Yuekai Zhang commited on
Commit
547de86
·
1 Parent(s): 1b118c4

add essentials

Browse files
Files changed (4) hide show
  1. README.md +8 -0
  2. encoder.onnx +3 -0
  3. requirements.txt +3 -0
  4. run.sh +28 -0
README.md ADDED
@@ -0,0 +1,8 @@
 
 
 
 
 
 
 
 
 
1
+ ### Reproduce
2
+
3
+ ```sh
4
+
5
+ docker run -it --name "oppo" --gpus all --net host nvcr.io/nvidia/tensorrt:23.03-py3
6
+ pip3 install -r requirements.txt
7
+ bash run.sh
8
+ ```
encoder.onnx ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:1096f60e4c2f92de7c24892a41fe76af7a391ff64d0b89b0a3c8d2e6d12e1311
3
+ size 72431039
requirements.txt ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ tensorrt==8.6.0
2
+ onnx
3
+ onnxruntime==1.13.1
run.sh ADDED
@@ -0,0 +1,28 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ polygraphy run \
2
+ ./encoder.onnx \
3
+ --fp16 \
4
+ --onnxrt \
5
+ --trt \
6
+ --atol 1e-3 --rtol 1e-3 \
7
+ --pool-limit workspace:1000000000 \
8
+ --save-engine=./encoder1_fp16.plan \
9
+ --verbose \
10
+ --onnx-outputs mark all \
11
+ --trt-outputs mark all \
12
+ --trt-min-shapes \
13
+ chunk_xs:[1,67,80] chunk_lens:[1] \
14
+ offset:[1,1] att_cache:[1,12,4,80,128] \
15
+ cnn_cache:[1,12,256,7] cache_mask:[1,1,80] \
16
+ --trt-opt-shapes \
17
+ chunk_xs:[16,67,80] chunk_lens:[16] \
18
+ offset:[16,1] att_cache:[16,12,4,80,128] \
19
+ cnn_cache:[16,12,256,7] cache_mask:[16,1,80] \
20
+ --trt-max-shapes \
21
+ chunk_xs:[32,67,80] chunk_lens:[32] \
22
+ offset:[32,1] att_cache:[32,12,4,80,128] \
23
+ cnn_cache:[32,12,256,7] cache_mask:[32,1,80] \
24
+ --input-shapes \
25
+ chunk_xs:[16,67,80] chunk_lens:[16] \
26
+ offset:[16,1] att_cache:[16,12,4,80,128] \
27
+ cnn_cache:[16,12,256,7] cache_mask:[16,1,80] \
28
+ --validate