diff --git a/milabench/testing.py b/milabench/testing.py index 9dcd455ee..06d47ad67 100644 --- a/milabench/testing.py +++ b/milabench/testing.py @@ -11,6 +11,26 @@ from milabench.utils import multilogger, validation_layers +here = Path(__file__).parent + + +def official_config(name): + p = here / ".." / "config" / f"{name}.yaml" + + if p.exists(): + return p.resolve() + + raise FileNotFoundError(f"{p} does not exist") + + +def resolved_config(name): + from .config import build_config + + p = official_config(name) + + return build_config(p) + + class ReplayPackage(BasePackage): """Disable some folder creation for replay purposes""" diff --git a/tests/conftest.py b/tests/conftest.py index 3d021c570..fb0391830 100644 --- a/tests/conftest.py +++ b/tests/conftest.py @@ -2,6 +2,8 @@ import stat from pathlib import Path + +from milabench.testing import official_config import pytest import voir.instruments.gpu as voirgpu @@ -21,20 +23,17 @@ def runs_folder(): def config(): def get_config(name): return here / "config" / f"{name}.yaml" - return get_config - -@pytest.fixture -def official_config(): - def get_config(name): - return here / ".." / "config" / f"{name}.yaml" - return get_config +@pytest.fixture(scope='module') +def module_tmp_dir(): + import tempfile + yield tempfile.mkdtemp() -@pytest.fixture -def standard_config(official_config): +@pytest.fixture(scope='module') +def standard_config(): return official_config("standard") diff --git a/tests/test_command_reg/test_command_reg_one_node.txt b/tests/test_command_reg/test_command_reg_one_node.txt index 7c4c59d27..1390e87e8 100644 --- a/tests/test_command_reg/test_command_reg_one_node.txt +++ b/tests/test_command_reg/test_command_reg_one_node.txt @@ -15,7 +15,7 @@ export MILABENCH_DIR_DATA=$BASE/data export MILABENCH_DIR_RUNS=$BASE/runs export MILABENCH_DIR_EXTRA=$BASE/extra/llm export MILABENCH_DIR_CACHE=$BASE/cache -export MILABENCH_CONFIG='{"system": {"arch": "cuda", "sshkey": null, "nodes": [{"ip": "127.0.0.1", "main": true, "name": "0", "port": 22, "user": "username", "hostname": "127.0.0.1", "aliaslist": [], "ipaddrlist": ["127.0.0.1"], "local": true}], "self": {"ip": "127.0.0.1", "main": true, "name": "0", "port": 22, "user": "username", "hostname": "127.0.0.1", "aliaslist": [], "ipaddrlist": ["127.0.0.1"], "local": true}}, "dirs": {"base": "$BASE", "venv": "$BASE/venv/torch", "data": "$BASE/data", "runs": "$BASE/runs", "extra": "$BASE/extra/llm", "cache": "$BASE/cache"}, "group": "llm", "install_group": "torch", "install_variant": "cuda", "run_name": "dev", "enabled": true, "capabilities": {"nodes": 1}, "max_duration": 800, "voir": {"options": {"stop": 30, "interval": "1s"}}, "validation": {"usage": {"gpu_load_threshold": 0.5, "gpu_mem_threshold": 0.5}}, "config_base": "$SRC/milabench/tests/../config", "config_file": "$SRC/milabench/tests/../config/standard.yaml", "definition": "$SRC/milabench/benchmarks/llama", "tags": ["inference", "llm", "nlp"], "plan": {"method": "per_gpu"}, "weight": 1.0, "name": "llama", "tag": ["llama"]}' +export MILABENCH_CONFIG='{"system": {"arch": "cuda", "sshkey": null, "nodes": [{"ip": "127.0.0.1", "main": true, "name": "0", "port": 22, "user": "username", "hostname": "127.0.0.1", "aliaslist": [], "ipaddrlist": ["127.0.0.1"], "local": true}], "self": {"ip": "127.0.0.1", "main": true, "name": "0", "port": 