remove dynamic=True for decode (#1974)

jcaip · web-flow · commit 2d11289cc1d6 · 2025-03-27T11:27:21.000-04:00
diff --git a/torchao/_models/llama/generate.py b/torchao/_models/llama/generate.py
@@ -863,7 +863,6 @@ def ffn_or_attn_only(mod, fqn):
             decode_one_token,
             mode="reduce-overhead",
             fullgraph=True,
-            dynamic=True,
         )
 
         if compile_prefill:
