BigsnarfDude

icon indicating a website link bigsnarfdude.github.io

icon location Canada icon location Standing on the shoulders of giants - ML, Deep Learning, and DFIR. Kaggle Expert. https://www.Kaggle.com/vincento. Python, Scala, Spaces, and VIM

Results 24 comments of BigsnarfDude

M1 perf. tests. Who has M2/M3 could you run tests and publish results?

``` transformer_lm % python main.py --gpu M1Pro 32GB - It/sec 1 M2Pro 32GB - It/sec 1.4 M1Ultra - It/sec 4.004 ```

M1 perf. tests. Who has M2/M3 could you run tests and publish results?

These MLX numbers are looking close to same performance slope here https://github.com/ggerganov/llama.cpp/discussions/4167

Loader of pretrain Resnet on imageNet

@thegodone found some code for alexnet and vgg with loading of pretrained here. might provide hints on a loader https://github.com/robertmccraith/mimm/tree/main/mimm/models

Is this loss curve normal

``` in the README.md loss curves expected | model | params | train loss | val loss | | ------| ------ | ---------- | -------- | | gpt2 | 124M...

Is this loss curve normal

i've trained 124m, medium, and large using both openwebtext and red-pajamas datasets. your iterations should be around 100k and you will reach same training and val loss as the gpt2...

Is there a way to set the default region?

I patched scalazon with pull request to allow .setEnpoint function to be called and ENDPOINT parameter to be passed in. Here is the patch if you are interested. Closing issue....

mistral example seems to hang on Loading model from disk.

can confirm m2pro 32gb works without issues. loads and outputs

LoRA training really slow

I've trained LORA on a M2PRO 32GB. Note batch-size to keep it from running outta memory `python3 lora.py --model converted_model2mlx --train --iters 1000 --batch-size 2` `Iter 410: Train loss 1.090,...

LoRA training really slow

`RAM Usage: 27.0/32.0GB - swap:0.3/2.0GB ` Never saw any jumps in swap after the initial model load. (dont have pycharm. i only use terminal and vim) can not recreate behaviour...

Example of a Convolutional Variational Autoencoder (CVAE) on MNIST

Nice work. I pulled your CVAE branch @menzHSE for the PR and tested with latest mlx-core and mlx-datasets. Tested on MBP M2Pro 32gb ``` ... Downloading http://yann.lecun.com/exdb/mnist/train-labels-idx1-ubyte.gz 32.0KiB (53.1MiB/s) Downloading...