dugujiujian1999
dugujiujian1999
@AlvaroMaza Hello! I have a pr right now. And I am changing this file too. Would you be comfortable if I added you as a Coauthor (Co-authored-by) to acknowledge your...
Looks like the import breaks something.
I believe it would be beneficial to reconsider the import method. Here's the suggested approach ``` if backend.backend() == "tensorflow": from keras.utils.module_utils import tensorflow as tf print("Using TensorFlow backend.") print("TensorFlow...
@AlvaroMaza Thank you. I've integrated the patch you provided and subsequently opened a new pr here. FYI: https://github.com/keras-team/keras/pull/19029
Running on colab T4 --- - batch_norm_op_jax_after (HEAD -> 936dd1345d794e91b3883bf99dec66dc8021e7fc) 102967424/102967424 ━━━━━━━━━━━━━━━━━━━━ 6s 0us/step 101/101 ━━━━━━━━━━━━━━━━━━━━ 165s 1s/step - loss: 0.5308 training: 1111 ms/step 101/101 ━━━━━━━━━━━━━━━━━━━━ 33s 269ms/step inferencing: 267...
@jackd i don't know. It takes less time after the patch. i use the code there: https://github.com/haifeng-jin/keras-benchmarking/tree/main/prs
``` import scipy sparse_weights = scipy.sparse.csr_matrix(my_sum.get_weights()) print(sparse_weights) ``` How about obtaining the weight?
@innat So do i. 👍
1. Save in keras format ``` import os os.environ['KERAS_BACKEND'] = 'jax' import keras import numpy as np window_size = 100 inference_window_size = None X = np.random.random((1024, window_size, 1)) Y =...
@lbortolotti Try this ``` spec_shape = str(spec_shape).replace("None", "b", 1).replace("None", "d") ``` --- https://github.com/keras-team/keras/blob/c4dd4fab5bd9491b32b4ab1e360d290ad8e8a238/keras/backend/common/variables.py#L422-L424 ↓ ``` if config.backend() == "jax": if str(e) == "b" or str(e) == "d": # JAX2TF tracing...