| 24 Sep 2024 |
 hexa | I think this one has been failing for me on the linear-operator package | 11:41:02 |
 connor (he/him) | As a sanity check — has anyone been able to successfully use torch.compile to speed up model training, or do they also get a python stack trace when torch tries to call into OpenAI’s triton | 15:23:08 |
| 25 Sep 2024 |
 SomeoneSerge (matrix works sometimes) | It used to work but now our t2iton is lagging 1 major version behind | 19:36:58 |
 Gaétan Lepage | Because those geniuses are not able to tag a freaking release | 20:20:55 |
| Gaétan Lepage | https://github.com/triton-lang/triton/issues/3535 | 20:21:18 |
| SomeoneSerge (matrix works sometimes) | unstable-yyyy-mm-dd is ok for us; there were some minor but unresolved issues with the PR that does the bump though | 20:23:04 |
| 26 Sep 2024 |
| connor (he/him) | In reply to @glepage:matrix.org
https://github.com/triton-lang/triton/issues/3535 Well that’s an infuriating read | 16:33:18 |
| Gaétan Lepage | It's OK, OpenAI is just a small startup with only a few people.
And deep learning is not even their main activity | 17:07:38 |
| connor (he/him) | Yeah and they're definitely not a for-profit organization | 17:20:14 |
 @adam:robins.wtf | "open" is in their name | 17:24:26 |
 nim65s | it's such a joke that I find it sad it was not opened one day earlier | 17:28:20 |
| Gaétan Lepage | "I propose a 200€ bounty for this PR. Please git tag the freaking commit. | 21:09:04 |
| Gaétan Lepage | * "I propose a 200€ bounty for this PR. Please git tag the freaking commit." | 21:09:07 |
| Gaétan Lepage | The ease of spinning up a release is a decreasing function of the project/company resources. | 21:09:40 |
| nim65s | same issue on a one-man project abandonned for the last year or so: https://github.com/bab2min/EigenRand/issues/56 | 21:47:05 |
| nim65s | * same issue on a one-man project abandonned for the last year or so: https://github.com/bab2min/EigenRand/issues/56 : <48h | 21:49:56 |
| 28 Sep 2024 |
|  shekhinah changed their profile picture. | 07:04:58 |
|  kaya 𖤐 changed their profile picture. | 16:55:46 |
| 1 Oct 2024 |
|  -_o joined the room. | 21:00:15 |
| 2 Oct 2024 |
| hexa | Gaétan Lepage: please take care of tensordict | 00:25:19 |
| hexa | 
Download image.png | 00:25:22 |
| Gaétan Lepage | Sure, I will have a look right now.
I have not faced any failure on my end, weird... | 06:21:33 |
| Gaétan Lepage | Is this on staging ? | 06:23:26 |
| Gaétan Lepage | All failures that I was able to find on hydra are timeouts or upstream dependency failures.
I was able to build tensordict on all architectures... | 07:05:50 |
| hexa | this is on trunk | 11:03:39 |
| hexa | then you probably need to increase meta.timeout | 11:04:00 |
| Gaétan Lepage | Now that you say it, I remember this package being stuck (indefinitly) during mass rebuilds.
I don't know if increasing the timeout will help. When everything works fine, it builds in ~1min...
Also, nothing has changed in the derivation for the past few months. | 11:47:12 |
 Kevin Mittman (UTC-7) | Back from vacation | 18:23:19 |
| Kevin Mittman (UTC-7) | Redacted or Malformed Event | 18:32:05 |
| Kevin Mittman (UTC-7) | In reply to @ss:someonex.net
Kevin Mittman Hi! Do you know how dcgm uses cuda and why it has to link several versions? See libdcgm_cublas_proxy${cudaMajor}.so | 18:34:06 |
