Ryan Landay comments

Results 14 comments of


                                            Ryan Landay

In emacs, M-backspace sends M-b "ackspace"

Hi, do you have any updates on this? This is pretty annoying for me and I imagine several dozen other emacs users as well.

Intermittent segmentation faults in llama_sample_top_p_top_k()

If I had to guess, the problem is that I’m passing an enormous value (16384) for repeat_last_n (since I was running into problems with repetitive output and wanted to basically...

Intermittent segmentation faults in llama_sample_top_p_top_k()

Thanks for the additional information. This code appears to only be getting called once per token generated, so I doubt there’d be hardly any performance impact from adding a bounds...

ggml : reduce the values of `ggml_op` enum

@ggerganov, it appears you did this already in 5b2b2dc6ae8086bff7c9b3c17fb435cf319b7185. I recommend closing this issue.

ggml : reduce the values of `ggml_op` enum

Oh never mind, I am looking at issues that are already completed.

TypeError: Input tensors need to be on the same GPU, but found the following tensor and device combinations

Oh, and I had gradient checkpointing enabled, as you can see from the stack trace. I already found one bug with using both gradient checkpointing and LoRA that I had...

TypeError: Input tensors need to be on the same GPU, but found the following tensor and device combinations

I have not solved it. I tried again on 2024-07-26 but I ran into the same problem. I suppose it’s been long enough that maybe I should try it again...

TypeError: Input tensors need to be on the same GPU, but found the following tensor and device combinations

I’m already doing that. I just finished reading the QLoRA paper and it does recommend using bfloat16 for optimal numerical precision, so I agree that that’s the right setting to...

sl pull hangs on sapling.ext.github.pr_marker.cleanup_landed_pr_hook

The workaround I found was to clone the repo with `sl clone` instead of `git clone`. There was another issue I ran into with `git clone` which was that bookmarks...

“ld: symbol(s) not found for architecture x86_64” when installing 3.14t-dev over a previous snapshot

Note: I accidentally wrote 2.4.1 for my pyenv version in the initial report but it was actually 2.4.11. I’ll update to 2.4.12 and try again but I assume I’ll get...