James
James
Thanks, I think I understand what is going on.
I made a change in the RemNote codebase which will hopefully fix this issue :)
Maybe changing the name of the tool from python -> ipython?
Would it work better by formatting tool calls using xml? On Tue, 14 Jan 2025 at 18:04, Erik Bjäreholt ***@***.***> wrote: > Just happened to me again :( > >...
I'm not up to date on this, but maybe consider passing stop sequences? Like: ```py stop=[''] ```
Ohhh gotcha, makes sense On Sun, 26 Jan 2025 at 15:12, Erik Bjäreholt ***@***.***> wrote: > @bjsi The current stop/abort method works fine, > this is for when we *don't*...
Is there anyone who would be interested in maintaining it? On Tue, 28 May 2024 at 03:51, Tarun Yerneni ***@***.***> wrote: > Hi James, > > Is there any way...
Commenting out this line in `env/lib/python3.12/site-packages/colbert/training/training.py` fixes it for me # colbert = torch.nn.parallel.DistributedDataParallel(colbert, device_ids=[config.rank], # output_device=config.rank, # find_unused_parameters=True)
M2 GPU backend, 500mb input tensor: ```sh Benchmarking 508.6 MB... Time: 0.79s Throughput: 676.3 MB/s ``` Throughput should get much better as the input size increases because the number of...
NV tests were failing due to a bug in [GPUOcelot](https://github.com/gpuocelot/gpuocelot) where it didn't support the `shf` instruction. There were no details about this error in the CI logs, I was...