Update voir version to 0.2.8 (#122)

mila-iqia · May 4, 2023 · 0c5c00e · 0c5c00e
1 parent 1eb4887
commit 0c5c00e
Show file tree

Hide file tree

Showing 28 changed files with 49 additions and 32 deletions.
diff --git a/benchmarks/_template/requirements.in b/benchmarks/_template/requirements.in
@@ -1 +1 @@
-voir>=0.2.5,<0.3
+voir>=0.2.8,<0.3
diff --git a/benchmarks/accelerate_opt/requirements.cuda.txt b/benchmarks/accelerate_opt/requirements.cuda.txt
@@ -315,7 +315,7 @@ varname==0.10.0
  # via
  # -c .pin-constraints-torch.txt
  # giving
-voir==0.2.6
+voir==0.2.8
  # via -r benchmarks/accelerate_opt/requirements.in
 xxhash==3.2.0
  # via

diff --git a/benchmarks/accelerate_opt/requirements.in b/benchmarks/accelerate_opt/requirements.in
@@ -8,4 +8,4 @@ accelerate
 deepspeed
 rich
 asyncssh
-voir>=0.2.4,<0.3
+voir>=0.2.8,<0.3
diff --git a/benchmarks/accelerate_opt/requirements.rocm.txt b/benchmarks/accelerate_opt/requirements.rocm.txt
@@ -315,7 +315,7 @@ varname==0.10.0
  # via
  # -c .pin-constraints-torch.txt
  # giving
-voir==0.2.5
+voir==0.2.8
  # via -r benchmarks/accelerate_opt/requirements.in
 xxhash==3.2.0
  # via

diff --git a/benchmarks/dlrm/requirements.cuda.txt b/benchmarks/dlrm/requirements.cuda.txt
@@ -336,7 +336,7 @@ varname==0.10.0
  # via
  # -c .pin-constraints-torch.txt
  # giving
-voir==0.2.6
+voir==0.2.8
  # via -r benchmarks/dlrm/requirements.in
 websocket-client==1.5.1
  # via

diff --git a/benchmarks/dlrm/requirements.in b/benchmarks/dlrm/requirements.in
@@ -11,4 +11,4 @@ torchx
 tensorboard
 
 # Following limits are for milabench
-voir>=0.2.4,<0.3
+voir>=0.2.8,<0.3
diff --git a/benchmarks/dlrm/requirements.rocm.txt b/benchmarks/dlrm/requirements.rocm.txt
@@ -336,7 +336,7 @@ varname==0.10.0
  # via
  # -c .pin-constraints-torch.txt
  # giving
-voir==0.2.5
+voir==0.2.8
  # via -r benchmarks/dlrm/requirements.in
 websocket-client==1.5.1
  # via

diff --git a/benchmarks/huggingface/requirements.cuda.txt b/benchmarks/huggingface/requirements.cuda.txt
@@ -175,5 +175,5 @@ varname==0.10.0
  # via
  # -c .pin-constraints-torch.txt
  # giving
-voir==0.2.6
+voir==0.2.8
  # via -r benchmarks/huggingface/requirements.in
diff --git a/benchmarks/huggingface/requirements.in b/benchmarks/huggingface/requirements.in
@@ -1,3 +1,3 @@
 torch
 transformers
-voir>=0.2.5,<0.3
+voir>=0.2.8,<0.3
diff --git a/benchmarks/huggingface/requirements.rocm.txt b/benchmarks/huggingface/requirements.rocm.txt
@@ -175,5 +175,5 @@ varname==0.10.0
  # via
  # -c .pin-constraints-torch.txt
  # giving
-voir==0.2.5
+voir==0.2.8
  # via -r benchmarks/huggingface/requirements.in
diff --git a/benchmarks/stable_baselines3/requirements.cuda.txt b/benchmarks/stable_baselines3/requirements.cuda.txt
@@ -304,7 +304,7 @@ urllib3==1.26.15
  # sentry-sdk
 varname==0.10.0
  # via giving
-voir==0.2.6
+voir==0.2.8
  # via -r benchmarks/stable_baselines3/requirements.in
 wandb==0.14.0
  # via -r benchmarks/stable_baselines3/requirements.in

diff --git a/benchmarks/stable_baselines3/requirements.in b/benchmarks/stable_baselines3/requirements.in
@@ -19,4 +19,4 @@ seaborn
 tqdm
 
 # Following limits are for milabench
-voir>=0.2.5,<0.3.0
+voir>=0.2.8,<0.3.0
diff --git a/benchmarks/stable_baselines3/requirements.rocm.txt b/benchmarks/stable_baselines3/requirements.rocm.txt
@@ -304,7 +304,7 @@ urllib3==1.26.15
  # sentry-sdk
 varname==0.10.0
  # via giving
-voir==0.2.5
+voir==0.2.8
  # via -r benchmarks/stable_baselines3/requirements.in
 wandb==0.14.0
  # via -r benchmarks/stable_baselines3/requirements.in

diff --git a/benchmarks/stargan/requirements.cuda.txt b/benchmarks/stargan/requirements.cuda.txt
@@ -152,5 +152,5 @@ varname==0.10.0
  # via
  # -c .pin-constraints-torch.txt
  # giving
-voir==0.2.6
+voir==0.2.8
  # via -r benchmarks/stargan/requirements.in
diff --git a/benchmarks/stargan/requirements.in b/benchmarks/stargan/requirements.in
@@ -1,4 +1,4 @@
 numpy
 torch
 torchvision
-voir>=0.2.4,<0.3
+voir>=0.2.8,<0.3
diff --git a/benchmarks/stargan/requirements.rocm.txt b/benchmarks/stargan/requirements.rocm.txt
@@ -152,5 +152,5 @@ varname==0.10.0
  # via
  # -c .pin-constraints-torch.txt
  # giving
-voir==0.2.5
+voir==0.2.8
  # via -r benchmarks/stargan/requirements.in
diff --git a/benchmarks/super-slomo/requirements.cuda.txt b/benchmarks/super-slomo/requirements.cuda.txt
@@ -157,5 +157,5 @@ varname==0.10.0
  # via
  # -c .pin-constraints-torch.txt
  # giving
-voir==0.2.6
+voir==0.2.8
  # via -r benchmarks/super-slomo/requirements.in
diff --git a/benchmarks/super-slomo/requirements.in b/benchmarks/super-slomo/requirements.in
@@ -4,4 +4,4 @@ numpy
 tqdm
 opencv-python
 
