Skip to content

Telemetry

Principles:

  • No surprises — you will be notified before we begin collecting data. You will be notified of any changes to the data being collected or how it is used.
  • Easy opt-out: You will be able to easily opt-out of data collection
  • Transparency — you will be able to review all data that is sent to us
  • We will not sell data or buy data about you.

Why telemetry is useful for the infinity project

All CLI arguments are currently logged, and system info: OS, GPU, Processor, arm/x86

Examples how this improves infinity: - GPU model effects: - if e.g. flash-attn build would be useful / worth the effort, or not worth it because most card are still - default parameters, e.g. batch size - Model Architecture: - Helps create custom flash attention builds - If enough people deploy a DebertaV2 based model, work on faster implementation for that. - Devices: - prioritizes: apple mps / neuron / AMD depelopment - CLI args, to deprecate old ones or less popular ones such as onnx/tensorrt gpu + optimum or ctranslate2. Or to see if the v1 of the cli is still used.

Disable Telemetry

You can disable tracking like the following:

# set 
export DO_NOT_TRACK="1"
# infinity specific setting
export INFINITY_ANONYMOUS_USAGE_STATS="0"

This is in line with various FOSS projects: - https://docs.vllm.ai/en/latest/serving/usage_stats.html#usage-stats-collection - https://docs.ray.io/en/latest/cluster/usage-stats.html - https://github.com/langflow-ai/langflow/blob/eedfe43e6983bfed3c9a15e197c0206ea21b14eb/docs/docs/Contributing/contributing-telemetry.md?plain=1#L25 - https://github.com/LineaLabs/lineapy/blob/eebe5d5862f9888eaf619ba9bbeaa21b3ea5d2e5/README.md?plain=1#L266 - https://github.com/dbt-labs/dbt-core/blob/78c05718c589a56bc49f7520b47474690ae1cbe0/core/dbt/contracts/project.py#L19

Review which data is sent:

https://github.com/michaelfeil/infinity/blob/main/libs/infinity_emb/infinity_emb/telemetry.py