mirror of
https://github.com/immich-app/ml-models.git
synced 2026-09-30 21:27:43 +08:00
60e028dee6
* feat: add ppocr to exporter * ocr: unpack scrfd heads * ocr: affine fold Co-authored-by: todorangrg <todorangrg@gmail.com> * rknn dynamic inputs * fp16 outputs * tiled transpose * refactor * target multiple aspect ratios * update cli * add 16:9 * better ci fixture integration * ci bug fixes * check cancellation * chore: bless ci fixtures * tweak runners * add metadata to rknn binaries * canvas->dim * fix fixture trigger * update onnxscript locals * chore: bless ci fixtures * download unconditionally * simplify fix-fixtures * use huge runner for server model * run compilation in subprocess * try/catch recv * remove 16/9 * fix detection mac utilization * chore: bless ci fixtures * conditional rknn compile * add rknn config to render --------- Co-authored-by: todorangrg <todorangrg@gmail.com> Co-authored-by: github-actions <41898282+github-actions[bot]@users.noreply.github.com>
128 lines
4.2 KiB
Plaintext
128 lines
4.2 KiB
Plaintext
CPUExecutionProvider
|
|
-24 Add
|
|
-24 LayerNormalization axis=-1 epsilon=9.999999747378752e-06 stash_type=1
|
|
+24 SkipLayerNormalization epsilon=9.999999747378752e-06
|
|
194 nodes, weights 59b60965f71a
|
|
FuseSkipLayerNorm x23 43fd47b1
|
|
FuseSkipLayerNorm, _EotSelectBeforeLayerNorm, _EotOneHotSelect x1 cec5f0c2
|
|
_EotOneHotSelect x3 ad46cb4f
|
|
_FlipCausalAttention x12 bf5bc373
|
|
_Fp16TokenEmbedding x2 777a3503
|
|
unstamped x153 d815eaaa
|
|
|
|
CUDAExecutionProvider
|
|
-24 Add
|
|
-24 LayerNormalization axis=-1 epsilon=9.999999747378752e-06 stash_type=1
|
|
+24 SkipLayerNormalization epsilon=9.999999747378752e-06
|
|
194 nodes, weights 59b60965f71a
|
|
FuseSkipLayerNorm x23 43fd47b1
|
|
FuseSkipLayerNorm, _EotSelectBeforeLayerNorm, _EotOneHotSelect x1 cec5f0c2
|
|
_EotOneHotSelect x3 ad46cb4f
|
|
_FlipCausalAttention x12 bf5bc373
|
|
_Fp16TokenEmbedding x2 777a3503
|
|
unstamped x153 d815eaaa
|
|
|
|
CoreMLExecutionProvider
|
|
+48 Reshape
|
|
+48 Transpose perm=(0, 2, 1, 3)
|
|
+24 MatMul
|
|
+13 Mul
|
|
+12 Add
|
|
-12 Attention is_causal=1 kv_num_heads=8 q_num_heads=8 qk_matmul_output_mode=0 softcap=0.0
|
|
+12 Softmax axis=-1
|
|
+12 Transpose perm=(0, 1, 3, 2)
|
|
-1 ReduceL2 keepdims=1 noop_with_empty_axes=0
|
|
+1 ReduceSum keepdims=1
|
|
+1 Sqrt
|
|
376 nodes, weights 7d064fbb71c8
|
|
DecomposeAttention, _FlipCausalAttention x168 e3c7aa4c
|
|
DecomposeReduceL2 x3 919a4b2e
|
|
_EotOneHotSelect x3 b5dd8d0a
|
|
_EotSelectBeforeLayerNorm, _EotOneHotSelect x1 9780cd73
|
|
_Fp16TokenEmbedding x2 777a3503
|
|
unstamped x199 98ea11b5
|
|
|
|
MIGraphXExecutionProvider
|
|
+48 Reshape
|
|
