Skip to content

profiles.md

Latest commit

 

History

History
184 lines (129 loc) · 3.99 KB

profiles.md

File metadata and controls

184 lines (129 loc) · 3.99 KB
Raw

Configuration Profiles

OpenMed supports configuration profiles to quickly switch between different settings for development, production, testing, or custom workflows.

Built-in Profiles

OpenMed ships with five built-in profiles:

Profile Backend Batch size Timeout Use case
dev Auto Backend default 600s Development and debugging
prod Auto Backend default 300s Production deployments
test Auto Backend default 60s Testing and CI
fast Auto Backend default 120s Quick experiments
low_resource CPU ONNX INT8 1 900s CPU-only machines with 4-8 GiB RAM

The low_resource profile lazily loads the official small PII model, caps inference at one note per batch and one worker, and never falls back to the Torch backend. The tested model snapshot is pinned for reproducible memory and accuracy behavior. Install its runtime with pip install "openmed[onnx-runtime]". See the low-resource benchmark for the memory envelope and reproduction commands.

Using Profiles

Python API

from openmed import OpenMedConfig, analyze_text

# Create config from profile
config = OpenMedConfig.from_profile("dev")
print(config.log_level)  # "DEBUG"

# Apply profile with overrides
config = OpenMedConfig.from_profile("prod", timeout=600)

# Apply profile to existing config
config = OpenMedConfig(default_org="MyOrg")
new_config = config.with_profile("dev")

Environment Variable

Set the OPENMED_PROFILE environment variable:

export OPENMED_PROFILE=prod
python my_script.py
from openmed import OpenMedConfig

# Profile is automatically applied
config = OpenMedConfig()
print(config.profile)  # "prod"

Custom Profiles

Creating Custom Profiles

Programmatic example:

from openmed.core.config import save_profile

settings = {
    "log_level": "INFO",
    "timeout": 450,
    "device": "cuda",
    "use_medical_tokenizer": True,
}

save_profile("myproject", settings)

Loading Custom Profiles

from openmed import OpenMedConfig

config = OpenMedConfig.from_profile("myproject")

Deleting Custom Profiles

from openmed.core.config import delete_profile

delete_profile("myproject")

Note: Built-in profiles (dev, prod, test, fast, low_resource) cannot be deleted.

Profile Storage

Custom profiles are stored in TOML format at:

~/.config/openmed/profiles/<name>.toml

Example profile file:

# OpenMed profile: myproject
# Custom profile configuration

log_level = "INFO"
timeout = 450
device = "cuda"
use_medical_tokenizer = true

Profile Precedence

Profile settings are applied in this order (later overrides earlier):

  1. OpenMedConfig defaults
  2. Environment variables
  3. Config file (~/.config/openmed/config.toml)
  4. Profile settings
  5. Explicit arguments
# Profile overrides config file, explicit args override profile
config = OpenMedConfig.from_profile("prod", timeout=900)
# timeout=900 overrides prod's timeout=300

Workflow Examples

Development Workflow

from openmed import OpenMedConfig, analyze_text

# Use dev profile for detailed logging
config = OpenMedConfig.from_profile("dev")

result = analyze_text(
    "Patient has chronic myeloid leukemia.",
    model_name="disease_detection_superclinical",
    config=config,
)

Production Deployment

from openmed import OpenMedConfig, BatchProcessor

# Production settings with minimal logging
config = OpenMedConfig.from_profile("prod")

processor = BatchProcessor(
    model_name="disease_detection_superclinical",
    config=config,
)

results = processor.process_files(clinical_notes)

CI/CD Testing

export OPENMED_PROFILE=test
pytest tests/

Switching Profiles Dynamically

from openmed import OpenMedConfig

# Start with dev
config = OpenMedConfig.from_profile("dev")

# Switch to prod for final run
config = config.with_profile("prod")