David Hall
David Hall
(obviously this is low priority and you have probably less bandwidth than I do, but I am somewhat interested in support for NIO Buffers. It's come up a few times...
also reproduces in version 2.0.0
fwiw, i get a segfault for any dimension >= 18, but not before
Also get a failure in sgetrf for dim >= 10
Huh ok. Thanks! netlib-java stopped working on ubuntu 20.04 since they stopped shipping gfortran3 and I didn't think to try On Thu, May 13, 2021 at 2:52 PM Ludovic Henry...
Looks like it's fixed! With tokenizers 0.15.2 and transformers 4.38.2: ``` 10000 0.010763333993963897 100000 0.11263033305294812 200000 0.2581357080489397 300000 0.4427691249875352 400000 0.5360061249812134 500000 0.7575962499831803 800000 1.2085898750228807 1000000 1.7549052499816753 10000000 23.24868266703561 ```...
Thanks! I can work around, particularly for these --> scalar conversions. (But for this case it also seems like a straightforward thing to do on the plugin-end?) I could be...
Sure, i meant those are the easy cases, and felt like a minimum. I think you can define something that's correct modulo floating point (and definitely not optimal) with transpose->reshape->reduce->reshape
(changing scan layers to a list comprehension works around)
ruled out vmap as a culprit. Same failure: ```python import jax import jax.random as jrandom from jax.sharding import PartitionSpec as P, Mesh, NamedSharding from jax.lax import with_sharding_constraint UP_PSPEC = P("data",...