Merge pull request #101 from anarkiwi/ghae2e

enable end to end test in gha

.github/workflows/test.yml

@@ -20,10 +20,15 @@ jobs:
       - name: Install poetry and dependencies
         # TODO: remove workaround for https://github.com/pytorch/pytorch/issues/100974 once torch upgraded.
         run: |
+          sudo apt-get -yq update && sudo apt-get install -yq libnvidia-compute-535-server nvidia-cuda-dev
           git submodule init && \
           git submodule update && \
           poetry config virtualenvs.create false && \
           poetry install && \
-          for i in torch ; do poetry run pip install --no-cache-dir "$i"=="$(grep $i pyproject.toml | grep -Eo '\"[0-9\.]+' | sed 's/\"//g')" || exit 1 ; done
-          python3 -c "import torch" && \
+          for i in torch torchvision ; do poetry run pip install --no-cache-dir "$i"=="$(grep $i\\s pyproject.toml | grep -Eo '\"[0-9\.]+' | sed 's/\"//g')" || exit 1 ; done
+          python -c 'import torch' && \
+          python -c 'import rfml.annotation_utils as annotation_utils' && \
           poetry run black . --check --force-exclude torchsig
+      - name: Run e2e test
+        run: |
+          mkdir -p testdata && DATA=$(realpath testdata) PYTHON=$(which python3) ./utils/end_to_end_siggen.sh

experiments/siggen_experiments.py

@@ -2,6 +2,9 @@
 
 torch.set_float32_matmul_precision("medium")
 from rfml.experiment import *
+import sys
+
+root = sys.argv[1]
 
 #
 # python rfml/siggen_experiments.py
@@ -16,12 +19,12 @@
     "siggen_experiment": {
         "class_list": ["am", "fm"],
         "train_dir": [
-            "/data/siggen/fm.sigmf-meta",
-            "/data/siggen/am.sigmf-meta",
+            f"{root}/fm.sigmf-meta",
+            f"{root}/am.sigmf-meta",
         ],
         "val_dir": [
-            "/data/siggen/fm.sigmf-meta",
-            "/data/siggen/am.sigmf-meta",
+            f"{root}/fm.sigmf-meta",
+            f"{root}/am.sigmf-meta",
         ],
         "iq_epochs": 10,
         "iq_train_limit": 0.5,

label_scripts/label_siggen.py

@@ -5,12 +5,15 @@
 
 import rfml.annotation_utils as annotation_utils
 import rfml.data as data_class
+import sys
+
+root = sys.argv[1]
 
 # generated with
 # $ for i in am fm ; do /scratch/iqtlabs/rfml/utils/siggen.py --samp_rate 1000000 --siggen $i --int_count 1000 ; done
 data_globs = {
-    "am": ["/data/siggen/am.sigmf-meta"],
-    "fm": ["/data/siggen/fm.sigmf-meta"],
+    "am": [f"{root}/am.sigmf-meta"],
+    "fm": [f"{root}/fm.sigmf-meta"],
 }
 
 

pyproject.toml

@@ -19,9 +19,10 @@ python-on-whales = "^0.73.0"
 sigmf = "^1.2.0"
 tqdm = "^4.66.4"
 torch = "2.0.1"
+torchvision = "0.15.2"
 torch-model-archiver = "^0.12.0"
 torchsig = {path = "torchsig"}
-cupy-cuda12x = "^13.2.0"
+cupy-cuda11x = "13.3.0"
 ultralytics = "^8.2.79"
 
 

rfml/train_iq.py

@@ -321,7 +321,7 @@ def train_iq(
             ),
             checkpoint_callback,
         ],
-        accelerator="gpu",
+        accelerator="auto",
         devices=1,
         logger=logger,
         # profiler="simple",

utils/end_to_end_siggen.sh

@@ -9,30 +9,40 @@ if [ "$0" != "./utils/end_to_end_siggen.sh" ] ; then
     exit 1
 fi
 
