use a separate script to download data and generate features

This commit is contained in:
Fangjun Kuang 2023-12-22 16:27:39 +08:00
parent 4fa40f5dcd
commit ed286895bc
4 changed files with 173 additions and 29 deletions

View File

@ -1,7 +1,55 @@
FROM k2fsa/icefall:torch1.13.0-cuda11.6
ARG PYTHON_VERSION=3.8
WORKDIR /workspace/icefall
FROM python:${PYTHON_VERSION}
RUN cd egs/librispeech/ASR && \
./prepare.sh --stop-stage 1 && \
ls -lh download data
RUN apt-get update -y && \
apt-get install -qq -y \
git \
git-lfs \
less \
vim \
&& \
apt-get clean && \
rm -rf /var/cache/apt/archives /var/lib/apt/lists
ARG K2_VERSION="1.24.4.dev20231220+cpu.torch1.13.0"
ARG KALDIFEAT_VERSION="1.25.3.dev20231221+cpu.torch1.13.0"
ARG TORCHAUDIO_VERSION="0.13.0"
ARG TORCH_VERSION="1.13.0"
LABEL authors="Fangjun Kuang <csukuangfj@gmail.com>"
LABEL k2_version=${K2_VERSION}
LABEL kaldifeat_version=${KALDIFEAT_VERSION}
LABEL github_repo="https://github.com/k2-fsa/icefall"
# Install dependencies
RUN pip install --no-cache-dir \
torch==${TORCH_VERSION} torchaudio==${TORCHAUDIO_VERSION} -f https://download.pytorch.org/whl/cpu/torch_stable.html \
k2==${K2_VERSION} -f https://k2-fsa.github.io/k2/cpu.html \
git+https://github.com/lhotse-speech/lhotse \
kaldifeat==${KALDIFEAT_VERSION} -f https://csukuangfj.github.io/kaldifeat/cpu.html \
kaldi_native_io \
kaldialign \
kaldifst \
kaldilm \
sentencepiece>=0.1.96 \
tensorboard \
typeguard \
dill \
onnx \
onnxruntime \
onnxmltools \
six \
multi_quantization \
typeguard \
numpy \
pytest \
graphviz
# RUN git clone https://github.com/k2-fsa/icefall /workspace/icefall && \
# cd /workspace/icefall && \
# pip install --no-cache-dir -r requirements.txt
#
# ENV PYTHONPATH /workspace/icefall:$PYTHONPATH
#
# WORKDIR /workspace/icefall

57
.github/scripts/docker/run.sh vendored Executable file
View File

@ -0,0 +1,57 @@
#!/usr/bin/env bash
set -ex
cd /icefall
export PYTHONPATH=/icefall:$PYTHONPATH
cd egs/librispeech/ASR
# We don't download the LM file since it is so large that it will
# cause OOM error for CI later.
mkdir -p download/lm
pushd download/lm
wget -q http://www.openslr.org/resources/11/librispeech-vocab.txt
wget -q http://www.openslr.org/resources/11/librispeech-lexicon.txt
wget -q http://www.openslr.org/resources/11/librispeech-lm-norm.txt.gz
ls -lh
gunzip librispeech-lm-norm.txt.gz
ls -lh
popd
pushd download/
wget -q https://huggingface.co/csukuangfj/librispeech-for-ci/resolve/main/LibriSpeech.tar.bz2
tar xf LibriSpeech.tar.bz2
rm LibriSpeech.tar.bz2
cd LibriSpeech
ln -s train-clean-100 train-clean-360
ln -s train-other-500 train-other-500
popd
mkdir -p data/manifests
lhotse prepare librispeech -j 2 -p dev-clean -p dev-other -p test-clean -p test-other -p train-clean-100 download/LibriSpeech data/manifests
ls -lh data/manifests
./local/compute_fbank_librispeech.py --dataset "dev-clean dev-other test-clean test-other train-clean-100" --perturb-speed False
ls -lh data/fbank
./prepare.sh --stage 5 --stop-stage 6
./zipformer/train.py \
--world-size 1 \
--num-epochs 1 \
--start-epoch 1 \
--use-fp16 0 \
--exp-dir zipformer/exp-small \
--causal 0 \
--num-encoder-layers 1,1,1,1,1,1 \
--feedforward-dim 64,96,96,96,96,96 \
--encoder-dim 32,64,64,64,64,64 \
--encoder-unmasked-dim 32,32,32,32,32,32 \
--base-lr 0.04 \
--full-libri 0 \
--enable-musan 0 \
--max-duration 30 \
--print-diagnostics 1

49
.github/workflows/build-cpu-docker.yml vendored Normal file
View File

@ -0,0 +1,49 @@
name: build-cpu-docker
on:
push:
branches:
- ci-train-2
workflow_dispatch:
concurrency:
group: build-cpu-docker-${{ github.ref }}
cancel-in-progress: true
jobs:
build-cpu-docker:
name: ${{ matrix.python-version }}
runs-on: ${{ matrix.os }}
strategy:
fail-fast: false
matrix:
os: [ubuntu-latest]
python-version: ["3.8", "3.9", "3.10"]
steps:
# refer to https://github.com/actions/checkout
- uses: actions/checkout@v4
with:
fetch-depth: 0
- name: Free space
shell: bash
run: |
df -h
rm -rf /opt/hostedtoolcache
df -h
- name: 'Login to GitHub Container Registry'
uses: docker/login-action@v2
with:
registry: ghcr.io
username: ${{ github.actor }}
password: ${{ secrets.GITHUB_TOKEN }}
- name: Build docker Image
shell: bash
run: |
cd .github/scripts/docker
docker build -t ghcr.io/csukuangfj/icefall:cpu-py${{ matrix.python-version }} --build-arg PYTHON_VERSION=${{ matrix.python-version }} .
docker image ls
docker push ghcr.io/csukuangfj/icefall:cpu-py${{ matrix.python-version }}

View File

@ -11,47 +11,37 @@ concurrency:
jobs:
train-librispeech:
name: ${{ matrix.image }}
name: ${{ matrix.python-version }}
runs-on: ${{ matrix.os }}
strategy:
fail-fast: false
matrix:
os: [ubuntu-latest]
python-version: ["3.8", "3.9", "3.10"]
steps:
# refer to https://github.com/actions/checkout
- uses: actions/checkout@v2
- uses: actions/checkout@v4
with:
fetch-depth: 0
- name: Rename
shell: bash
run: |
cp -v .github/scripts/docker/Dockerfile ./Dockerfile
- name: Free space
shell: bash
run: |
df -h
rm -rf /opt/hostedtoolcache
df -h
echo "pwd: $PWD"
echo "github.workspace ${{ github.workspace }}"
- name: Log in to Docker Hub
uses: docker/login-action@v2
- name: Run the build process with Docker
uses: addnab/docker-run-action@v3
with:
username: ${{ secrets.DOCKER_USERNAME }}
password: ${{ secrets.DOCKER_PASSWORD }}
image: ghcr.io/csukuangfj/icefall:cpu-py${{ matrix.python-version }}
options: |
--volume ${{ github.workspace }}/:/icefall
shell: bash
run: |
ls -lh /icefall
- name: 'Login to GitHub Container Registry'
uses: docker/login-action@v2
with:
registry: ghcr.io
username: ${{ github.actor }}
password: ${{ secrets.GITHUB_TOKEN }}
- name: Build docker Image
shell: bash
run: |
cp -v .github/scripts/docker/Dockerfile ./Dockerfile
docker build -t ghcr.io/k2-fsa/icefall:librispeech .
docker image ls
/icefall/.github/scripts/docker/run.sh