Recent Releases of Zeus
Zeus - Zeus Daemon v0.2.0
Change Highlights
CPU and DRAM energy measurements
Zeus daemon now also supports CPU and DRAM energy measurements with RAPL, which also requires root privileges just for measurement. Zeus daemon has also been integrated into the Zeus Python library, so as long as you have the daemon deployed and you set the ZEUSD_SOCK_PATH
environment variable, you'll be all set!
What's Changed
- [Feat] Implement CPU and DRAM monitoring for
zeusd
by @wbjin in https://github.com/ml-energy/zeus/pull/137 - Incorporate Zeusd for CPU and DRAM monitoring in ZeusMonitor by @michahn01 in https://github.com/ml-energy/zeus/pull/150
- Trace GPU ID in Zeusd GPU routes by @jaywonchung in https://github.com/ml-energy/zeus/pull/152
Consumption - Computation and Communication
- Python
Published by jaywonchung 3 months ago

Zeus - Zeus v0.11.0
Change Highlights
zeus
!
Renamed to Until now we used zeus-ml
because the name zeus
was taken on PyPI, but now we're finally able to move to zeus
:
pip install zeus
Prometheus Metrics
Zeus power and energy measurements can now be exported as Prometheus metrics! We currently support three metrics:
- Energy consumption of a fixed code range (Histogram)
- Power draw over time (Gauge)
- Cumulative energy consumption over time (Counter)
We wrote up a detailed metric monitoring guide and integration examples.
AMD GPU enhancements
We created an official distribution of ROCm AMDSMI Python bindings (GitHub, PyPI) and integrated it with Zeus. Before this, users had to cd
into their ROCm installation's AMDSMI distribution directory and run pip install
, which isn't very convenient.
Carbon Emission Estimations
The new [zeus.monitor.carbon.CarbonEmissionMonitor]https://ml.energy/zeus/reference/monitor/carbon/#zeus.monitor.carbon.CarbonEmissionMonitor) takes in a carbon intensity provider (e.g., from ElectricityMaps) and provides an estimate for operational carbon emissions. The window-based API is essentially the same as ZeusMonitor
.
Full Changelog
- [Misc] Reorganize Zeus NSDI 23 paper artifacts by @jaywonchung in https://github.com/ml-energy/zeus/pull/126
- [Docs] Add
BUILD_SOCIAL_CARD
env, skip social card build by default by @jaywonchung in https://github.com/ml-energy/zeus/pull/130 - [Feat]
CarbonIntensityProvider
and ElectricityMaps implementation by @danielhou0515 in https://github.com/ml-energy/zeus/pull/129 - [Misc] Fix link in PLO example README by @jaywonchung in https://github.com/ml-energy/zeus/pull/136
- Fix typo in profiler script by @dkopczyk in https://github.com/ml-energy/zeus/pull/138
- [Feat]
amdsmi
bindings integration by @parthraut in https://github.com/ml-energy/zeus/pull/132 - Make sure to assign EmptyCPUs to cpus if there is a permission error by @wbjin in https://github.com/ml-energy/zeus/pull/139
- [Feat] Implement CPU and DRAM monitoring for
zeusd
by @wbjin in https://github.com/ml-energy/zeus/pull/137 - [Fix] Fix tests failing due to deprecated
app
argument in httpx client by @jaywonchung in https://github.com/ml-energy/zeus/pull/140 - Out of Bounds Power Limit in
GlobalPowerLimitOptimizer
by @parthraut in https://github.com/ml-energy/zeus/pull/143 - [CI] Upgrade
actions/cache
to V4 by @jaywonchung in https://github.com/ml-energy/zeus/pull/144 - [Misc] Update Perseus paper link by @jaywonchung in https://github.com/ml-energy/zeus/pull/145
- [feat]
CarbonEmissionMonitor
by @danielhou0515 in https://github.com/ml-energy/zeus/pull/148 - Update
zeusd
dependencies following dependabot suggestions by @jaywonchung in https://github.com/ml-energy/zeus/pull/149 - [Feat] Prometheus metric export by @sharonsyh in https://github.com/ml-energy/zeus/pull/134
- Pytorch Fully Sharded Data Parallel (FSDP) Integration by @parthraut in https://github.com/ml-energy/zeus/pull/147
- Rename package from
zeus-ml
tozeus
by @jaywonchung in https://github.com/ml-energy/zeus/pull/151 - Incorporate Zeusd for CPU and DRAM monitoring in ZeusMonitor by @michahn01 in https://github.com/ml-energy/zeus/pull/150
- Trace GPU ID in Zeusd GPU routes by @jaywonchung in https://github.com/ml-energy/zeus/pull/152
New Contributors
- @dkopczyk made their first contribution in https://github.com/ml-energy/zeus/pull/138
- @michahn01 made their first contribution in https://github.com/ml-energy/zeus/pull/150
Consumption - Computation and Communication
- Python
Published by jaywonchung 3 months ago