-voir>=0.2.4,<0.3
+voir>=0.2.8,<0.3
diff --git a/benchmarks/super-slomo/requirements.rocm.txt b/benchmarks/super-slomo/requirements.rocm.txt
@@ -157,5 +157,5 @@ varname==0.10.0
  # via
  # -c .pin-constraints-torch.txt
  # giving
-voir==0.2.5
+voir==0.2.8
  # via -r benchmarks/super-slomo/requirements.in
diff --git a/benchmarks/timm/requirements.cuda.txt b/benchmarks/timm/requirements.cuda.txt
@@ -168,5 +168,5 @@ varname==0.10.0
  # via
  # -c .pin-constraints-torch.txt
  # giving
-voir==0.2.6
+voir==0.2.8
  # via -r benchmarks/timm/requirements.in
diff --git a/benchmarks/timm/requirements.in b/benchmarks/timm/requirements.in
@@ -3,4 +3,4 @@ torchvision
 pyyaml
 huggingface_hub
 safetensors>=0.2
-voir>=0.2.5,<0.3
+voir>=0.2.8,<0.3
diff --git a/benchmarks/timm/requirements.rocm.txt b/benchmarks/timm/requirements.rocm.txt
@@ -168,5 +168,5 @@ varname==0.10.0
  # via
  # -c .pin-constraints-torch.txt
  # giving
-voir==0.2.5
+voir==0.2.8
  # via -r benchmarks/timm/requirements.in
diff --git a/benchmarks/torchvision/requirements.cuda.txt b/benchmarks/torchvision/requirements.cuda.txt
@@ -154,5 +154,5 @@ varname==0.10.0
  # via
  # -c .pin-constraints-torch.txt
  # giving
-voir==0.2.6
+voir==0.2.8
  # via -r benchmarks/torchvision/requirements.in
diff --git a/benchmarks/torchvision/requirements.in b/benchmarks/torchvision/requirements.in
@@ -1,4 +1,4 @@
 torch
 torchvision
 tqdm
-voir>=0.2.5,<0.3
+voir>=0.2.8,<0.3
diff --git a/benchmarks/torchvision/requirements.rocm.txt b/benchmarks/torchvision/requirements.rocm.txt
@@ -154,5 +154,5 @@ varname==0.10.0
  # via
  # -c .pin-constraints-torch.txt
  # giving
-voir==0.2.5
+voir==0.2.8
  # via -r benchmarks/torchvision/requirements.in
diff --git a/milabench/cli.py b/milabench/cli.py
@@ -11,7 +11,7 @@
 
 from coleo import Option, config as configuration, default, run_cli, tooled
 from omegaconf import OmegaConf
-from voir.instruments.gpu import get_gpu_info
+from voir.instruments.gpu import deduce_backend, get_gpu_info, select_backend
 
 from milabench.alt_async import proceed
 from milabench.utils import blabla
@@ -138,6 +138,22 @@ def get_base_defaults(base, arch="none", run_name="none"):
  }
 
 
+def deduce_arch():
+ """Deduce the arch for installation and preparation purposes"""
+ arch_guess = os.environ.get("MILABENCH_GPU_ARCH", None)
+
+ if arch_guess is not None:
+ return arch_guess
+
+ return deduce_backend()
+
+
+def init_arch():
+ """Initialize the monitor for the given arch"""
+ arch = deduce_arch()
+ return select_backend(arch)
+
+
 def _get_multipack(
  config_path,
  base=None,
@@ -177,15 +193,13 @@ def _get_multipack(
  sys.exit(1)
  overrides = merge(overrides, {"*": {"dirs": {"venv": venv}}})
 
- arch_guess = os.environ.get("MILABENCH_GPU_ARCH", None)
- arch = get_gpu_info(arch_guess)["arch"]
-
  if run_name is None:
  run_name = blabla() + ".{time}"
+
  now = str(datetime.today()).replace(" ", "_")
  run_name = run_name.format(time=now)
 
- base_defaults = get_base_defaults(base=base, arch=arch, run_name=run_name)
+ base_defaults = get_base_defaults(base=base, arch=deduce_arch(), run_name=run_name)
 
  config = build_config(base_defaults, config_path, overrides)
 
@@ -305,6 +319,9 @@ def run():
 
  mp = get_multipack(run_name=run_name)
 
+ # Initialize the backend here so we can retrieve GPU stats
+ init_arch()
+
  success = run_with_loggers(
  mp.do_run(repeat=repeat),
  loggers=[

diff --git a/milabench/structs.py b/milabench/structs.py
@@ -2,7 +2,7 @@
 from contextlib import asynccontextmanager, contextmanager
 from dataclasses import dataclass
 from itertools import count
-from typing import List, Dict
+from typing import Dict, List
 
 from voir.proc import LogEntry
 

diff --git a/tests/yoshua-benchio/requirements.in b/tests/yoshua-benchio/requirements.in
@@ -1 +1 @@
-voir>=0.2.5,<0.3
+voir>=0.2.8,<0.3