add torchao quantization

2025-07-21 19:32:12 +08:00
parent 3ae30e7b19
commit 925dd4f75c
1 changed files with 5 additions and 1 deletions
@@ -28,7 +28,7 @@ def is_module_installed(module_name):


 def get_available_quant_ops():
-    available_ops = []
+    available_ops = [("torchao", True)]

    vllm_installed = is_module_installed("vllm")
    if vllm_installed:
@@ -394,6 +394,10 @@ class ModularConfigManager:
                mm_type = (
                    f"W-{dit_scheme}-channel-sym-A-{dit_scheme}-channel-sym-dynamic-Q8F"
                )
+            elif quant_backend == "torchao":
+                mm_type = (
+                    f"W-{dit_scheme}-channel-sym-A-{dit_scheme}-channel-sym-dynamic-Torchao"
+                )
            else:
                mm_type = "Default"