Zeus - Zeus v0.10.1
This is a maintenance release aimed at enhancing usability and fixing small bugs.
What's Changed
- Feat: Catch
PermissionError
and raise with more information by @wbjin in https://github.com/ml-energy/zeus/pull/111 - Feat: Alternative RAPL directory inside Docker containers by @wbjin in https://github.com/ml-energy/zeus/pull/115
- Feat: added utility function to retrieve CPU index from PID by @danielhou0515 in https://github.com/ml-energy/zeus/pull/117
- Docs: More documentation on CPU monitoring by @wbjin in https://github.com/ml-energy/zeus/pull/118
- Feat:
python -m zeus.show_env
by @jaywonchung in https://github.com/ml-energy/zeus/pull/119 - Feat:
getAverageMemoryPowerUsage
by @jaywonchung in https://github.com/ml-energy/zeus/pull/122 - Fix: Add
getAverageMemoryPowerUsage
toGPUs
as well by @jaywonchung in https://github.com/ml-energy/zeus/pull/124
New Contributors 🎉
- @danielhou0515 made their first contribution in https://github.com/ml-energy/zeus/pull/117
Full Changelog: https://github.com/ml-energy/zeus/compare/zeus-v0.10.0...zeus-v0.10.1
Consumption - Computation and Communication
- Python
Published by jaywonchung 8 months ago

