Martin Leitner-Ankerl

I've removed my `OLLAMA_MAX_VRAM` setting, then downloaded the latest RC with ```sh curl -fsSL https://ollama.com/install.sh | OLLAMA_VERSION="0.1.33-rc7" sh ``` Unfortunately there's not much differnce for me. `llama3:70b-instruct` crashes with CUDA...

Sure, I ran it 3 times, first run it gave me an error quite early and the CLI just hang, the other 2 runs looked the same and gave me...

I gave it a try, and get the same crashes. But I discovered something: I usually have the [Steam](https://store.steampowered.com/games/) client running in the background, and when I close it, `ollama...

I have Fedora 40, and as far as I know have all ROCm libraries installed. I can run `rocm-smi --showmeminfo vram` which shows the total RAM and used RAM: ```...

I googled a bit, I think an API like here in [P1010R1](https://www.open-std.org/jtc1/sc22/wg21/docs/papers/2018/p1010r1.html#approach) sounds good to me: * `T* uninitialized_data()` - Returns a pointer to storage that would back elements [size(),...

Nice nanobench benchmarks @Andersama :smile: I also don't like `insert_from_capacity`. Digging more into standardization, the last status of P1010 is that it should be revised, based on the accepted `resize_and_overwrite`:...

I did a git bisect, it seems in 2cd2773752371d6d6e933b5cd15f185fa8eaaaff the strange runaway time was introduced, most likely when indicatif was upgraded from 0.17.4 to 0.17.5 Before that, it was working...

I can't actually program in rust, I'd prefer if you can do that :)

Hi @vmilea, these are all very sensible points. I'll see that I can fix all of them. About (2), I'm thinking whats the best way to implement this is. E.g....

Thanks for reviewing @jonatack and @sipa! I've addressed all comments with e715861