+48 Transpose perm=(0, 2, 1, 3)
|
|
+24 MatMul
|
|
+12 Add
|
|
-12 Attention is_causal=1 kv_num_heads=8 q_num_heads=8 qk_matmul_output_mode=0 softcap=0.0
|
|
+12 Mul
|
|
+12 Softmax axis=-1
|
|
+12 Transpose perm=(0, 1, 3, 2)
|
|
374 nodes, weights 7d064fbb71c8
|
|
DecomposeAttention, _FlipCausalAttention x168 e3c7aa4c
|
|
_EotOneHotSelect x3 b5dd8d0a
|
|
_EotSelectBeforeLayerNorm, _EotOneHotSelect x1 9780cd73
|
|
_Fp16TokenEmbedding x2 777a3503
|
|
unstamped x200 84ebb197
|
|
|
|
NvTensorRTRTXExecutionProvider
|
|
+48 Reshape
|
|
+48 Transpose perm=(0, 2, 1, 3)
|
|
+24 MatMul
|
|
+12 Add
|
|
-12 Attention is_causal=1 kv_num_heads=8 q_num_heads=8 qk_matmul_output_mode=0 softcap=0.0
|
|
+12 Mul
|
|
+12 Softmax axis=-1
|
|
+12 Transpose perm=(0, 1, 3, 2)
|
|
374 nodes, weights 7d064fbb71c8
|
|
DecomposeAttention, _FlipCausalAttention x168 e3c7aa4c
|
|
_EotOneHotSelect x3 b5dd8d0a
|
|
_EotSelectBeforeLayerNorm, _EotOneHotSelect x1 9780cd73
|
|
_Fp16TokenEmbedding x2 777a3503
|
|
unstamped x200 84ebb197
|
|
|
|
OpenVINOExecutionProvider
|
|
+48 Reshape
|
|
+48 Transpose perm=(0, 2, 1, 3)
|
|
+24 MatMul
|
|
+12 Add
|
|
-12 Attention is_causal=1 kv_num_heads=8 q_num_heads=8 qk_matmul_output_mode=0 softcap=0.0
|
|
+12 Mul
|
|
+12 Softmax axis=-1
|
|
+12 Transpose perm=(0, 1, 3, 2)
|
|
374 nodes, weights 7d064fbb71c8
|
|
DecomposeAttention, _FlipCausalAttention x168 e3c7aa4c
|
|
_EotOneHotSelect x3 b5dd8d0a
|
|
_EotSelectBeforeLayerNorm, _EotOneHotSelect x1 9780cd73
|
|
_Fp16TokenEmbedding x2 777a3503
|
|
unstamped x200 84ebb197
|
|
|
|
RKNPU static
|
|
+60 MatMul
|
|
+48 Add
|
|
+48 Reshape
|
|
+48 Slice
|
|
+48 Transpose perm=(0, 2, 1, 3)
|
|
-12 Attention is_causal=1 kv_num_heads=8 q_num_heads=8 qk_matmul_output_mode=0 softcap=0.0
|
|
+12 Mul
|
|
+12 Softmax axis=-1
|
|
+12 Transpose perm=(0, 1, 3, 2)
|
|
494 nodes, weights 3eac0e390a47
|
|
contract {"dims":[{}]}
|
|
opset 19
|
|
DecomposeAttention, _FlipCausalAttention x168 e3c7aa4c
|
|
SplitLargeReduction x132 d3b9957d
|
|
_EotOneHotSelect x3 b5dd8d0a
|
|
_EotSelectBeforeLayerNorm, _EotOneHotSelect x1 9780cd73
|
|
_Fp16TokenEmbedding x2 777a3503
|
|
unstamped x188 694b0bba
|
|
|
|
TensorrtExecutionProvider
|
|
+48 Reshape
|
|
+48 Transpose perm=(0, 2, 1, 3)
|
|
+24 MatMul
|
|
+12 Add
|
|
-12 Attention is_causal=1 kv_num_heads=8 q_num_heads=8 qk_matmul_output_mode=0 softcap=0.0
|
|
+12 Mul
|
|
+12 Softmax axis=-1
|
|
+12 Transpose perm=(0, 1, 3, 2)
|
|
374 nodes, weights 7d064fbb71c8
|
|
DecomposeAttention, _FlipCausalAttention x168 e3c7aa4c
|
|
_EotOneHotSelect x3 b5dd8d0a
|
|
_EotSelectBeforeLayerNorm, _EotOneHotSelect x1 9780cd73
|
|
_Fp16TokenEmbedding x2 777a3503
|
|
unstamped x200 84ebb197
|