Zeus - Zeus v0.10.0: Broader support
What's New
CPU and DRAM energy measurement
We implemented support for Intel RAPL, which allows CPU and DRAM energy measurement on supported CPUs.
Generally speaking, most Intel CPUs support would support both and some AMD CPUs will support RAPL, albeit only CPU measurement.
JAX support
We added preliminary JAX support. Check out our full example here.
API usage is mostly identical:
monitor = ZeusMonitor(sync_execution_with="jax") # JAX!
monitor.begin_window("computations")
# Run computation
measurement = monitor.end_window("computations")
Zeus Daemon
Our energy optimizers require changing setting on the GPU, including power limit and frequency. This requires admin privileges. More details in our docs.
Zeus Daemon lets you circumvent this by running as a standalone daemon process on the node that implements privileged operations on your behalf, so that you don't have to give the entire Zeus-integrated application admin privileges.
We wrote the Zeus Daemon in Rust: Check out the source code and crates.io for details.
Breaking Changes
ZeusMonitor.begin_window
and ZeusMonitor.end_window
's second parameter sync_cuda
was renamed to sync_execution
.
This is because JAX asynchronously runs CPU code as well, and we would like to synchronize both CUDA and CPU computations. This created the need to generalize sync_cuda
to sync_execution
.
Changelog
- Docs: Add warnings about instantiating
ZeusMonitor
as a global variable. by @jaywonchung in https://github.com/ml-energy/zeus/pull/68 - Docs: Fix typo by @Sunt-ing in https://github.com/ml-energy/zeus/pull/69
- Docs: Improve the GPU energy monitoring demo by @Sunt-ing in https://github.com/ml-energy/zeus/pull/70
- Feat: Detect and reject unofficial
pynvml
bindings by @jaywonchung in https://github.com/ml-energy/zeus/pull/71 - Fix: Pandas warnings from
PowerMonitor
by @jaywonchung in https://github.com/ml-energy/zeus/pull/75 - Feat: Zeus daemon by @jaywonchung in https://github.com/ml-energy/zeus/pull/81
- Test: Allow
zeusd
dev and testing on MacOS by @jaywonchung in https://github.com/ml-energy/zeus/pull/82 - Refactor: Reorg
zeus.device.gpu
by @jaywonchung in https://github.com/ml-energy/zeus/pull/83 - Feat: Integrate
zeusd
intozeus.device.gpu
by @jaywonchung in https://github.com/ml-energy/zeus/pull/85 - Chore: Fix typo in GitHub Actions by @jaywonchung in https://github.com/ml-energy/zeus/pull/86
- Chore:
zeusd
debug outputs and doc comments by @jaywonchung in https://github.com/ml-energy/zeus/pull/87 - Feat: Add CPU measurement (via Intel RAPL) to ZeusMonitor by @wbjin in https://github.com/ml-energy/zeus/pull/90
- Fix: RAPL DRAM measurements not to be included in package measurements by @wbjin in https://github.com/ml-energy/zeus/pull/92
- Chore: Run checks in PRs from forks by @jaywonchung in https://github.com/ml-energy/zeus/pull/95
- Docs: Fix attribute name in
ZeusMonitor
example by @HGangloff in https://github.com/ml-energy/zeus/pull/96 - Feat: Add zero energy warning in
ZeusMonitor
by @sharonsyh in https://github.com/ml-energy/zeus/pull/93 - Feat: Add jax support in CUDA sync by @HGangloff in https://github.com/ml-energy/zeus/pull/97
- Docs: Refine JAX integration and example by @jaywonchung in https://github.com/ml-energy/zeus/pull/99
- Feat: Multi arch docker build by @sharonsyh in https://github.com/ml-energy/zeus/pull/104
- News: Add Perseus news and write Perseus blog by @jaywonchung in https://github.com/ml-energy/zeus/pull/107
- Feat: Multi-Arch Docker Build - Pushing to symbioticlab/zeus and mlenergy/zeus by @sharonsyh in https://github.com/ml-energy/zeus/pull/106
- Feat: RAPL Monitor for monitoring wraparounds for a rapl file by @wbjin in https://github.com/ml-energy/zeus/pull/105
- Test: Tests for CPU monitoring onn ZeusMonitor by @wbjin in https://github.com/ml-energy/zeus/pull/100
- Chore: Fix lint warnings from ruff by @wbjin in https://github.com/ml-energy/zeus/pull/108
New Contributors 🎉
- @Sunt-ing made their first contribution in https://github.com/ml-energy/zeus/pull/69
- @wbjin made their first contribution in https://github.com/ml-energy/zeus/pull/90
- @HGangloff made their first contribution in https://github.com/ml-energy/zeus/pull/96
- @sharonsyh made their first contribution in https://github.com/ml-energy/zeus/pull/93
Full Changelog: https://github.com/ml-energy/zeus/compare/v0.9.1...zeus-v0.10.0
Consumption - Computation and Communication
- Python
Published by jaywonchung 9 months ago

Zeus - v0.9.0: Batch size optimizer and big cleanups
What's new
- The batch size optimizer is now a full-fledged server that can be deployed independently, with Docker Compose, or on Kubernetes + KubeFlow.
- GPU abstraction: We created an abstraction layer over GPU vendors (NVIDIA and AMD). We're on our way to supporting AMD GPUs.
- Completely revamped documentation under https://ml.energy/zeus.
Deprecated
- See #20 (
ZeusDataLoader
,ZeusMaster
, and the C++ Zeus monitor)
Consumption - Computation and Communication
- Python
Published by jaywonchung 12 months ago

Zeus - v0.8.0: Energy-efficient large model training
This release features Perseus, an optimizer for energy-efficient large model training.
See the Perseus docs for details.
Consumption - Computation and Communication
- Python
Published by jaywonchung over 1 year ago

Zeus - v0.7.1: Moved to under `ml-energy`!
We moved our repository to under ml-energy
. No feature changes :)
Consumption - Computation and Communication
- Python
Published by jaywonchung over 1 year ago

Zeus - v0.7.0: Python-based power monitor
What's New
- We used to have a C++ power monitor under
zeus_monitor
, but we've deprecated that. There's no need for high speed polling because NVML power counters do not update that quick anyway.- In order to poll power consumption programmatically, use
zeus.monitor.power.PowerMonitor
.
- In order to poll power consumption programmatically, use
- CLI power & energy monitor:
python -m zeus.monitor power
python -m zeus.monitor energy
- We switched from the old
setup.py
to the new package metadata standardpyproject.toml
. - Docker image sizes are drastically smaller now! The compressed image used to be 8.48 GB, but now it's down to 2.71 GB.
Consumption - Computation and Communication
- Python
Published by jaywonchung over 1 year ago

