[WIP] Quark Quantizer Support

cmake/deps.txt

@@ -14,4 +14,4 @@ pybind11;https://github.com/pybind/pybind11/archive/refs/tags/v2.10.1.zip;769b6a
 googletest;https://github.com/google/googletest/archive/530d5c8c84abd2a46f38583ee817743c9b3a42b4.zip;5e3a61db2aa975cfd0f97ba92c818744e7fa7034
 microsoft_wil;https://github.com/microsoft/wil/archive/refs/tags/v1.0.230629.1.zip;e4a542a323c070376f7c2d1973d0f7ddbc1d2fa5
 directx_headers;https://github.com/microsoft/DirectX-Headers/archive/refs/tags/v1.613.1.zip;47653509a3371eabb156360f42faf582f314bf2e
-onnxruntime_extensions;https://github.com/microsoft/onnxruntime-extensions.git;77a1812f55dbfed1fc4d9d219bdc4951ef7a6db2
+onnxruntime_extensions;https://github.com/microsoft/onnxruntime-extensions.git;4e10ee046a2f035351f3fe88740bd8215a18fdb9

src/python/py/models/builder.py

@@ -304,8 +304,7 @@ def __init__(self, config, io_dtype, onnx_dtype, ep, cache_dir, extra_options):
         }
         if self.quant_type is not None:
             # Create quantized attributes from quantization config
-            self.quant_attrs["bits"] = config.quantization_config["bits"]
-            self.quant_attrs["group_size"] = config.quantization_config["group_size"]
+            self.quant_attrs["config"] = config.quantization_config
             self.quant_attrs["use_g_idx"] = config.quantization_config["desc_act"] if "desc_act" in config.quantization_config else False
 
     def make_genai_config(self, model_name_or_path, extra_kwargs, out_dir):
@@ -2101,7 +2100,15 @@ def make_model(self, input_path):
                 from onnxruntime_genai.models.quantized_model import QuantModel
             q_size = self.num_attn_heads * self.head_size
             kv_size = self.num_kv_heads * self.head_size
-            model = QuantModel.from_pretrained(self.quant_type, input_path, self.quant_attrs["bits"], self.quant_attrs["group_size"], self.quant_attrs["use_g_idx"], q_size, kv_size, self.intermediate_size, self.num_layers)
+            model = QuantModel.from_pretrained(
+                self.quant_type,
+                input_path = input_path,
+                quant_attrs = self.quant_attrs,
+                q_size = q_size,
+                kv_size = kv_size,
+                intermediate_size = self.intermediate_size,
+                num_layers = self.num_layers,
+            )
         else:
             # Load PyTorch model
             extra_kwargs = {"num_hidden_layers": self.num_layers} if "num_hidden_layers" in self.extra_options else {}