mlx-examples
mlx-examples copied to clipboard
ml-explore
Examples in the MLX framework
When replacing LayerNorm with BatchNorm (merged in https://github.com/ml-explore/mlx/pull/217) with default params, the cifar example fails with ValueError with mlx 0.0.7: ``` File "../mlx-examples/cifar/main.py", line 48, in train_epoch mx.eval(model.parameters(), optimizer.state) ValueError:...
Hi, Is it possible to convert a [LoRA](https://github.com/ml-explore/mlx-examples/tree/main/lora) model trained with MLX back into the HuggingFace format to publish on the HuggingFace hub, and preferably merge it with the main...
This PR implements a example for the "Prompt Lookup Decoding" technique: https://github.com/apoorvumang/prompt-lookup-decoding using a local Mistral model. - This approach works best on input-grounded tasks such as summarization, document QA,...
_Continuation of #202. Decided to merge the Prompt Lookup Decoding under the Speculative Decoding example._ This PR implements a example for the "Prompt Lookup Decoding" technique: https://github.com/apoorvumang/prompt-lookup-decoding - This approach...
How small should loss reach, 0.1 or 1.0?
Based on lora.py and meant to perform supervised instruction fine-tuning for the various supported models, separating the model-specific logic from the common training and supporting logic (and management of training...
I downloaded pre-converted model from mlx-community from HuggingFace using the command ```bash huggingface-cli download --local-dir-use-symlinks False --local-dir tinyllama-1.1B-Chat-v0.6 mlx-community/tinyllama-1.1B-Chat-v0.6 ``` and tried to run ```bash python lora.py --model ./tinyllama-1.1B-Chat-v0.6 \...
Hi @awni 👋 I've tried to add several new features to the Whisper implementation through this PR, following the implementation of the original repository: - word-level timestamps (https://github.com/openai/whisper/pull/869) - word-level...
Since we already support the hf format model, it is better to merge the deepseek example with the hf llm example to reduce maintenance overhead. so, linear scale rope support...
I have run into a situation where my code continues to run for some time while nan values start to appear in mx arrays. I want to catch this early...