22, "user": "username", "hostname": "127.0.0.1", "aliaslist": [], "ipaddrlist": ["127.0.0.1"], "local": true}}, "dirs": {"base": "$BASE", "venv": "$BASE/venv/torch", "data": "$BASE/data", "runs": "$BASE/runs", "extra": "$BASE/extra/llm", "cache": "$BASE/cache"}, "group": "llm", "install_group": "torch", "install_variant": "cuda", "run_name": "dev", "enabled": true, "capabilities": {"nodes": 1}, "max_duration": 800, "voir": {"options": {"stop": 30, "interval": "1s"}}, "validation": {"usage": {"gpu_load_threshold": 0.5, "gpu_mem_threshold": 0.5}}, "config_base": "$SRC/milabench/config", "config_file": "$SRC/milabench/config/standard.yaml", "definition": "$SRC/milabench/benchmarks/llama", "tags": ["inference", "llm", "nlp"], "plan": {"method": "per_gpu"}, "weight": 1.0, "name": "llama", "tag": ["llama"]}' export OMP_NUM_THREADS=8 echo "---" diff --git a/tests/test_command_reg/test_command_reg_two_nodes.txt b/tests/test_command_reg/test_command_reg_two_nodes.txt index 6eda9ea72..3ecda3b0f 100644 --- a/tests/test_command_reg/test_command_reg_two_nodes.txt +++ b/tests/test_command_reg/test_command_reg_two_nodes.txt @@ -15,7 +15,7 @@ export MILABENCH_DIR_DATA=$BASE/data export MILABENCH_DIR_RUNS=$BASE/runs export MILABENCH_DIR_EXTRA=$BASE/extra/llm export MILABENCH_DIR_CACHE=$BASE/cache -export MILABENCH_CONFIG='{"system": {"arch": "cuda", "sshkey": null, "nodes": [{"ip": "127.0.0.1", "main": true, "name": "0", "port": 22, "user": "username", "hostname": "127.0.0.1", "aliaslist": [], "ipaddrlist": ["127.0.0.1"], "local": true}, {"ip": "192.168.0.11", "main": false, "name": "1", "port": 22, "user": "username", "hostname": "192.168.0.11", "aliaslist": [], "ipaddrlist": ["192.168.0.11"], "local": false}], "self": {"ip": "127.0.0.1", "main": true, "name": "0", "port": 22, "user": "username", "hostname": "127.0.0.1", "aliaslist": [], "ipaddrlist": ["127.0.0.1"], "local": true}}, "dirs": {"base": "$BASE", "venv": "$BASE/venv/torch", "data": "$BASE/data", "runs": "$BASE/runs", "extra": "$BASE/extra/llm", "cache": "$BASE/cache"}, "group": "llm", "install_group": "torch", "install_variant": "cuda", "run_name": "dev", "enabled": true, "capabilities": {"nodes": 1}, "max_duration": 800, "voir": {"options": {"stop": 30, "interval": "1s"}}, "validation": {"usage": {"gpu_load_threshold": 0.5, "gpu_mem_threshold": 0.5}}, "config_base": "$SRC/milabench/tests/../config", "config_file": "$SRC/milabench/tests/../config/standard.yaml", "definition": "$SRC/milabench/benchmarks/llama", "tags": ["inference", "llm", "nlp"], "plan": {"method": "per_gpu"}, "weight": 1.0, "name": "llama", "tag": ["llama"]}' +export MILABENCH_CONFIG='{"system": {"arch": "cuda", "sshkey": null, "nodes": [{"ip": "127.0.0.1", "main": true, "name": "0", "port": 22, "user": "username", "hostname": "127.0.0.1", "aliaslist": [], "ipaddrlist": ["127.0.0.1"], "local": true}, {"ip": "192.168.0.11", "main": false, "name": "1", "port": 22, "user": "username", "hostname": "192.168.0.11", "aliaslist": [], "ipaddrlist": ["192.168.0.11"], "local": false}], "self": {"ip": "127.0.0.1", "main": true, "name": "0", "port": 22, "user": "username", "hostname": "127.0.0.1", "aliaslist": [], "ipaddrlist": ["127.0.0.1"], "local": true}}, "dirs": {"base": "$BASE", "venv": "$BASE/venv/torch", "data": "$BASE/data", "runs": "$BASE/runs", "extra": "$BASE/extra/llm", "cache": "$BASE/cache"}, "group": "llm", "install_group": "torch", "install_variant": "cuda", "run_name": "dev", "enabled": true, "capabilities": {"nodes": 