Ne wmulti run system

config/examples/system.yaml

@@ -26,3 +26,33 @@ system:
       ip: 192.168.11.13
       main: false 
       user: username
+
+
+
+
+multirun:
+  runs:
+    # Force batch size to populate the sizing model
+    - name: "bs{sizer.batch_size}"
+      matrix:
+        sizer.auto: 1
+        sizer.batch_size: [1, 2, 4, 8, 16, 32, 64, 128]
+        sizer.save: "scaling.yaml"
+
+    # Matrix run
+    - name: "c{sizer.capacity}_m{sizer.multiple}_w{cpu.n_workers}"
+      matrix:
+        cpu.auto: 1
+        cpu.n_workers: [2, 4, 8, 16, 32]
+        sizer.auto: 1
+        sizer.capacity: [4Go, 8Go, 16Go, 32Go, 64Go, All]
+        sizer.multiple: 8
+        sizer.save: "scaling.yaml"
+
+    # Auto run
+    - name: "auto"
+      matrix:
+        cpu.auto: 1
+        sizer.auto: 1
+        sizer.multiple: 8
+        sizer.save: "scaling.yaml"

milabench/_version.py

@@ -1,5 +1,5 @@
 """This file is generated, do not modify"""
 
-__tag__ = "v1.0.0_RC1-13-gde92a7e"
-__commit__ = "de92a7ea9dea1da24e8105e4566d5e6daef8464c"
-__date__ = "2024-10-03 15:48:10 +0000"
+__tag__ = "v1.0.0_RC1-17-gbaf5304"
+__commit__ = "baf53044e78d0989600359e9496e9aae682bf640"
+__date__ = "2024-10-10 16:12:31 +0000"

milabench/cli/run.py

@@ -23,6 +23,7 @@
 from ..report import make_report
 from ..sizer import MemoryUsageExtractor
 from ..summary import make_summary
+from ..system import multirun, apply_system
 
 
 # fmt: off
@@ -73,20 +74,8 @@ def _fetch_arch(mp):
 
 
 @tooled
-def cli_run(args=None):
-    """Run the benchmarks."""
-    if args is None:
-        args = arguments()
-
-    layers = validation_names(args.validations)
-
-    dash_class = {
-        "short": ShortDashFormatter,
-        "long": LongDashFormatter,
-        "no": None,
-    }.get(args.dash, None)
-
-    mp = get_multipack(run_name=args.run_name)
+def run(args, name):
+    mp = get_multipack(run_name=name)
     arch = _fetch_arch(mp)
 
     # Initialize the backend here so we can retrieve GPU stats
@@ -136,3 +125,26 @@ def cli_run(args=None):
             )
 
     return success
+
+
+@tooled
+def cli_run(args=None):
+    """Run the benchmarks."""
+    if args is None:
+        args = arguments()
+
+    layers = validation_names(args.validations)
+
+    dash_class = {
+        "short": ShortDashFormatter,
+        "long": LongDashFormatter,
+        "no": None,
+    }.get(args.dash, None)
+
+    success = 0
+    for name, conf in multirun():
+        with apply_system(conf):
+            # mark the run later so we can resume multirun more easily
+            success += run(args, name or args.name)
+
+    return success

milabench/config.py

@@ -100,11 +100,15 @@ def combine_args(args, kwargs):
         yield kwargs
     else:
         key, values = args.popitem()
-        for value in values:
-            kwargs[key] = value
+
+        try:
+            for value in values:
+                kwargs[key] = value
+                yield from combine_args(deepcopy(args), kwargs)
+        except:
+            kwargs[key] = values
             yield from combine_args(deepcopy(args), kwargs)
 
-
 def expand_matrix(name, bench_config):
     if "matrix" not in bench_config:
         return [(name, bench_config)]

milabench/system.py

@@ -1,4 +1,5 @@
 import contextvars
+from copy import deepcopy
 import ipaddress
 import os
 import socket
@@ -15,7 +16,7 @@
 from .merge import merge
 
 system_global = contextvars.ContextVar("system", default=None)
-
+multirun_global = contextvars.ContextVar("multirun", default=None)
 
 def get_gpu_capacity(strict=False):
     try:
@@ -79,6 +80,60 @@ def as_environment_variable(name):
     return "MILABENCH_" + "_".join(map(str.upper, frags))
 
 
+def multirun():
+    multirun = multirun_global.get()
+
+    if len(multirun) == 0:
+        yield None, dict()
+
+    runs = multirun.get("runs", dict())
+
+    from .config import combine_args
+    import time
+    from types import SimpleNamespace
+
+    def unflatten(dct):
+        result = {}
+        for k, v in dct.items():
+            l = result
+            frags = k.split(".")
+            for frag in frags[:-1]:
+                l = l.setdefault(frag, SimpleNamespace())
+            setattr(l, frags[-1], v)
+
+        return result
+
+    for run_matrix in runs:
+        arguments = run_matrix["matrix"]
+
+        for run in combine_args(arguments, dict()):
+            template_name = run_matrix["name"]
+
+            ctx = unflatten(run)
+            ctx['time'] = int(time.time())
+            run_name = template_name.format(**ctx)
+
+            yield run_name, run
+
+
+@contextmanager
+def apply_system(config: dict):
+    system = system_global.get()
+    old = deepcopy(system)
+
+    for k, v in config.items():
+        frags = k.split(".")
+
+        lookup = system.setdefault("options", {})
+        for f in frags[:-1]:
+            lookup = lookup.setdefault(f, {})
+        lookup[frags[-1]] = v
+
+
+    yield    
+    system_global.set(old)
+
+
 def option(name, etype, default=None):
     options = dict()
     system = system_global.get()
@@ -464,6 +519,9 @@ def build_system_config(config_file, defaults=None, gpu=True):
         config = merge(defaults, config)
 
     system = config.get("system", {})
+    multirun = config.get("multirun", {})
+
+    multirun_global.set(multirun)
     system_global.set(system)
 
     # capacity is only required if batch resizer is enabled

scripts/article/run_rocm.sh

@@ -116,7 +116,7 @@ milabench prepare $ARGS
 
 #
 #   Run the benchmakrs
-milabench run $ARGS 
+milabench run $ARGS --system $MILABENCH_WORDIR/system.yaml
 
 #
 #   Display report

tests/test_system_matrix.py

@@ -0,0 +1,37 @@
+
+
+
+
+
+from milabench.system import multirun, build_system_config, enable_offline, option, apply_system, SizerOptions
+
+from milabench.testing import official_config
+
+
+def test_system_matrix():
+    with enable_offline(True):
+        sys = build_system_config(official_config("examples/system"))
+
+        n = 0
+        for name, conf in multirun():
+            print(name, conf)
+            n += 1
+
+        assert n == 39
+
+
+def test_apply_system_matrix():
+    with enable_offline(True):
+        sys = build_system_config(official_config("examples/system"))
+
+        for name, conf in multirun():
+            with apply_system(conf):
+
+                # Apply system worked and changed the config
+                for k, v in conf.items():
+                    assert option(k, lambda x: x) == v
+
+
+
+if __name__ == "__main__":
+    test_apply_system_matrix()