-if [ ! -w /data ] ; then
-    echo /data must exist and be writable
+DATA="${DATA:=/data}"
+
+if [ ! -w $DATA ] ; then
+    echo $DATA must exist and be writable
     exit 1
 fi
 
-rm -rf /data/siggen
-mkdir -p /data/siggen
-mkdir -p /data/siggen/test
-mkdir -p /data/model_store
+PYTHON="${PYTHON:=python3}"
+NUMBERS="${NUMBERS:=4}"
+SRATE="${SRATE:=500000}"
+
+echo using python $PYTHON
 
-git clone https://github.com/iqtlabs/gamutrf-deploy /data/gamutrf-deploy
+rm -rf $DATA/siggen
+mkdir -p $DATA/siggen
+mkdir -p $DATA/siggen/test
+mkdir -p $DATA/model_store
+
+git clone https://github.com/iqtlabs/gamutrf-deploy $DATA/gamutrf-deploy
 
 # generate both training and test data
 docker build -f utils/Dockerfile.siggen utils -t iqtlabs/rfml-siggen
 for i in am fm ; do
     for o in /data/siggen/$i /data/siggen/test/$i ; do
-        docker run -v /data/siggen:/data/siggen -u $(id -u ${USER}):$(id -g ${USER}) -t iqtlabs/rfml-siggen /siggen.py --samp_rate 1000000 --siggen $i --int_count 100 --sample_file $o
+        docker run -v $DATA/siggen:/data/siggen -t iqtlabs/rfml-siggen /siggen.py --samp_rate $SRATE --siggen $i --int_count $NUMBERS --sample_file $o
     done
 done
 
+sudo chown -R $(id -u):$(id -g) $DATA/siggen
+
 # label data and produce model
-python label_scripts/label_siggen.py
-python experiments/siggen_experiments.py
-cp models/siggen_experiment.mar /data/model_store/torchsig_model.mar
+$PYTHON label_scripts/label_siggen.py $DATA/siggen
+$PYTHON experiments/siggen_experiments.py $DATA/siggen
+cp models/siggen_experiment.mar $DATA/model_store/torchsig_model.mar
 
 # run inference
 ./utils/gamutrf_offline_siggen.sh

utils/gamutrf_offline_siggen.sh

@@ -2,15 +2,21 @@
 
 set -e
 
-docker compose -f /data/gamutrf-deploy/offline.yml down --remove-orphans
-docker compose -f /data/gamutrf-deploy/torchserve-cuda.yml down --remove-orphans
-VOL_PREFIX=/data docker compose -f /data/gamutrf-deploy/torchserve-cuda.yml up -d
+PYTHON="${PYTHON:=python3}"
+TORCHSERVE="${TORCHSERVE:=torchserve.yml}"
+OFFLINE="${OFFLINE:=offline.yml}"
+DATA="${DATA:=/data}"
+
+
+docker compose -f $DATA/gamutrf-deploy/$OFFLINE down --remove-orphans
+docker compose -f $DATA/gamutrf-deploy/$TORCHSERVE down --remove-orphans
+VOL_PREFIX=$DATA docker compose -f $DATA/gamutrf-deploy/$TORCHSERVE up -d
 
 for i in am fm ; do
-    sudo rm -rf /data/samples
-    VOL_PREFIX=/data RECORDING=/logs/siggen/test/$i.sigmf-meta docker compose -f /data/gamutrf-deploy/offline.yml up gamutrf
+    sudo rm -rf $DATA/samples
+    VOL_PREFIX=$DATA RECORDING=/logs/siggen/test/$i.sigmf-meta docker compose -f $DATA/gamutrf-deploy/$OFFLINE up gamutrf
     echo $i
-    ./utils/count_labels.py /data/samples/*/*sigmf-meta $i
+    $PYTHON ./utils/count_labels.py $DATA/samples/*/*sigmf-meta $i
 done
 
-VOL_PREFIX=/data docker compose -f /data/gamutrf-deploy/torchserve-cuda.yml down
+VOL_PREFIX=$DATA docker compose -f $DATA/gamutrf-deploy/$TORCHSERVE down