1}, "max_duration": 800, "voir": {"options": {"stop": 30, "interval": "1s"}}, "validation": {"usage": {"gpu_load_threshold": 0.5, "gpu_mem_threshold": 0.5}}, "config_base": "$SRC/milabench/config", "config_file": "$SRC/milabench/config/standard.yaml", "definition": "$SRC/milabench/benchmarks/llama", "tags": ["inference", "llm", "nlp"], "plan": {"method": "per_gpu"}, "weight": 1.0, "name": "llama", "tag": ["llama"]}' export OMP_NUM_THREADS=8 echo "---" diff --git a/tests/test_mock.py b/tests/test_mock.py index 69ab8e2ca..1b9cd7cd7 100644 --- a/tests/test_mock.py +++ b/tests/test_mock.py @@ -1,20 +1,11 @@ +from contextlib import contextmanager +import os +import milabench.alt_async import milabench.commands.executors +from milabench.testing import resolved_config -import traceback -from pytest import fixture - - -@fixture -def args(standard_config, tmp_path): - return [ - "--base", str(tmp_path), - "--config", str(standard_config) - ] - - -async def mock_exec(command, phase="run", timeout=False, timeout_delay=600, **kwargs): - return [0] +import pytest def run_cli(*args): @@ -27,18 +18,72 @@ def run_cli(*args): assert not exc.code -def test_milabench(monkeypatch, args): - monkeypatch.setenv("MILABENCH_GPU_ARCH", "cuda") - monkeypatch.setattr(milabench.commands, "execute_command", mock_exec) +def benchlist(enabled=True): + standard = resolved_config("standard") + + for key, value in standard.items(): + if value.get("enabled", False): + if key[0] != "_": + yield key + + +# We want to reuse this fixtures for each bench +# so we do not run some steps multiple times +@pytest.fixture(scope='module') +def args(standard_config, module_tmp_dir): + return [ + "--base", str(module_tmp_dir), + "--config", str(standard_config) + ] - run_cli("install", *args) - run_cli("prepare", *args) +def mock_voir_run(argv, info, timeout=None, constructor=None, env=None, **options): + from voir.proc import Multiplexer + mp = Multiplexer(timeout=timeout, constructor=constructor) + mp.start(["sleep", "1"], info=info, env=env, **options) + return mp + + +def count_file_like(path, name): + try: + acc = 0 + for file in os.listdir(path + "/runs"): + if file.startswith(name): + acc += 1 + return acc + except FileNotFoundError: + return 0 + + +@contextmanager +def filecount_inc(path, name): + """Check that a new file was created after running""" + old = count_file_like(path, name) + yield + new = count_file_like(path, name) + + assert new == old + 1 + + +@pytest.mark.parametrize("bench", benchlist()) +def test_milabench(monkeypatch, args, bench, module_tmp_dir): + from milabench.cli.dry import assume_gpu + + monkeypatch.setenv("MILABENCH_GPU_ARCH", "cuda") + + with filecount_inc(module_tmp_dir, "install"): + run_cli("install", *args, "--select", bench) + + with filecount_inc(module_tmp_dir, "prepare"): + run_cli("prepare", *args, "--select", bench) # # use Mock GPU-SMI # - monkeypatch.setenv("MILABENCH_GPU_ARCH", "mock") - from milabench.cli.dry import assume_gpu - with assume_gpu(8): - run_cli("run", *args, "--no-report") + with monkeypatch.context() as ctx: + ctx.setattr(milabench.alt_async, "voir_run", mock_voir_run) + ctx.setenv("MILABENCH_GPU_ARCH", "mock") + + with filecount_inc(module_tmp_dir, bench): + with assume_gpu(8): + run_cli("run", *args, "--no-report", "--select", bench, "--run-name", str(bench))