the split q, k, v
         qkv_weights = {}
         copy_fn = partial(copy_weights_phi, config, qkv_weights)
-    elif config.mlp_class_name in ("LLaMAMLP", "GemmaMLP", "LLaMAMoE"):
-        # holder to reconstitute the split q, k, v
-        qkv_weights = {}
-        copy_fn = partial(copy_weights_hf_llama, config, qkv_weights)
     elif model_name.lower().startswith("qwen2.5"):
         # holder to reconstitute the split q, k, v
         qkv_weights = {}
         copy_fn = partial(copy_weights_qwen_2_5, config, qkv_weights)
+    elif config.mlp_class_name in ("LLaMAMLP", "GemmaMLP", "LLaMAMoE"):
+        # holder to reconstitute the split q, k, v
+        qkv_weights = {}
+        copy_fn = partial(copy_weights_hf_llama, config, qkv_weights)
     else:
         copy_fn = copy_weights_gpt_neox
 

diff --git a/litgpt/scripts/convert_lit_checkpoint.py b/litgpt/scripts/convert_lit_checkpoint.py
@@ -384,11 +384,11 @@ def convert_lit_checkpoint(checkpoint_dir: Path, output_dir: Path) -> None:
         copy_fn = partial(copy_weights_gemma_2, config)
     elif config.name.lower().startswith("phi"):
         copy_fn = partial(copy_weights_phi, config)
+    elif config.name.lower().startswith("qwen2.5"):
+        copy_fn = partial(copy_weights_qwen_2_5, config)
     elif config.mlp_class_name in ("LLaMAMLP", "GemmaMLP", "LLaMAMoE"):
         untie_weights = "Gemma" in config.name
         copy_fn = partial(copy_weights_llama, config, untie_weights=untie_weights)
-    elif config.name.lower().startswith("qwen2.5"):
-        copy_fn = partial(copy_weights_qwen_2_5, config)
     else:
         copy_fn = copy_weights_gpt_neox