Merge 3670ef56cf into 9448bf1caa

Updates the requirements on [safetensors](https://github.com/huggingface/safetensors) to permit the latest version.
- [Release notes](https://github.com/huggingface/safetensors/releases)
- [Changelog](https://github.com/huggingface/safetensors/blob/main/RELEASE.md)
- [Commits](https://github.com/huggingface/safetensors/compare/v0.6.0-rc0...v0.7.0)

---
updated-dependencies:
- dependency-name: safetensors
  dependency-version: 0.7.0
  dependency-type: direct:production
...

Signed-off-by: dependabot[bot] <support@github.com>

modules/image_models.py

@@ -2,7 +2,6 @@ import time
 
 import modules.shared as shared
 from modules.logging_colors import logger
-from modules.torch_utils import get_device
 from modules.utils import resolve_model_path
 
 
@@ -11,13 +10,14 @@ def get_quantization_config(quant_method):
     Get the appropriate quantization config based on the selected method.
 
     Args:
-        quant_method: One of 'none', 'bnb-8bit', 'bnb-4bit'
+        quant_method: One of 'none', 'bnb-8bit', 'bnb-4bit',
+                      'torchao-int8wo', 'torchao-fp4', 'torchao-float8wo'
 
     Returns:
         PipelineQuantizationConfig or None
     """
     import torch
-    from diffusers import BitsAndBytesConfig, QuantoConfig
+    from diffusers import BitsAndBytesConfig, TorchAoConfig
     from diffusers.quantizers import PipelineQuantizationConfig
 
     if quant_method == 'none' or not quant_method:
@@ -46,6 +46,30 @@ def get_quantization_config(quant_method):
             }
         )
 
+    # torchao int8 weight-only
+    elif quant_method == 'torchao-int8wo':
+        return PipelineQuantizationConfig(
+            quant_mapping={
+                "transformer": TorchAoConfig("int8wo")
+            }
+        )
+
+    # torchao fp4 (e2m1)
+    elif quant_method == 'torchao-fp4':
+        return PipelineQuantizationConfig(
+            quant_mapping={
+                "transformer": TorchAoConfig("fp4_e2m1")
+            }
+        )
+
+    # torchao float8 weight-only
+    elif quant_method == 'torchao-float8wo':
+        return PipelineQuantizationConfig(
+            quant_mapping={
+                "transformer": TorchAoConfig("float8wo")
+            }
+        )
+
     else:
         logger.warning(f"Unknown quantization method: {quant_method}. Loading without quantization.")
         return None
@@ -77,11 +101,13 @@ def load_image_model(model_name, dtype='bfloat16', attn_backend='sdpa', cpu_offl
         attn_backend: 'sdpa', 'flash_attention_2', or 'flash_attention_3'
         cpu_offload: Enable CPU offloading for low VRAM
         compile_model: Compile the model for faster inference (slow first run)
-        quant_method: Quantization method - 'none', 'bnb-8bit', 'bnb-4bit'
+        quant_method: 'none', 'bnb-8bit', 'bnb-4bit', or torchao options (int8wo, fp4, float8wo)
     """
     import torch
     from diffusers import DiffusionPipeline
 
+    from modules.torch_utils import get_device
+
     logger.info(f"Loading image model \"{model_name}\" with quantization: {quant_method}")
     t0 = time.time()
 

modules/shared.py

@@ -60,7 +60,7 @@ group.add_argument('--image-attn-backend', type=str, default=None, choices=['sdp
 group.add_argument('--image-cpu-offload', action='store_true', help='Enable CPU offloading for image model.')
 group.add_argument('--image-compile', action='store_true', help='Compile the image model for faster inference.')
 group.add_argument('--image-quant', type=str, default=None,
-                   choices=['none', 'bnb-8bit', 'bnb-4bit'],
+                   choices=['none', 'bnb-8bit', 'bnb-4bit', 'torchao-int8wo', 'torchao-fp4', 'torchao-float8wo'],
                    help='Quantization method for image model.')
 
