bconsolvo's picture
Update AMD NPU model files to Ryzen AI 1.8.0
70c348b verified
Raw
History Blame
4.23 kB
DynamicDispatch Offload - not offloaded
+-------------------+-------+------------------------+------------------------+
| Op Type | Count | Inputs | Outputs |
+===================+=======+========================+========================+
| CastAvx | 1 | [1,'height','width',1� | [1,'height','width',1� |
| | | - FLOAT16 | - BFLOAT16 |
| CastAvx | 1 | [1,'floor(2.0*floor(2� | [1,'floor(2.0*floor(2� |
| | | - BFLOAT16 | - FLOAT16 |
| DynamicPad | 1 | [1,16,'height','width� | [1] - FLOAT16 |
| | | - FLOAT16 | |
| Gather | 2 | [1] - INT64 | [1] - INT64 |
| | | [1] - INT64 | |
| Mul | 2 | [1] - INT64 | [1] - INT64 |
| | | [1] - INT64 | |
| Shape | 1 | [1,16,'height','width� | [1] - INT64 |
| | | - FLOAT16 | |
| Slice | 1 | [1] - FLOAT16 | [1] - FLOAT16 |
| | | [1] - INT64 | |
| | | [1] - INT64 | |
| | | [1] - INT64 | |
| Slice | 1 | [1] - FLOAT16 | [1,3,'floor(2.0*floor� |
| | | [1] - INT64 | - FLOAT16 |
| | | [1] - INT64 | |
| | | [1] - INT64 | |
| Transpose | 1 | [1] - FLOAT16 | [1,'height','width',1� |
| | | | - FLOAT16 |
| Transpose | 1 | [1,'floor(2.0*floor(2� | [1] - FLOAT16 |
| | | - FLOAT16 | |
+-------------------+-------+------------------------+------------------------+
| Not offloaded sum | 12 | | |
+-------------------+-------+------------------------+------------------------+
DynamicDispatch Offload - offloaded
+--------------------+-------------+---------------------+--------------------+
| Op Type | Count | Inputs | Outputs |
+====================+=============+=====================+====================+
| DynamicDispatch | 1 | [1,'height','width� | [1,'floor(2.0*flo� |
| | | - BFLOAT16 | - BFLOAT16 |
+--------------------+-------------+---------------------+--------------------+
| Offloaded sum | 1 | | |
+--------------------+-------------+---------------------+--------------------+
| Offloaded Op Types | SDGroupNorm | | |
| | SDConv | | |
| | SDResize | | |
| | SDGemm | | |
| | SDAdd | | |
| | SDMHA_VAE | | |
+--------------------+-------------+---------------------+--------------------+
| Offloaded sum (dd | 88 | | |
| fusion) | | | |
| Offload Ratio (dd | 88.0% | | |
| fusion) | | | |
+--------------------+-------------+---------------------+--------------------+