automl
automl copied to clipboard
google
Google Brain AutoML
https://github.com/google/automl/blob/master/lion/lion_pytorch.py#L79: Now: ```python update = exp_avg * beta1 + grad * (1 - beta1) p.add_(torch.sign(update), alpha=-group['lr']) # Decay the momentum running average coefficient exp_avg.mul_(beta2).add_(grad, alpha=1 - beta2) ``` Can be:...
I'm trying to do quantization aware training on EfficientDet keras model. I tried use Nvidia [TensorFlow-2.x-Quantization-Toolkit](https://docs.nvidia.com/deeplearning/tensorrt/tensorflow-quantization-toolkit/docs/index.html), but I got error: `ValueError: to_quantize can only either be a tf.keras Sequential or...
hello, I try to use Lion optimizer, but I got the error : module 'tensorflow.keras.optimizers' has no attribute 'legacy'. Then I check the TF2 api doc(v2.11), it does not have...
Hi, Thanks for this amazing repo, I successfully trained a model on custom data and achieved great results. But when i try training on 2 gpu's using --strategy==gpus i met...
The doc says `mu` is inferred from `grads` and `updates` if `mu_dtype=None`. But [this line](https://github.com/google/automl/blob/ecbc7a3b5ab1944d5f39a80a21fd07b7606583f0/lion/lion_optax.py#LL108C45-L108C45) actually turns `jnp.bfloat16` and `jnp.float16` to `jnp.float32` when `mu_dtype=None`. Example on GPUs: ```python >>> jax.__version__...
When I use the Lion_pytorch, it comes out "UserWarning: __floordiv__ is deprecated, and its behavior will change in a future version of pytorch. It currently rounds toward 0 (like the...
Hi! When exporting a model with `python3 model_inspect.py --runmode=saved_model --tflite_path=model.tflite`, the exported TensorFlow Lite model is incompatible with mobile libraries like MediaPipe and the TensorFlow Lite Task Library, since they...
Hi anyone reproduce lite version mAP, in my side i try to retrain with COCO2017 to reproduce the mAP for example, lite1 my train result: 29.5% pretrained accuracy: 31.50% Thanks
