[misc] fix import error (#9299)

src/llamafactory/model/model_utils/embedding.py

@@ -94,7 +94,7 @@ def _description_based_initialization(
             if len(valid_token_ids) == 0:
                 # Fallback: use mean of all existing embeddings
                 logger.warning_rank0(
-                    f"Description for token {i+1}/{num_new_tokens} contains no valid tokens. "
+                    f"Description for token {i + 1}/{num_new_tokens} contains no valid tokens. "
                     "Using mean of existing embeddings."
                 )
                 base_embedding = embed_weight[:-num_new_tokens].mean(dim=0)

src/llamafactory/model/model_utils/kv_cache.py

@@ -28,11 +28,11 @@ if TYPE_CHECKING:
 
 def configure_kv_cache(config: "PretrainedConfig", model_args: "ModelArguments", is_trainable: bool) -> None:
     if not is_trainable:
-        setattr(config, "use_cache", model_args.use_cache)
+        setattr(config, "use_cache", model_args.use_kv_cache)
         if hasattr(config, "text_config"):
-            setattr(config.text_config, "use_cache", model_args.use_cache)
+            setattr(config.text_config, "use_cache", model_args.use_kv_cache)
 
-        if model_args.use_cache:
+        if model_args.use_kv_cache:
             logger.info_rank0("KV cache is enabled for faster generation.")
         else:
             logger.info_rank0("KV cache is disabled.")

src/llamafactory/model/model_utils/moe.py

@@ -154,7 +154,17 @@ def configure_moe(config: "PretrainedConfig", model_args: "ModelArguments", is_t
     ]:
         setattr(text_config, "output_router_logits", True)
 
-    if model_type in ["ernie4_5_moe", "granitemoe", "jamba", "llama4", "mixtral", "olmoe", "phimoe", "qwen2_moe", "qwen3_moe"]:
+    if model_type in [
+        "ernie4_5_moe",
+        "granitemoe",
+        "jamba",
+        "llama4",
+        "mixtral",
+        "olmoe",
+        "phimoe",
+        "qwen2_moe",
+        "qwen3_moe",
+    ]:
         setattr(config, "router_aux_loss_coef", model_args.moe_aux_loss_coef)
 
     elif text_config and getattr(text_config, "model_type", None) in ["qwen3_moe"]: