Evgenii Zheltonozhskii

```python config = tf.ConfigProto() # Turns on XLA JIT compilation. config.graph_options.optimizer_options.global_jit_level = tf.OptimizerOptions.ON_1 run_metadata = tf.RunMetadata() sess = tf.Session(config=config) tf.global_variables_initializer().run(session=sess) ``` I've added theses lines to enable XLA

The standard fetching is still available, it's just that you'll need to download only a couple of hundreds of newest papers from API

it seems that @karpathy is rewriting it (from scratch?): https://www.reddit.com/r/MachineLearning/comments/obne9p/d_is_arxivsanity_down_what_people_use_these_days/h3q422o Hopefully it'll be part of the features of new version.

Looks like I'm facing the issue once again

ping @karpathy

@karpathy sorry to bother, but looks like it's happening once again. For "top resent" too

@karelin I've never had empty 3-day period except when the bug occurred. The fact that yours is empty too is additional evidence. Obviously for one-month period there will be papers...

@karpathy Well, today is Tuesday, so I suppose that's not the reason

@karpathy Just want to note that the same thing still happens from time to time (i.e. today), and last week too

After launching a local instance, I've noted that sometimes when I try to update PDFs, arxiv bans me (makes sense given I downloaded order of 100000 papers from there). If...