mlx
mlx copied to clipboard
ml-explore
MLX: An array framework for Apple silicon
In my iOS project, I am adding dependency on mlx-swift (source code) and mlx is added as submodule for that. iOS project treats redundant-move, conditional-uninitialized, pessimizing-move, unused-lambda-capture, sometimes-uninitialized warnings as...
**Describe the bug** Opening `mlx-community/Qwen3-Embedding-8B-4bit-DWQ` from HF fails with `ValueError: Missing parameters: lm_head.weight.` The error output is: ``` Traceback (most recent call last): File "/Users/guy/Library/Application Support/CodeRunner/Debuggers/pdb.crDebugger/pdb_server.py", line 40, in main...
**Describe the bug** The gradient accumulation implementation with mx.eval causes a sudden high memory usage of > 10 GB. Before mx.eval it is around 600 MB which is reasonable. **To...
**Describe the bug** The exact same kernel produces different results with `mx.fast.metal_kernel` when running with an M1 Max processor than when running with an M3 Max. When using our old...
mlx_lm.server did not work for me after pip install mlx-lm openai. Can you please help? I have a 8GB Mac M1. I have other model locally successfully using MLX
I use mlx_lm.lora --model /Users/caozhihui/ai_model/Qwen/Qwen3-8B --train --data ./data to fine-tune. After fine-tuning, the model will directly reply after multiple rounds of inquiries. this is dataset {"messages": [{"role": "system", "content": "你是一名专业、温柔、亲切的高级客服，擅长解答用户关于电影票“出票失败”等问题。你所负责的渠道是【ocr】"},...
**Describe the bug** A slice in NumPy writes through to underlying array, MLX doesn't. MLX behaves like Python, not NumPy. **To Reproduce** ```python import mlx.core as mx a = mx.array([1,...
Hi! I'm trying to port my quantum computing and tensor network code from PyTorch to MLX, but I find `mlx.core.linalg.qr` and `mlx.core.linalg.svd` do not support matrices of complex64. These two...
