make linter happy

Author: lmxyy

docs/source/usage/attention.rst

@@ -12,6 +12,6 @@ and 50-series GPUs compared to FlashAttention-2, without precision loss.
 
 The key change from `Basic Usage <./basic_usage>`_ is use ``transformer.set_attention_impl("nunchaku-fp16")`` to enable FP16 attention.
 While FlashAttention-2 is the default, FP16 attention offers better performance on modern NVIDIA GPUs.
-Switch back with ``transformer.set_attention_impl("flash-attention2")``. 
+Switch back with ``transformer.set_attention_impl("flash-attention2")``.
 
 For more details, see :meth:`~nunchaku.models.transformers.transformer_flux.NunchakuFluxTransformer2dModel.set_attention_impl`.

docs/source/usage/lora.rst

@@ -97,9 +97,9 @@ The syntax is as follows:
 This command composes the specified LoRAs with their respective strength values and saves the result to the output file,
 which can subsequently be loaded using :meth:`~nunchaku.models.transformers.transformer_flux.NunchakuFluxTransformer2dModel.update_lora_params` for optimized inference performance.
 
-Following composition, users may either load the file directly 
-(via the ComfyUI LoRA loader or :meth:`~nunchaku.models.transformers.transformer_flux.NunchakuFluxTransformer2dModel.update_lora_params`) 
-or utilize :mod:`nunchaku.lora.flux.convert` to convert the composed LoRA to Nunchaku's format and export it. 
+Following composition, users may either load the file directly
+(via the ComfyUI LoRA loader or :meth:`~nunchaku.models.transformers.transformer_flux.NunchakuFluxTransformer2dModel.update_lora_params`)
+or utilize :mod:`nunchaku.lora.flux.convert` to convert the composed LoRA to Nunchaku's format and export it.
 The syntax is as follows:
 
 .. code-block:: bash

docs/source/usage/qencoder.rst

@@ -12,11 +12,11 @@ Nunchaku provides a quantized T5 encoder for FLUX.1 to reduce GPU memory usage.
 The key changes from `Basic Usage <./basic_usage>`_ are:
 
 **Loading Quantized T5 Encoder** (line 11):
-Use :class:`~nunchaku.models.text_encoders.t5_encoder.NunchakuT5EncoderModel` to load the quantized encoder. 
+Use :class:`~nunchaku.models.text_encoders.t5_encoder.NunchakuT5EncoderModel` to load the quantized encoder.
 This reduces GPU memory usage while maintaining quality. Supports local or Hugging Face remote paths.
 
 **Pipeline Integration** (line 14):
-Pass the quantized encoder to the pipeline via the ``text_encoder_2`` parameter, 
+Pass the quantized encoder to the pipeline via the ``text_encoder_2`` parameter,
 replacing the default T5 encoder.
 
 .. note::

nunchaku/models/transformers/transformer_flux.py

@@ -653,7 +653,7 @@ def set_attention_impl(self, impl: str):
         ----------
         impl : str
             Attention implementation to use. Supported values:
-            
+
             - ``"flash-attention2"`` (default): Standard FlashAttention-2.
             - ``"nunchaku-fp16"``: Uses FP16 attention accumulation, up to 1.2× faster than FlashAttention-2 on NVIDIA 30-, 40-, and 50-series GPUs.
         """
@@ -783,7 +783,7 @@ def update_lora_params(self, path_or_state_dict: str | dict[str, torch.Tensor]):
         ----------
         path_or_state_dict : str or dict
             Path to a LoRA weights file or a state dict. The path supports:
-            
+
             - Local file path, e.g., ``"/path/to/your/lora.safetensors"``
             - HuggingFace repo with file, e.g., ``"user/repo/lora.safetensors"``
               (automatically downloaded and cached)