chore(docker): remove docker-compose, use individual build/run commands

Replace docker-compose.benchmark.yml with per-image docker build/run
instructions. Each benchmark is built and tested independently via
--build-arg BENCHMARK=<name> on Dockerfile.benchmark.

Co-Authored-By: Claude <noreply@anthropic.com>

docker/docker-compose.benchmark.yml

@@ -1,68 +0,0 @@
-# Benchmark evaluation services for LeRobot.
-#
-# Each service builds Dockerfile.benchmark with the matching BENCHMARK arg,
-# giving an isolated environment per benchmark suite.
-#
-# Usage:
-#   Build all:        docker compose -f docker/docker-compose.benchmark.yml build
-#   Build one:        docker compose -f docker/docker-compose.benchmark.yml build libero
-#   Run smoke tests:  docker compose -f docker/docker-compose.benchmark.yml run --rm libero bash docker/smoke_test_benchmark.sh
-#   Interactive:      docker compose -f docker/docker-compose.benchmark.yml run --rm libero
-
-x-benchmark-base: &benchmark-base
-  build:
-    context: .
-    dockerfile: docker/Dockerfile.benchmark
-  image: lerobot-benchmark-${BENCHMARK:-libero}
-  environment:
-    - MUJOCO_GL=egl
-    - DEVICE=cuda
-  deploy:
-    resources:
-      reservations:
-        devices:
-          - driver: nvidia
-            count: 1
-            capabilities: [gpu]
-  volumes:
-    # Mount HuggingFace cache from host so model weights are not re-downloaded each run.
-    - ${HF_HOME:-~/.cache/huggingface}:/home/user_lerobot/.cache/huggingface
-  stdin_open: true
-  tty: true
-
-services:
-  libero:
-    <<: *benchmark-base
-    image: lerobot-benchmark-libero
-    build:
-      context: .
-      dockerfile: docker/Dockerfile.benchmark
-      args:
-        BENCHMARK: libero
-
-  libero_plus:
-    <<: *benchmark-base
-    image: lerobot-benchmark-libero-plus
-    build:
-      context: .
-      dockerfile: docker/Dockerfile.benchmark
-      args:
-        BENCHMARK: libero_plus
-
-  robomme:
-    <<: *benchmark-base
-    image: lerobot-benchmark-robomme
-    build:
-      context: .
-      dockerfile: docker/Dockerfile.benchmark
-      args:
-        BENCHMARK: robomme
-
-  robocasa:
-    <<: *benchmark-base
-    image: lerobot-benchmark-robocasa
-    build:
-      context: .
-      dockerfile: docker/Dockerfile.benchmark
-      args:
-        BENCHMARK: robocasa

docker/smoke_test_benchmark.sh

@@ -1,13 +1,14 @@
 #!/usr/bin/env bash
 # Smoke-test a benchmark container: verifies imports and CLI entry-points.
 #
-# Run inside the container (BENCHMARK env var must be set):
+# Build and run for a specific benchmark:
-#   bash docker/smoke_test_benchmark.sh
+#   docker build --build-arg BENCHMARK=libero -f docker/Dockerfile.benchmark -t lerobot-benchmark-libero .
+#   docker run --gpus all --rm -e BENCHMARK=libero lerobot-benchmark-libero bash docker/smoke_test_benchmark.sh
 #
-# Or run all benchmarks via docker compose:
+# Test all benchmarks individually:
-#   for svc in libero libero_plus robomme robocasa; do
+#   for b in libero libero_plus robomme robocasa; do
-#     docker compose -f docker/docker-compose.benchmark.yml run --rm "$svc" \
+#     docker build --build-arg BENCHMARK=$b -f docker/Dockerfile.benchmark -t lerobot-benchmark-$b .
-#       bash docker/smoke_test_benchmark.sh
+#     docker run --gpus all --rm -e BENCHMARK=$b lerobot-benchmark-$b bash docker/smoke_test_benchmark.sh
 #   done
 
 set -euo pipefail