Zeus - v0.6.1: `approx_instant_energy`
What's New
approx_instant_energy
in ZeusMonitor
- Sometimes, the NVML energy counter update period is longer than the measurement window, in which case energy consumption may be return as
0.0
. In this case, whenapprox_instant_energy=True
,ZeusMonitor
will approximate the energy consumption of the window as instant power consumption multiplied by the duration of the measurement window:\textrm{Energy} = \int_0^T \textrm{Power}(t) dt \approx \textrm{Power}(T) \cdot T
Consumption - Computation and Communication
- Python
Published by jaywonchung over 1 year ago

Zeus - v0.6.0: `OptimumSelector`
What's New
OptimumSelector
- Until know, the optimal power limit for
GlobalPowerLimitOptimizer
was the one that minimizes the Zeus time-energy cost. Not everyone would want that. - Now,
OptimumSelector
is an abstract base class with which you can implement your own optimal power limit selection policy. - Pre-implemented one are
Time
,Energy
,ZeusCost
, andMaxSlowdownConstraint
. These are thoroughly tested.
wait_steps
- Now, you can specify
wait_steps
inGlobalPowerLimitOptimizer
, and it'll wait for the specified number of steps before profiling and optimizing. wait_steps
is set to 1 by default to because users may havetorch.backends.cudnn.benchmark = True
andDataLoader
workers usually need time to warm up before ramping up to their normal fetch throughput.
Breaking Changes
GlobalPowerLimitOptimizer
now takes an instance ofOptimumSelector
in its constructor, instead ofeta_knob
. If you want to recover the functionality of v0.5.0, modify your code like this:# Before plo = GlobalPowerLimitOptimizer(..., eta_knob=0.5, ...)
# After from zeus.optimizer.power_limit import ZeusCost plo = GlobalPowerLimitOptimizer(..., optimum_selector=ZeusCost(eta_knob=0.5), ...)
Consumption - Computation and Communication
- Python
Published by jaywonchung almost 2 years ago

Zeus - v0.5.0: Big refactor, `GlobalPowerLimitOptimizer`
What's New
Callback-based architecture
zeus.callback.Callback
is the new backbone for Zeus componentsGlobalPowerLimitOptimizer
is the shiny new way to online-profile and optimize the power limit of DNN training.EarlyStopController
monitors and manages all sorts of conditions to determine whether training should stop.
Extensive testing
tests/
is richer than ever. With deep component tests with exhaustive parametrization, there are now around 1500 test cases.- Especially,
zeus.util.testing.ReplayZeusMonitor
exposes the same public API asZeusMonitor
but replays the measurement window logs produced byZeusMonitor
, instead of doing actual measurement. With this, Zeus can now be tested without any actual GPUs.
Consumption - Computation and Communication
- Python
Published by jaywonchung almost 2 years ago

Zeus - v0.4.0: `ZeusMonitor`
What's New
- Just measuring energy with Zeus has been non-trivial. Now,
ZeusMonitor
is the only way to measure time and energy consumed by an arbitrary set of GPUs from executing an arbitrary range of code. There should be one-- and preferably only one --obvious way to do it.ZeusDataLoader
was refactored to build aroundZeusMonitor
.ZeusMonitor
is quite thoroughly tested now.
Consumption - Computation and Communication
- Python
Published by jaywonchung almost 2 years ago

Zeus - v0.3.0: `ZeusMonitorContext` for in-training-loop profiling
What's New
ZeusMonitorContext
allows users to profile their per-iteration energy and time consumption.- It's aimed for those who would like to get a feel for the energy consumption of their DNN training job with a couple additional lines (as opposed to modified lines).
- Documentation and integration example: here
Consumption - Computation and Communication
- Python
Published by jaywonchung over 2 years ago

Zeus - v0.2.0: Single-Node Data Parallel Support
New Features
- Single-node multi-GPU data parallel training support added (#2)
zeus_monitor
is built at Docker image build time and baked into the image (#6)
Breaking Changes
ZeusDataLoader
's profile window for each power limit is now based on the number of iterations, not time. (#2)- This was done to ease synchronization between GPUs while profiling power limits.
- The
ZEUS_PROFILE_PARAMS
environment variable is now parsed as a comma separated string of the number of warmup and measure iterations. ZeusMaster
's constructor now takes argumentsprofile_warmup_iters
andprofile_measure_iters
.
Consumption - Computation and Communication
- Python
Published by jaywonchung over 2 years ago
