iAkashPaul

Anyone got any recommendations for FOSS Generative Design instead of F360.

That was always understood. I just didn't want confusion with the word dang.

@dang could help here

One quick check for any RAG system is to ask what all can the bot answer about. Generating scalable metadata at ingestion along with knowledge graphs make for a good closed domain experience

Someone make an E7 type case with a slide under keyboard for Android/USB-C on the bottom right side.

It's so weird that 10 years after seeing Andy Barry's drone dodging tree branches using high FPS low-res cameras(120FPS@320*240 via FPGAs), we still don't have cheap enough sensors to do this over USB3/4 today.

That watermark detection rights at the end is real sus

A go-to method is to ingest different chunksizes based on the document hierarchy & then use langchain with a bunch of retrievers depending on the doc type.

Then create an index about the metadata of each doc. So that you can ask the RAGbot what all it can answer about.

Another way to ensure it stays on-domain is to generate synthetic questions & check for similarity against user queries. There's a whole rabbit hole of query decomposition to avoid straying off topic as well.

Exactly, I don't think this project uses VAD for pausing LLM generation or interruptions in general which is key to good assistant interactions

Reminded me of Andy Barry's work on fixed wing UAVs dodging branches almost 10 years ago!

I use Maccy for this alongside nifty macOS configurations

There's a better parallel/batching that works on the 30s chunks resulting in 40X. From HF at https://github.com/Vaibhavs10/insanely-fast-whisper

This is again not native PyTorch so there's still room to have better RTFX numbers.

Sure but only for companies banking on grabbing users before stopping free-tiers or raising prices to match actual GPU/compute costs like MS/OpenAI/GitHub Copilot etc.

The rest are able to deliver great classification, summarisation, extraction, QnA & automated customer care & art(?).

Reminds a lot of Kal by https://kal-store.com/

Whoa this took me back to using iPodWizard.net for all the firmware changes I did to customize my iPod & the dual boot with Rockbox was neat too!!

From testing out the 1024px outputs it seems to work okay for text2img tasks only, other tasks get deepfried results. Also got it working using SDXL-LCM with similar results.

One of the ways we are creating a closed domain internal bot is by never relaying the user input as it is & combining it with vector search on hypothetical questions & limited user input. Kinda like induction. Along with making the model not use general knowledge & just the context strictly.

Would be great to get Maccy & Rectangle working on Ubuntu!

Fullcontrol & SimpleFOC based interaction or robotics is something that's extremely piqued my interest in electronics

llama.cpp can run the Q4 variant of the same at 30tok/s on an M1 Pro unlike the 20tok/s being quoted

Neat, I was thinking of virtual Link in GBA ROMs to talk with LLMs for dialogue :)

For people reading this & wondering how to deal with ARCON/ARCOS based access - just install code-server on the server & use VSCode for file-management, IDE & shell access all in a single tab.

No more session/SFTP expiry.

L4s are the only newer GPUs that you can get on GCP(g2) while A10 in AWS in the 1x config(g5), H100s are only in 8x variant otherwise.

Lambdalabs if you're lucky will grant 1x H100 at $2/hr :)

Every other serverless provider charges 3-5X the typical hourly cost.

I've tried the large-v2 on translate task but the results aren't great. Guess there needs to be another round of distillation with translate task thrown in too.

INT4 or FP4 ones are of interest to me

True, Intel is trying its best at both hardware & recently software contributions like Neural Compressor, so let's see how far they take nvidia to task

Yeah even that as well

At this point the M3 Max in a MacBook Pro with 96/128GB V/RAM needs to be benched against various Nvidia cards like 4090 & the H100 for $/performance numbers

Been hearing quick OS updates with project Treble for 6+ years now & no OEM has it as good as Pixel series. Lineage & PixelExperience ROMs are what's pushed my older devices to the latest OSes & be functional beyond manufacturers EOL

Think it could be the recent debacle of making clone apps as a marketing move & one CS rep misusing access to harass a customer over a project domain/name.