Kave Bahraman
Kave Bahraman
hey man. this issue is a bit vague to me. do you mean shuffling the order in which spectrum fill result_spectrum?
oh got it. thanks.
I have played with this, running Rust's release binary achieves faster results than Python's code; however, Python is about 2x faster than debug binary. Maybe there is something happening with...
I think you should submit the issue on tokenizers too. Python implementation is using rust bindings, no? Could U provide the full example to replicate the problem? Have you used...
I was able to replicate the result(omitted the http part). Rust is almost 2x slower than Python version. I checked the tokenizer, and they took the same time to execute,...
It appears that I didn't read instructions fully! Thank you for your response. P.S: Site doesn't work for me neither
Hi @msminhas93 how are you building you binary?
> You may want to check whether your simd instructions were properly detected, e.g. by printing `candle::utils::with_avx()` for x86. Shouldn't he build with RUSTFLAGS="-C target-feature=+avx" for this?
> Flamegraphs are pretty hard to look at as indeed lots of things will be hidden by rayon. Instead could you use the tracing functionality, e.g. something similar to [this](https://github.com/huggingface/candle/blob/15ed0b11cef868bbcf58dfa87c796db84fdcaff2/candle-examples/examples/bert/main.rs#L24),...
@LaurentMazare I'll be happy to help and learn in the way.