Chen Qian
Chen Qian
We have a runnable script in `examples/glue_benchmark/glue.py`, it would be nice to have a keras-io example to illustrate each step.
We need to verify that `DeBERTa` works fine on the common benchmark, as the implementation is a bit complex.
Because of the capacity limit, we cannot do TPU testing on every PR push. We have two options here: 1) type gcbrun/ manually to trigger TPU testing, which should happen...
Now if the `max_seq_len` goes over the model limit, it will error out. We need to be able to generate arbitrarily long sequence. To achieve this, one thing we can...
We can write one mega guide, or a few ones covering the following topics: - Use `keras_nlp.models.GPT2CausalLM` to generate text. - Finetune `keras_nlp.models.GPT2CausalLM` to generate text matching a certain style,...
We have added a `generate()` method to `GPT2CausalLM`, and we need a way to benchmark this API since performance is a key to text generation. More details will be added...
We need guidelines for our contributors working on model specs, such as compilation defaults. What we need to do: - provide a guideline on what should be tested, and what...
Our accelerator testing is taking too long, and there are some tests we don't necessarily want to verify on GPU specifically, such as saving testing.
We need to do some manual checks to ensure the quality of contrastive search, and do the switch.
# What does this PR do? Add a monitor of mlflow logger so that it sets run status as failed if main thread exits unexpectedly # What issue(s) does this...