 # Model loader

modules/ui.py

@@ -280,25 +280,26 @@ def list_interface_input_elements():
         'include_past_attachments',
     ]
 
-    # Image generation elements
-    elements += [
-        'image_prompt',
-        'image_neg_prompt',
-        'image_width',
-        'image_height',
-        'image_aspect_ratio',
-        'image_steps',
-        'image_cfg_scale',
-        'image_seed',
-        'image_batch_size',
-        'image_batch_count',
-        'image_model_menu',
-        'image_dtype',
-        'image_attn_backend',
-        'image_compile',
-        'image_cpu_offload',
-        'image_quant',
-    ]
+    if not shared.args.portable:
+        # Image generation elements
+        elements += [
+            'image_prompt',
+            'image_neg_prompt',
+            'image_width',
+            'image_height',
+            'image_aspect_ratio',
+            'image_steps',
+            'image_cfg_scale',
+            'image_seed',
+            'image_batch_size',
+            'image_batch_count',
+            'image_model_menu',
+            'image_dtype',
+            'image_attn_backend',
+            'image_compile',
+            'image_cpu_offload',
+            'image_quant',
+        ]
 
     return elements
 
@@ -531,25 +532,29 @@ def setup_auto_save():
         'paste_to_attachment',
         'include_past_attachments',
 
-        # Image generation tab (ui_image_generation.py)
-        'image_prompt',
-        'image_neg_prompt',
-        'image_width',
-        'image_height',
-        'image_aspect_ratio',
-        'image_steps',
-        'image_cfg_scale',
-        'image_seed',
-        'image_batch_size',
-        'image_batch_count',
-        'image_model_menu',
-        'image_dtype',
-        'image_attn_backend',
-        'image_compile',
-        'image_cpu_offload',
-        'image_quant',
     ]
 
+    if not shared.args.portable:
+        # Image generation tab (ui_image_generation.py)
+        change_elements += [
+            'image_prompt',
+            'image_neg_prompt',
+            'image_width',
+            'image_height',
+            'image_aspect_ratio',
+            'image_steps',
+            'image_cfg_scale',
+            'image_seed',
+            'image_batch_size',
+            'image_batch_count',
+            'image_model_menu',
+            'image_dtype',
+            'image_attn_backend',
+            'image_compile',
+            'image_cpu_offload',
+            'image_quant',
+        ]
+
     for element_name in change_elements:
         if element_name in shared.gradio:
             shared.gradio[element_name].change(

modules/ui_image_generation.py

@@ -18,7 +18,6 @@ from modules.image_models import (
 )
 from modules.logging_colors import logger
 from modules.text_generation import stop_everything_event
-from modules.torch_utils import get_device
 from modules.utils import gradio
 
 ASPECT_RATIOS = {
@@ -291,8 +290,11 @@ def on_gallery_select(evt: gr.SelectData, current_page):
     if evt.index is None:
         return "", "Select an image to view its settings"
 
+    if not _image_cache:
+        get_all_history_images()
+
     # Get the current page's images to find the actual file path
-    all_images = get_all_history_images()
+    all_images = _image_cache
     total_images = len(all_images)
 
     # Calculate the actual index in the full list
@@ -471,9 +473,9 @@ def create_ui():
                             with gr.Column():
                                 shared.gradio['image_quant'] = gr.Dropdown(
                                     label='Quantization',
-                                    choices=['none', 'bnb-8bit', 'bnb-4bit'],
+                                    choices=['none', 'bnb-8bit', 'bnb-4bit', 'torchao-int8wo', 'torchao-fp4', 'torchao-float8wo'],
                                     value=shared.settings['image_quant'],
-                                    info='Quantization method for reduced VRAM usage. Quanto supports lower precisions (2-bit, 4-bit, 8-bit).'
+                                    info='BnB: bitsandbytes quantization. torchao: int8wo, fp4, float8wo.'
                                 )
 
                                 shared.gradio['image_dtype'] = gr.Dropdown(
@@ -507,9 +509,7 @@ def create_ui():
                             info="Enter HuggingFace path. Use : for branch, e.g. user/model:main"
                         )
                         shared.gradio['image_download_btn'] = gr.Button("Download", variant='primary')
-                        shared.gradio['image_model_status'] = gr.Markdown(
-                            value=f"Model: **{shared.settings['image_model_menu']}** (not loaded)" if shared.settings['image_model_menu'] != 'None' else "No model selected"
-                        )
+                        shared.gradio['image_model_status'] = gr.Markdown(value="")
 
 
 def create_event_handlers():
@@ -667,7 +667,7 @@ def generate(state):
     """
     import torch
 
