[ExecuTorch][Llama][ez] Remove unwrap_tensor_subclass from 4w quantization path

SS-JIA · web-flow · commit 4ca6270da0a3 · 2026-04-17T15:24:14.000Z
Differential Revision: D101252037 Pull Request resolved: #18957
diff --git a/examples/models/llama/source_transformation/quantize.py b/examples/models/llama/source_transformation/quantize.py
@@ -190,7 +190,6 @@ def filter_fn(m, fqn):
             ),
         )
         quantize_(model, q_config)
-        model = unwrap_tensor_subclass(model)
 
         return model
     else:

Original file line number	Diff line number	Diff line change
`@@ -190,7 +190,6 @@ def filter_fn(m, fqn):`
`190`	`190`	`),`
`191`	`191`	`)`
`192`	`192`	`quantize_(model, q_config)`
`193`		`- model = unwrap_tensor_subclass(model)`
`194`	`193`
`195`	`194`	`return model`
`196`	`195`	`else:`