Remove quanto for now (requires messy compilation)

modules/image_models.py

@@ -11,7 +11,7 @@ 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', 'quanto-8bit', 'quanto-4bit', 'quanto-2bit'
+        quant_method: One of 'none', 'bnb-8bit', 'bnb-4bit'
 
     Returns:
         PipelineQuantizationConfig or None
@@ -46,30 +46,6 @@ def get_quantization_config(quant_method):
             }
         )
 
-    # Quanto 8-bit quantization
-    elif quant_method == 'quanto-8bit':
-        return PipelineQuantizationConfig(
-            quant_mapping={
-                "transformer": QuantoConfig(weights_dtype="int8")
-            }
-        )
-
-    # Quanto 4-bit quantization
-    elif quant_method == 'quanto-4bit':
-        return PipelineQuantizationConfig(
-            quant_mapping={
-                "transformer": QuantoConfig(weights_dtype="int4")
-            }
-        )
-
-    # Quanto 2-bit quantization
-    elif quant_method == 'quanto-2bit':
-        return PipelineQuantizationConfig(
-            quant_mapping={
-                "transformer": QuantoConfig(weights_dtype="int2")
-            }
-        )
-
     else:
         logger.warning(f"Unknown quantization method: {quant_method}. Loading without quantization.")
         return None
@@ -101,7 +77,7 @@ 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', 'quanto-8bit', 'quanto-4bit', 'quanto-2bit'
+        quant_method: Quantization method - 'none', 'bnb-8bit', 'bnb-4bit'
     """
     import torch
     from diffusers import DiffusionPipeline

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', 'quanto-8bit', 'quanto-4bit', 'quanto-2bit'],
+                   choices=['none', 'bnb-8bit', 'bnb-4bit'],
                    help='Quantization method for image model.')
 
 # Model loader

modules/ui_image_generation.py

@@ -471,7 +471,7 @@ def create_ui():
                             with gr.Column():
                                 shared.gradio['image_quant'] = gr.Dropdown(
                                     label='Quantization',
-                                    choices=['none', 'bnb-8bit', 'bnb-4bit', 'quanto-8bit', 'quanto-4bit', 'quanto-2bit'],
+                                    choices=['none', 'bnb-8bit', 'bnb-4bit'],
                                     value=shared.settings['image_quant'],
                                     info='Quantization method for reduced VRAM usage. Quanto supports lower precisions (2-bit, 4-bit, 8-bit).'
                                 )

requirements/full/requirements.txt

@@ -11,7 +11,6 @@ huggingface-hub==0.36.0
 jinja2==3.1.6
 markdown
 numpy==2.2.*
-optimum-quanto==0.2.7
 pandas
 peft==0.18.*
 Pillow>=9.5.0

requirements/full/requirements_amd.txt

@@ -9,7 +9,6 @@ huggingface-hub==0.36.0
 jinja2==3.1.6
 markdown
 numpy==2.2.*
-optimum-quanto==0.2.7
 pandas
 peft==0.18.*
 Pillow>=9.5.0

requirements/full/requirements_amd_noavx2.txt

@@ -9,7 +9,6 @@ huggingface-hub==0.36.0
 jinja2==3.1.6
 markdown
 numpy==2.2.*
-optimum-quanto==0.2.7
 pandas
 peft==0.18.*
 Pillow>=9.5.0

requirements/full/requirements_apple_intel.txt

@@ -9,7 +9,6 @@ huggingface-hub==0.36.0
 jinja2==3.1.6
 markdown
 numpy==2.2.*
-optimum-quanto==0.2.7
 pandas
 peft==0.18.*
 Pillow>=9.5.0

requirements/full/requirements_apple_silicon.txt

@@ -9,7 +9,6 @@ huggingface-hub==0.36.0
 jinja2==3.1.6
 markdown
 numpy==2.2.*
-optimum-quanto==0.2.7
 pandas
 peft==0.18.*
 Pillow>=9.5.0

requirements/full/requirements_cpu_only.txt

@@ -9,7 +9,6 @@ huggingface-hub==0.36.0
 jinja2==3.1.6
 markdown
 numpy==2.2.*
-optimum-quanto==0.2.7
 pandas
 peft==0.18.*
 Pillow>=9.5.0

requirements/full/requirements_cpu_only_noavx2.txt

@@ -9,7 +9,6 @@ huggingface-hub==0.36.0
 jinja2==3.1.6
 markdown
 numpy==2.2.*
-optimum-quanto==0.2.7
 pandas
 peft==0.18.*
 Pillow>=9.5.0

requirements/full/requirements_noavx2.txt

@@ -11,7 +11,6 @@ huggingface-hub==0.36.0
 jinja2==3.1.6
 markdown
 numpy==2.2.*
-optimum-quanto==0.2.7
 pandas
 peft==0.18.*
 Pillow>=9.5.0

requirements/full/requirements_nowheels.txt

@@ -9,7 +9,6 @@ huggingface-hub==0.36.0
 jinja2==3.1.6
 markdown
 numpy==2.2.*
-optimum-quanto==0.2.7
 pandas
 peft==0.18.*
 Pillow>=9.5.0