This is with reference to #3448.
The error currently being faced is that- (div_2, sym_size_int_3669, mul_102, mul_114, reshape_default_3, mul_213, div_113, mul_2962, clone_54, select_1, clone_67, expand_122, slice_16, clone_68, expand_123, mul_4869, expand_184,
expand_185, mul_7245, expand_246, expand_247, mul_9621, expand_308, expand_309, mul_11997, expand_370, expand_371, mul_14373, expand_432, expand_433, mul_16749, expand_494, expand_495, expand_556, expand_557,
mul_21501, expand_618, expand_619, mul_23877, expand_680, expand_681, mul_26253, expand_742, expand_743, mul_28629, expand_804, expand_805, mul_31005, expand_866, expand_867, mul_33381, expand_928, expand_929,
mul_35757, expand_990, expand_991, mul_38133, expand_1052, expand_1053, mul_40509, expand_1114, expand_1115, mul_42885, expand_1176, expand_1177, mul_45261, expand_1238, expand_1239, mul_47637, expand_1300, exp
and_1301, mul_50013, expand_1362, expand_1363, mul_52389, expand_1424, expand_1425, mul_54765, expand_1486, expand_1487, mul_57141, expand_1548, expand_1549, mul_59517, expand_1610, expand_1611)
There are symints in the output, whose dtype are not appended here

if "val" in output.meta:
           output_meta = output.meta["val"]
           if isinstance(output_meta, (FakeTensor, torch.Tensor)):
               if truncate_doulbe and output_meta.dtype == torch.float64:
                   output_dtypes.append(dtype.float32)
               else:
                   output_dtypes.append(dtype._from(output_meta.dtype))

but it is present in the engine output. So there are missing output dtypes.

Update: the above is corrected by #3488