EasyDeL
EasyDeL copied to clipboard
erfanzar
Accelerate, Optimize performance with streamlined training and serving options with JAX.
Hi, I'm testing the attention mechanisms on kaggle TPU vm v3-8. it said below: ``` pallas_flash is Failed : Mosaic kernels cannot be automatically partitioned. Please wrap the call in...
**Describe the bug** Hi, when I'm finetuning gemma. the checkpoint size was a fixed value at the begining. Then it became bigger and bigger. Finally, when it reached 5.99GB, it...
**Describe the bug** Can't train with multiple VM's; TPU v-4-32 It stops after loading the model, won't even load the data Been trying for two days, maybe my set-up is...
**Describe the bug** I want to train a reward model using Easydel with sequence classification. The classifier has been implemented in the Flax sequence classifier classes for each model, but...
**Describe the bug** ``` --------------------------------------------------------------------------- TypeError Traceback (most recent call last) Cell In[3], line 22 19 dtype = jnp.bfloat16 21 # Load the pretrained model with automatic sharding ---> 22...
- Introduced `relu_squared` activation function to the ACT2FN dictionary in `easydel/infra/utils.py`. - Added unit tests for `relu_squared` and existing activation functions in a new test file `tests/test_activations.py`. - Updated CHANGELOG...
## Description While reviewing the codebase, I identified several areas where performance could be improved and memory usage reduced by leveraging `int1` kernels or boolean operations. These optimizations can be...
can the vinference or the regular inference classes be used on colab ? as the last time i tried it wasn't able to pick up the tpu's on the machine...
This pull request introduces improvements to how tool parsers are managed and used during streaming inference in both the `esurge` and `vsurge` servers. The main goal is to ensure that...