Quality Time

Code for the paper "Carbon-Aware Quality Adaptation for Energy-Intensive Services".

The datasets and full instructions on how to reproduce all experiments can be provided upon request. We will update the repository with the datasets and instructions in the future.

Setup

python -m venv venv
. venv/bin/activate
pip install --upgrade pip
pip install -r requirements.txt

Usage

Run experiments using python main.py, where you can override individual parameterizations using hydra. For example:

> python main.py optimizer=qt_online region=CISO vp=1,24,168 qor_target=.5 requests_dataset=wiki_en,wiki_de result_dir=final_results

Would run 6 experiments and store the results in ./final_results: three validity periods (vp) on two datasets (requests_dataset) with the qt_online optimizer in the CISO region and a minimum quality of responses (qor_target) of 0.5.

Full configuration:

> python main.py --help
main is powered by Hydra.

== Configuration groups ==
Compose your configuration from those groups (group=option)

optimizer: greedy, qt, qt_online
user_groups: p0, p10, p100, p20, p30, p40, p5, p50, p60, p70, p80, p90, p95, p96, p97, p98, p99


== Config ==
Override anything in the config (foo.bar=value)

seed: 0
qor_target: 0.5
requests_dataset: wiki_en
region: CISO
vp: 24
model_qualities:
- bad
- good
machines:
- _target_: src.scenario.Machine
  name: AWS_p4d.24xlarge
  power_usage: 3.7818
  pue: 1
  performance:
    llama8B: 41710
    good: 18176
  embedded_carbon: 135
result_dir: results
optimizer:
  _target_: src.optimizer.Qt
  callback:
    _target_: src.util.Callback
    gap: 0.001
    time_limit: 3600
  relax: false
  silent: false
user_groups:
  name: p0
  groups:
  - name: best-effort
    weight: 1
    slo_lower:
      bad: 1
      good: 0
    slo_upper:
      bad: 0
      good: 1