Downloads · 30 days
0
model_repo_whisper_large_v2
yuekai/model_repo_whisper_large_v2
model_repo_whisper_large_v2 is a machine learning model from yuekai. Use it for the machine learning task on the model card, and read the license before you ship it in a product.
https://huggingface.co/spaces/yuekai/triton-asr-client
Access
Public
Updated Aug 31, 2023
Repo size
3.2 GB
Likes
0
Public
Repo makeup
Click a slice to open those files.
.onnx3.2 GB · 100%
At a glance
- Access
- Public
- Created
- Aug 17, 2023
- Updated
- Aug 31, 2023
- SHA
- dfd1b0ba
Spaces
Model card
From the Hugging Face model README
Client
https://huggingface.co/spaces/yuekai/triton-asr-client
https://github.com/yuekaizhang/Triton-ASR-Client
Server
docker pull soar97/triton-whisper:23.06
docker run -it --name "whisper-server" --gpus all --net host -v $your_mount_dir --shm-size=2g soar97/triton-whisper:23.06
apt-get install git-lfs
git-lfs install
git clone https://huggingface.co/yuekai/model_repo_whisper_large_v2.git
export CUDA_VISIBLE_DEVICES="1"
model_repo_path=./model_repo_whisper
tritonserver --model-repository $model_repo_path \
--pinned-memory-pool-byte-size=2048000000 \
--cuda-memory-pool-byte-size=0:4096000000 \
--http-port 10086 \
--metrics-port 10087
Benchmark Results
Decoding on a single V100 GPU, audios are padding to 30s, using aishell1 test set files
| Model | Backend | Concurrency | RTF |
|---|---|---|---|
| Large-v2 | ONNX FP16 | 4 | 0.14 |
| Module | Time Distribution |
|---|---|
| feature_extractor | 0.8% |
| encoder | 9.6% |
| decoder | 67.4% |
| greedy search | 22.2% |
- Created
- Aug 17, 2023
- Updated
- Aug 31, 2023