free-music-demixer VS khoj

Hi HN,

Over the Christmas break I added multi-threading to the WASM Demucs module in freemusicdemixer

Demucs (v4 hybrid transformer) is a much higher quality model than the previous default, but it ran very slowly when limited to one worker: ~17 minutes for an average 4-minute song

I have since implemented multi-threading with WebWorkers.

If you raise the "MAX MEMORY" setting to 16 GB or 32 GB, your track will demix within 7-5 minutes, producing state-of-the-art results.

There is also support for the Demucs 6-source model which adds piano and guitar stems.

Please reach out and be loud about any bugs or UX issues you encounter!: https://github.com/sevagh/free-music-demixer/issues

* Post-processing step (bigger impact)

I tried to tackle the post-processing step in my C++ code (which would win ~1 dB in quality across all targets) but it's too tricky for now [2]. Maybe some other day.

1: https://github.com/sevagh/free-music-demixer/blob/main/examp...

2: https://github.com/sigsep/open-unmix-pytorch/blob/master/ope...

There are already several RAG chat open source solutions available. Two that immediately come to mind are:

Danswer

https://github.com/danswer-ai/danswer

Khoj

https://github.com/khoj-ai/khoj

I'm a fan of Khoj. Been using it for months. https://github.com/khoj-ai/khoj

https://github.com/khoj-ai/khoj

This is the easiest I found, on here too.

Thanks for the feedback. Does your machine have a GPU? 32GB CPU RAM should be enough but GPU speeds up response time.

We have fixes for the seg fault[1] and improvement to the query speed[2] that should be released by end of day today[3].

Update khoj to version 0.10.1 with pip install --upgrade khoj-assistant to see if that improves your experience.

The number of documents/pages/entries doesn't scale memory utilization as quickly and doesn't affect the search, chat response time as much

[1]: The seg fault would occur when folks sent multiple chat queries at the same time. A lock and some UX improvements fixed that

[2]: The query time improvements are done by increasing batch size, to trade-off increased memory utilization for more speed

[3]: The relevant pull request for reference: https://github.com/khoj-ai/khoj/pull/393

We recently integrated Llama 2 into Khoj. I wanted to share a short real-world evaluation of using Llama 2 for the chat with docs use-cases and hear which models have worked best for you all. The standard benchmarks (ARC, HellaSwag, MMLU etc.) are not tuned for evaluating this

Self-host your Notion assistant using the instructions here. You'll need Python >= 3.8 to get started.

Here's an early example: https://github.com/khoj-ai/khoj