Awni Hannun
Awni Hannun
- Fixes scatter for large sizes - Probably will be part of a fix for some reductions for large sizes TODO: measure impact on runtime to see if we need...
Run the following with: ``` mpirun -n 4 -- python ``` ```python import mlx.core as mx import socket hostname = socket.gethostname() world = mx.distributed.init() rank = world.rank() print(f"Distributed available: {mx.distributed.is_available()}")...
This PR introduces two optimizations to handle much longer prompts and generations: ### Longer prompts - During prompt processing / cache prefiling, split the prompt into segments (512 is about...
Simple fix, as title.
The error message from running: `chat-with-mlx` is: ``` TypeError: unsupported operand type(s) for |: 'type' and 'type' ``` Which was only added in Python 3.10 and up. You might want...
I'm not sure if there is any change to MLX Swift or the examples to make it happen other than setting the right capability. But just want to open this...
It would be great to get the scikit-learn Array API back-end to be compatible with MLX (which is mostly conformant with the array API). Here is an example which currently...
The non-recursive destructor didn't catch two cases: - Destruction of arrays that have duplicate inputs - Destruction of arrays with siblings So this fixes that. Close https://github.com/ml-explore/mlx-swift/issues/139 Added some tests...
- Bump nanobind - Closes #1380
E.g.: ```bash mlx_lm.lora --model mlx-community/codegemma-7b-it-8bit --train --adapter-path adapters_codegemma_7B --data training_data --iters 500 ``` Can result in: ``` libc++abi: terminating due to uncaught exception of type std::runtime_error: [METAL] Command buffer execution...