-    from modules.torch_utils import clear_torch_cache
+    from modules.torch_utils import clear_torch_cache, get_device
 
     clear_torch_cache()
 

requirements/full/requirements.txt

@@ -21,10 +21,11 @@ python-docx==1.1.2
 pyyaml
 requests
 rich
-safetensors==0.6.*
+safetensors==0.7.*
 scipy
 sentencepiece
 tensorboard
+torchao==0.14.*
 transformers==4.57.*
 triton-windows==3.5.1.post21; platform_system == "Windows"
 tqdm

requirements/full/requirements_amd.txt

@@ -19,10 +19,11 @@ python-docx==1.1.2
 pyyaml
 requests
 rich
-safetensors==0.6.*
+safetensors==0.7.*
 scipy
 sentencepiece
 tensorboard
+torchao==0.14.*
 transformers==4.57.*
 triton-windows==3.5.1.post21; platform_system == "Windows"
 tqdm

requirements/full/requirements_amd_noavx2.txt

@@ -19,10 +19,11 @@ python-docx==1.1.2
 pyyaml
 requests
 rich
-safetensors==0.6.*
+safetensors==0.7.*
 scipy
 sentencepiece
 tensorboard
+torchao==0.14.*
 transformers==4.57.*
 triton-windows==3.5.1.post21; platform_system == "Windows"
 tqdm

requirements/full/requirements_apple_intel.txt

@@ -19,10 +19,11 @@ python-docx==1.1.2
 pyyaml
 requests
 rich
-safetensors==0.6.*
+safetensors==0.7.*
 scipy
 sentencepiece
 tensorboard
+torchao==0.14.*
 transformers==4.57.*
 triton-windows==3.5.1.post21; platform_system == "Windows"
 tqdm

requirements/full/requirements_apple_silicon.txt

@@ -19,10 +19,11 @@ python-docx==1.1.2
 pyyaml
 requests
 rich
-safetensors==0.6.*
+safetensors==0.7.*
 scipy
 sentencepiece
 tensorboard
+torchao==0.14.*
 transformers==4.57.*
 triton-windows==3.5.1.post21; platform_system == "Windows"
 tqdm

requirements/full/requirements_cpu_only.txt

@@ -19,10 +19,11 @@ python-docx==1.1.2
 pyyaml
 requests
 rich
-safetensors==0.6.*
+safetensors==0.7.*
 scipy
 sentencepiece
 tensorboard
+torchao==0.14.*
 transformers==4.57.*
 triton-windows==3.5.1.post21; platform_system == "Windows"
 tqdm

requirements/full/requirements_cpu_only_noavx2.txt

@@ -19,10 +19,11 @@ python-docx==1.1.2
 pyyaml
 requests
 rich
-safetensors==0.6.*
+safetensors==0.7.*
 scipy
 sentencepiece
 tensorboard
+torchao==0.14.*
 transformers==4.57.*
 triton-windows==3.5.1.post21; platform_system == "Windows"
 tqdm

requirements/full/requirements_noavx2.txt

@@ -21,10 +21,11 @@ python-docx==1.1.2
 pyyaml
 requests
 rich
-safetensors==0.6.*
+safetensors==0.7.*
 scipy
 sentencepiece
 tensorboard
+torchao==0.14.*
 transformers==4.57.*
 triton-windows==3.5.1.post21; platform_system == "Windows"
 tqdm

requirements/full/requirements_nowheels.txt

@@ -19,10 +19,11 @@ python-docx==1.1.2
 pyyaml
 requests
 rich
-safetensors==0.6.*
+safetensors==0.7.*
 scipy
 sentencepiece
 tensorboard
+torchao==0.14.*
 transformers==4.57.*
 triton-windows==3.5.1.post21; platform_system == "Windows"
 tqdm