rltakashige

[MEDIUM] Fix Local Network permissions in MacOS App

5

**Context:** The MacOS app requires Local Network access (in System Settings -> Privacy & Security) in order to allow devices to discover each other. **Issue:** This is configured correctly upon...

[HARD] Support arbitrary tensor parallel splits

14

**Context:** Tensor parallelism allows models to be split among multiple devices by distributing weights among devices. auto_parallel strategies for tensor parallelism use MLX LM's `shard_linear` and `shard_inplace` functions to split...

MacOS app uninstaller

1

When deleting the app, the plists persist. The main issue here is network setup. A simple uninstaller or even a cleanup script would prevent us leaving the host computers in...

## Motivation Simplification of API + moving model specific code to the runner ## Test Plan ### Manual Testing Tested that GPT OSS outputs are parsed correctly on the dashboard....

Fix placement pipeline edge case

1

## Motivation Issue discovered by some users. See https://discord.com/channels/1262421488506769518/1305414037453930496/1461864134831505716 ## Changes Fixed the placement algorithm to handle cases such as a (97.5GB, 1.25GB, 1.25GB) and for a model with 20...

## Motivation Some models hang in model loading with fast synch enabled. Handling loading by forcing each layer to be evaled seems to resolve this issue. ## Changes ## Why...