iAkashPaul

9 ms·

91.
by iAkashPaul 3y ago
Summarisation for calls, emails. Lots of extraction tasks & closed domain chatbots. Deployment is usually FastAPI for business logic, Langchain or MS/Guidance library, LLM hosted via. HF-TGI server
92.
by iAkashPaul 3y ago
Anyone got any recommendations for FOSS Generative Design instead of F360.
93.
by iAkashPaul 3y ago
That was always understood. I just didn't want confusion with the word dang.
94.
by iAkashPaul 3y ago
@dang could help here
95.
by iAkashPaul 3y ago
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
96.
by iAkashPaul 3y ago
Someone make an E7 type case with a slide under keyboard for Android/USB-C on the bottom right side.
97.
by iAkashPaul 3y ago
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.
98.
by iAkashPaul 3y ago
That watermark detection rights at the end is real sus
99.
by iAkashPaul 3y ago
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 RAG
100.
by iAkashPaul 3y ago
Exactly, I don't think this project uses VAD for pausing LLM generation or interruptions in general which is key to good assistant interactions
101.
by iAkashPaul 3y ago
Reminded me of Andy Barry's work on fixed wing UAVs dodging branches almost 10 years ago!
102.
by iAkashPaul 3y ago
I use Maccy for this alongside nifty macOS configurations
103.
by iAkashPaul 3y ago
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
104.
by iAkashPaul 3y ago
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, summar
105.
by iAkashPaul 3y ago
Reminds a lot of Kal by https://kal-store.com/
106.
by iAkashPaul 3y ago
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!!
107.
by iAkashPaul 3y ago
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.
108.
by iAkashPaul 3y ago
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
109.
by iAkashPaul 3y ago
Would be great to get Maccy & Rectangle working on Ubuntu!
110.
by iAkashPaul 3y ago
Fullcontrol & SimpleFOC based interaction or robotics is something that's extremely piqued my interest in electronics
111.
by iAkashPaul 3y ago
llama.cpp can run the Q4 variant of the same at 30tok/s on an M1 Pro unlike the 20tok/s being quoted
112.
by iAkashPaul 3y ago
Neat, I was thinking of virtual Link in GBA ROMs to talk with LLMs for dialogue :)
113.
by iAkashPaul 3y ago
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 expi
114.
by iAkashPaul 3y ago
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 charg
115.
by iAkashPaul 3y ago
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.
116.
by iAkashPaul 3y ago
INT4 or FP4 ones are of interest to me
117.
by iAkashPaul 3y ago
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
118.
by iAkashPaul 3y ago
Yeah even that as well
119.
by iAkashPaul 3y ago
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
120.
by iAkashPaul 3y ago
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 manufac

More ›