add experimental tensorrt quantization

Signed-off-by: Vladimir Mandic <mandic00@live.com>

CHANGELOG.md

@@ -46,6 +46,9 @@
   - **nunchaku**: update to `nunchaku==1.0.0`  
     *note*: nunchaku updated the repo which will trigger re-download of nunchaku models when first used  
     nunchaku is currently available for: *Flux.1 Dev/Schnell/Kontext/Krea/Depth/Fill*, *Qwen-Image/Qwen-Lightning*, *SANA-1.6B*  
+  - **tensorrt**: new quantization engine from nvidia  
+    *experimental*: requires new pydantic package which *may* break other things, to enable start sdnext with `--new` flag  
+    *note*: this is model quantization only, no support for tensorRT inference yet  
 - **Other**
   - refactor reuse-seed and add functionality to all tabs  
   - refactor modernui js codebase  
@@ -53,6 +56,7 @@
   - remove samplers filtering  
   - allow both flow-matching and discrete samplers for sdxl models  
   - cleanup command line parameters  
+  - add `--new` command line flag to enable testing of new packages without breaking existing installs  
 - **Fixes**
   - normalize path hanlding when deleting images  
   - unified compile upscalers  

modules/model_quant.py

@@ -111,9 +111,9 @@ def create_trt_config(kwargs = None, allow: bool = True, module: str = 'Model',
         if trt is None:
             return kwargs
         trt_config_data = {
-            "fp8": {"quant_type": "FP8", "quant_method": "modelopt", "modules_to_not_convert": []},
             "int8": {"quant_type": "INT8", "quant_method": "modelopt", "modules_to_not_convert": []},
             "int4": {"quant_type": "INT4", "quant_method": "modelopt", "block_quantize": 128, "channel_quantize": -1, "modules_to_not_convert": ["conv", "patch_embed"]},
+            "fp8": {"quant_type": "FP8", "quant_method": "modelopt", "modules_to_not_convert": []},
             "nf4": {"quant_type": "NF4", "quant_method": "modelopt", "block_quantize": 128, "channel_quantize": -1, "scale_block_quantize": 8, "scale_channel_quantize": -1, "modules_to_not_convert": ["conv"]},
             "nvfp4": {"quant_type": "NVFP4", "quant_method": "modelopt", "block_quantize": 128, "channel_quantize": -1, "modules_to_not_convert": ["conv"]},
         }

modules/shared.py

@@ -224,8 +224,8 @@ options_templates.update(options_section(("quantization", "Model Quantization"),
     "layerwise_quantization_nonblocking": OptionInfo(False, "Layerwise non-blocking operations", gr.Checkbox),
 
     "trt_quantization_sep": OptionInfo("<h2>TensorRT</h2>", "", gr.HTML),
-    "trt_quantization": OptionInfo([], "Quantization enabled", gr.CheckboxGroup, {"choices": ["Model", "TE", "LLM", "VAE"]}),
-    "trt_quantization_type": OptionInfo("nvfp4", "Quantization type", gr.Dropdown, {"choices": ["int8", "int4", "fp8", "nf4", "nvfp4"]}),
+    "trt_quantization": OptionInfo([], "Quantization enabled", gr.CheckboxGroup, {"choices": ["Model"]}),
+    "trt_quantization_type": OptionInfo("int8", "Quantization type", gr.Dropdown, {"choices": ["int8", "int4", "fp8", "nf4", "nvfp4"]}),
 
     "nncf_compress_sep": OptionInfo("<h2>NNCF: Neural Network Compression Framework</h2>", "", gr.HTML, {"visible": cmd_opts.use_openvino}),
     "nncf_compress_weights": OptionInfo([], "Quantization enabled", gr.CheckboxGroup, {"choices": ["Model", "TE", "VAE"], "visible": cmd_opts.use_openvino}),