Відео

I am very happy that it helps you. :D

Full guided video tutorials take a lot of time and effort, but I have thought about it. Even these quick videos have taken more time than I guessed they would. I have a bunch more features I want to put into this project. Once it is a bit more feature complete I may take some time to do a real tutorials. But, my skills are more aligned with coding than video production...

I was thinking about sending you a message about if you could help me figure out CI for mac?

Well, I did the CI for mac after thinking about it yesterday. :D Hope to see you soon!

Not real soon, although it wouldn't be that difficult to manage. I can't run it on my server or desktop. Once Candle can do quantized models with Cuda I'll add it. :D Since I don't have much time to work on this I've been working on local changes that I'll share soon. I have added Phi models, and plan to add Llama very soon. Once Llama is working I'll push changes.

Fantastic! That's exactly what I created this for. I'll be improving things a bit soon - once the new year settles in.

Thank you for pointing this out. I'll be more careful next time.

Feel free! I have a lot of changes coming soon, but I have to pass through a busy period first. The biggest improvement that needs doing is fixing the websocket to work as a mpsc.

Glad you like it. Big improvements coming soon-ish.

Glad it was helpful!

I plan on doing RAG for text/PDF eventually, but it's a bit down in the list of priorities. I will try to remember to comment here when get to it. :D

@@danielcloughyou’re awesome; ya on your own time whenever you get to it. Putting this out for free is enough and grateful :)

🤗

Mostly because lepto-axum is slower to develop because it would restart the backend (and reload the model) every time I make a change in the frontend.

I think that is a great idea. I am going to finish updating the frontend a bit, probably tomorrow, then I could make it work with shuttle.rs. First I will want to incorporate the lama.cpp format .gguf in this project - because serverless won't run on GPU - and the more standard format is already a requested feature. I skipped it initially because it's currently CPU only with candle. Not sure about an ETA to have a serverless example up, but I think it's a great idea to make it a priority. Thanks for the suggestion.

I have made a serverless branch. Github in video description. ua-cam.com/video/awuE_-6omxY/v-deo.html

The main benefits, from my perspective are: 1) no python/pip (faster, simpler package management, no GIL) 2) strongly typed 3) memory safe If planning on running the project serverless you will have lightweight binaries to publish. I actually wasn't even going to use wasm before a friend was like, 'you might as well make the project "Pure Rust"'.

My anecdote: If you're building desktop applications or things that are designed to be run by end users, it's nice to get a single (small-ish) executable. My application is 22mb and takes about 100 megs of ram. It's a single file with no install or virtualenv setup. My Python virtualenv is 4 gigabytes, for comparison, and if I made a stripped down Docker container that still used the GPU it would be around the same size because of the Torch library overhead.

Thank you! Cheers!

Han si Chi hu dai fa Chuan Shu ka Ma ghji ti Pu xa

San liu Hui Chuan Shu nenng Cheng FA te Hai ka Ma ghji ti Tai zai Tai Hai she Tai Zen xa

It's a rare form of Wudang Wuxingquan.

For couple of years I’ve mastered the form from this video (and few scraps found in chinese videoservices). Still making improvements in my routine though, not all details are clearly visible. Next month will demonstrate it on local wushu competitions.