float / onnx /backends.generated.yaml
mohamedemam's picture
Add the converted ONNX artifacts (onnx/)
cf23f0c verified
Raw History Blame Contribute Delete
4.18 kB
version: 2
target:
torch: 2.7.1+cu126
cuda: true
gpu: NVIDIA RTX A6000
cc: '8.6'
vram_gb: 47.4
ort: 1.23.2
trt_ep_listed: true
trt_libs_loadable: true
trt: true
cuda_ep: true
objective: end_to_end_ms
default_mode: eager
chosen_combo:
emotion: onnx
audio: onnx
fmt: onnx
refenc: compile
e2e:
kind: global_generator
eager_ms: 18.77132902154699
eager_mad_ms: 0.012680015061050653
compile_ms: 16.538165014935657
compile_mad_ms: 0.04346101195551455
compile_s: 3.469
speedup: 1.1350309423442417
significant: true
components:
emotion:
mode: onnx
provider: trt
onnx_path: onnx/emotion_v1.onnx
metrics:
eager_ms: 16.313770989654586
eager_mad_ms: 0.2377640048507601
compile_ms: 12.844707001931965
compile_mad_ms: 0.22261301637627184
compile_s: 3.432
onnx_parity:
max_abs: 0.000633
mean_abs: 0.000164
cosine: 1.0
ok: true
onnx_ms: 2.843594003934413
onnx_mad_ms: 0.004879984771832824
onnx_provider: trt
onnx_providers:
- TensorrtExecutionProvider
- CUDAExecutionProvider
- CPUExecutionProvider
audio:
mode: onnx
provider: trt
onnx_path: onnx/wav2vec_v1.onnx
metrics:
eager_ms: 7.668967999052256
eager_mad_ms: 0.03063099575228989
compile_ms: 4.126863001147285
compile_mad_ms: 0.03980001201853156
compile_s: 6.241
onnx_parity:
max_abs: 0.015054
mean_abs: 0.000658
cosine: 0.99999
ok: true
onnx_ms: 1.743127009831369
onnx_mad_ms: 0.004930014256387949
onnx_provider: trt
onnx_providers:
- TensorrtExecutionProvider
- CUDAExecutionProvider
- CPUExecutionProvider
direction:
mode: eager
onnx_path: onnx/direction_v1.onnx
metrics:
eager_ms: 0.037949997931718826
eager_mad_ms: 0.0008099887054413557
compile_ms: 0.07996099884621799
compile_mad_ms: 0.001040025381371379
compile_s: 0.025
onnx_parity:
max_abs: 0.000194
mean_abs: 3.9e-05
cosine: 1.0
ok: true
onnx_ms: 0.181223003892228
onnx_mad_ms: 0.003029999788850546
onnx_provider: trt
onnx_providers:
- TensorrtExecutionProvider
- CUDAExecutionProvider
- CPUExecutionProvider
reason: no candidate beat eager by 1.10x outside the noise band
fmt:
mode: onnx
provider: trt
onnx_path: onnx/fmt_step_v1.onnx
metrics:
eager_ms: 5.844598985277116
eager_mad_ms: 0.013921002391725779
compile_ms: 4.951315990183502
compile_mad_ms: 0.020679988665506244
compile_s: 3.774
onnx_parity:
max_abs: 0.020788
mean_abs: 0.002348
cosine: 0.999996
ok: true
onnx_ms: 2.2519149933941662
onnx_mad_ms: 0.01677899854257703
onnx_provider: trt
onnx_providers:
- TensorrtExecutionProvider
- CUDAExecutionProvider
- CPUExecutionProvider
decoder:
mode: eager
onnx_path: onnx/decoder_v1.onnx
metrics:
eager_ms: 25.683184998342767
eager_mad_ms: 0.08247100049629807
compile_ms: 24.97261398821138
compile_mad_ms: 0.13203200069256127
compile_s: 9.272
onnx_parity:
max_abs: 0.673292
mean_abs: 0.003701
cosine: 0.999929
ok: true
onnx_ms: 149.67813200200908
onnx_mad_ms: 2.0826529944315553
onnx_provider: cuda
onnx_providers:
- CUDAExecutionProvider
- CPUExecutionProvider
reason: no candidate beat eager by 1.10x outside the noise band
refenc:
mode: compile
onnx_path: onnx/refenc_v1.onnx
metrics:
eager_ms: 5.775298981461674
eager_mad_ms: 0.004929985152557492
compile_ms: 5.231279996223748
compile_mad_ms: 0.04106998676434159
compile_s: 3.105
onnx_parity:
max_abs: 0.091488
mean_abs: 0.007204
cosine: 0.999985
ok: true
onnx_ms: 11.51109699276276
onnx_mad_ms: 0.11942200944758952
onnx_provider: trt
onnx_providers:
- TensorrtExecutionProvider
- CUDAExecutionProvider
- CPUExecutionProvider