[29] Logging in json format #68
Conversation
|
|
||
| # TODO: performance: we repeatedly open the file for each call. Better for multiprocessing | ||
| # but we can probably do better and rely for example on the logging module. | ||
| with open(os.path.join(self.path_run, "metrics.json"), "ab") as f: |
There was a problem hiding this comment.
I suggest that we start with this simple version, we can always improve performance if it turns out to be a bottleneck
| ] | ||
|
|
||
| [tool.uv.sources] | ||
| flash-attn = { url = "https://github.com/Dao-AILab/flash-attention/releases/download/v2.7.4.post1/flash_attn-2.7.4.post1+cu12torch2.6cxx11abiFALSE-cp312-cp312-linux_x86_64.whl" } |
There was a problem hiding this comment.
I had to make this change to use uv on the hpc2020 cluster. I am not sure if this is going to be a breaking change for people. @clessig , do we assume that different HPCs can use different versions of CUDA? That sounds like a nightmare.
There was a problem hiding this comment.
We do not assume it, we know it ;) One can write a script that detects the available CUDA (and the python version if this is a variable) and then assembles the string that defines the wheel to be downloaded. @tjhunter : To what extent could one integrate this into pyproject toml?
There was a problem hiding this comment.
And could we open an issues to track this? :)
There was a problem hiding this comment.
I have the script in branch of the private repo but not committed yet:
#57
|
Ass discussed, will be followed up by #90 |
Closes #29
First part: prototyping the new format.