MLOS
MLOS copied to clipboard
microsoft
MLOS is a project to enable autotuning for systems.
Makes config errors load time apparent instead of runtime apparent. Closes #818
A more significant rewrite of named tuple support inside mlos_core. This is based on comments in #811 as well as conversations with @bpkroth
- Creates the managed identity (MI) in the specified RG. If existing, operation is idempotent. - Assigns the MI as a user-assigned MI to the created control VM - Grants...
Merging data between storage backends is a little problematic atm due to the fact that the primary keys can overlap. For instance, Rather than using `config_id` (an autoincrement int that's...
Prior to #725 we could specify something like this in a reusable optimizer config: ```jsonc // MLOS Core optimizer { "class": "mlos_bench.optimizers.MlosCoreOptimizer", // Note: these variables can be overridden by...
In `_check_operation_status()` of `AzureDeploymentService`, debug logging relying on a JSON response from the management operation should be done after checking for response status code is 200.  In this case,...
For starters, `mlos_webui` seems an appropriate moniker to me. It should support 1. [ ] Visualization of `ExperimentData` from a shared `Storage` location. 3. [ ] Execution of existing `mlos_bench`...
Recently a [new version (v2.2.0) of FLAML](https://github.com/microsoft/FLAML/releases/tag/v2.2.0) was released and it introduces some new warnings and behavior: ``` INFO flaml.tune.searcher.blendsearch:blendsearch.py:169 No low-cost partial config given to the search algorithm. For...
If an experiment happens to fail on the first trial (maybe quota / provisioning issues) and then MLOS crashes after due to our user scripts expecting a result metric, trying...