Set TF-associated dependencies #1143
Conversation
|
Do you have a preference for 2.4 vs 2.3 specifically at the cost of a dubious pinning for gast? |
|
For a stable environment I wouldn't mind. |
|
Ok, @scarlehoff please confirm explicitly you are happy with @RoyStegeman 's. |
|
I'll need to check (I'll do so later today) but from first inspection looks ok. The main change needed is that this PR will need to be merged and change nnpdf4 to point to that merge commit. |
|
I don't think that is strictly required: The only change is the environment and that we are pinning explicitly, no? |
|
Ah, I thought we wanted the env to be also (more or less) what you get when you do a clean "conda install nnpdf". But in that the only changes are some specific versions you are right |
|
It's not strictly required, and also the opt_einsum would cause a conflict with the environment, but I think we can bump the version number of the env without causing any new conflicts. Although I wouldn't know why we would want that. The point is that this env installs tf 2.3.0 but has the opt_einsum requirement of tf2.4.1. |
|
Scratch this. I can't do it right now, can you change the packages to the 2.3 versions? |
|
So does this PR now produce a package which is compatible with the environment in @RoyStegeman's comment? If so, what needs to be done to merge? Do we then need to re-tag? |
But at this point I really want @scarlehoff to confirm it is fine this time given how baldly I screwed up. |
|
Ah ok, I was scanning the comments for very specific tokens and missed that. |
|
Ok, the environment seem fine. After 20k of a global fit was still under 5 GB of RAM and the time is unchanged from my normal conda environment. The tests also pass, not sure what else to check, but if someone else has a look it would be nice. |
|
I had a look at various other things (like the pandas indexing in certain places) and that also seems fine to me. Michael, imho it makes sense to consider Roys environment final. I am afk to edit various wikis right now.
… On 12 Mar 2021, at 18:14, Juacrumar ***@***.***> wrote:
Ok, the environment seem fine. After 20k of a global fit was still under 5 GB of RAM and the time is unchanged from my normal conda environment. The tests also pass, not sure what else to check, but if someone else has a look it would be nice.
—
You are receiving this because you commented.
Reply to this email directly, view it on GitHub, or unsubscribe.
|
|
Ok I'm going to launch the closure fits with that env, at the very worst I'll just rerun them again again when we run into the next problem 😉 people seem to be getting impatient |
|
Greetings from your nice fit 🤖 !
Check the report carefully, and please buy me a ☕ , or better, a GPU 😉! |
|
Fit looks identical to when we merged #1081 see #1081 (comment) |
|
Right, but that's good, I didn't really expect this PR to affect the outcome of the fit. Or maybe I'm missing the point you're trying to make? |
|
I was just saying it was successful! |
|
The timings are also within minutes of each other so everything seems fine.
better safe than sorry! |
|
Good. Please merge this. |
See #1126 (comment)