Use python3 (vllm image has no python alias)

run_job.sh

@@ -18,8 +18,8 @@ cd "$(dirname "$0")"
 echo "=== [setup] packages ==="
 # vllm + torch come from the base image (vllm/vllm-openai). Per TRL docs,
 # transformers>=5.2.0 may need --no-deps next to vllm's pinned stack.
-python -m pip install --quiet 'transformers>=5.2.0' --no-deps
-python -m pip install --quiet \
+python3 -m pip install --quiet 'transformers>=5.2.0' --no-deps
+python3 -m pip install --quiet \
     'trl[peft]>=1.16.0' \
     'openenv[harbor]' \
     'harbor[hf-sandbox]>=0.24.0' \
@@ -28,14 +28,14 @@ python -m pip install --quiet \
 
 echo "=== [setup] OpenEnv repo (harbor_env lives in envs/, not the wheel) ==="
 git clone --depth 1 https://github.com/huggingface/OpenEnv.git /opt/OpenEnv
-SITE=$(python -c "import site; print(site.getsitepackages()[0])")
+SITE=$(python3 -c "import site; print(site.getsitepackages()[0])")
 echo "/opt/OpenEnv/envs" > "$SITE/openenv_examples.pth"
 
 echo "=== [setup] pinned tasks (1,000 train / 250 test, hash-checked) ==="
-python prepare_tasks.py --output /app/prepared
+python3 prepare_tasks.py --output /app/prepared
 
 echo "=== [check] imports ==="
-python - <<'EOF'
+python3 - <<'EOF'
 import trl, openenv, harbor_env  # noqa: F401
 from trl.experimental.async_grpo.openenv_harness import HarnessRolloutWorker  # noqa: F401
 print("trl", trl.__version__, "| openenv", openenv.__version__)
@@ -56,7 +56,7 @@ PREPARED=/app/prepared bash serve_env.sh &
 ENV_PID=$!
 
 echo "=== [wait] servers ==="
-python - <<'EOF'
+python3 - <<'EOF'
 import time, urllib.request
 for name, url in [("vllm", "http://127.0.0.1:8001/v1/models"),
                   ("env",  "http://127.0.0.1:8000/health")]:
@@ -76,7 +76,7 @@ CUDA_VISIBLE_DEVICES=0 \
 VLLM_URL=http://127.0.0.1:8001 \
 ENV_URL=http://127.0.0.1:8000 \
 TRAIN_SPLIT=/app/prepared/datasets/train \
-    python train_lora.py
+    python3 train_lora.py
 
 kill $VLLM_PID $ENV_PID 2>/dev/null || true
 echo "=== done ==="