[misc] code lint (#10439)

Co-authored-by: gemini-code-assist[bot] <176961590+gemini-code-assist[bot]@users.noreply.github.com>

src/llamafactory/model/model_utils/liger_kernel.py

@@ -45,7 +45,7 @@ def apply_liger_kernel(
         from liger_kernel.transformers import apply_liger_kernel_to_gemma3 as apply_liger_kernel
     elif model_type == "gemma3_text":
         from liger_kernel.transformers import apply_liger_kernel_to_gemma3_text as apply_liger_kernel
-    elif model_type in ["glm", "glm4"]: # for glm4-9b, glm4-32B respectively
+    elif model_type in ["glm", "glm4"]:  # for glm4-9b, glm4-32B respectively
         from liger_kernel.transformers import apply_liger_kernel_to_glm4 as apply_liger_kernel
     elif model_type == "glm4v":
         from liger_kernel.transformers import apply_liger_kernel_to_glm4v as apply_liger_kernel

src/llamafactory/model/model_utils/visual.py

@@ -44,15 +44,16 @@ class CompositeModel:
     language_model_keys: list[str]
     lora_conflict_keys: list[str]
 
-
     def get_projectors(self, module: "torch.nn.Module") -> list["torch.nn.Module"]:
         mm_projectors: list[torch.nn.Module] = []
         for projector_key in self.projector_keys:
             project_module = module
             for key in projector_key.split("."):
                 project_module = getattr(project_module, key, None)
-                if project_module is None: # i,e gemma4 bigger one, there is no embed_audio
-                    logger.warning_rank0(f"Projector key {projector_key} not found in module {module.__class__.__name__}.")
+                if project_module is None:  # i,e gemma4 bigger one, there is no embed_audio
+                    logger.warning_rank0(
+                        f"Projector key {projector_key} not found in module {module.__class__.__name__}."
+                    )
                     break
 
             if project_module is not None:

src/llamafactory/model/patcher.py

@@ -119,7 +119,7 @@ def patch_qwen3_5_forward(model: "PreTrainedModel") -> None:
                 cache_params=past_key_values,
                 cache_position=cache_position,
                 attention_mask=attention_mask,
-                position_ids=position_ids, # passing position_ids to linear attention
+                position_ids=position_ids,  # passing position_ids to linear attention
             )
         elif self.layer_type == "full_attention":
             hidden_states, _ = self.self_attn(
@@ -163,11 +163,7 @@ def patch_qwen3_5_forward(model: "PreTrainedModel") -> None:
             position_ids = position_ids[0]
 
         # `prepare_fa_kwargs_from_position_ids` would crash on None; guard for safety.
-        cu_seqlens = (
-            prepare_fa_kwargs_from_position_ids(position_ids)[0][0]
-            if position_ids is not None
-            else None
-        )
+        cu_seqlens = prepare_fa_kwargs_from_position_ids(position_ids)[0][0] if position_ids is not None else None
 
         # FLA varlen kernels expect [B, T, D] layout, not [B, D, T] like the
         # standard causal-conv1d path that the upstream forward uses.
@@ -232,6 +228,7 @@ def patch_qwen3_5_forward(model: "PreTrainedModel") -> None:
 
     if model.config.architectures[0] == "Qwen3_5ForConditionalGeneration":
         from transformers.models.qwen3_5.modeling_qwen3_5 import Qwen3_5DecoderLayer, Qwen3_5GatedDeltaNet
+
         Qwen3_5DecoderLayer.forward = _patched_decoder_forward
         Qwen3_5GatedDeltaNet.forward = _patch_gdn_forward
     elif model.config.architectures[0] == "Qwen3_5MoeForConditionalGeneration":
@@ -239,6 +236,7 @@ def patch_qwen3_5_forward(model: "PreTrainedModel") -> None:
             Qwen3_5MoeDecoderLayer,
             Qwen3_5MoeGatedDeltaNet,
         )
+
         Qwen3_5MoeDecoderLayer.forward = _patched_decoder_forward
         Qwen3_5MoeGatedDeltaNet.forward = _patch_gdn_forward