ironbound

9 ms·

121.
by ironbound 2y ago
$1,300.00
122.
by ironbound 2y ago
For people new to this maybe check out this video, it explains how the internals run pretty quickly https://m.youtube.com/watch?v=UKcWu1l_UNw In theory if Anthropic puts research into the mechanics of the models internals,
123.
Diffusion LLM Has Arrived(arxiv.org)
2 points by ironbound 2y ago|0 comments
124.
by ironbound 2y ago
Testing extreme NVME offload (4 x Gen5x4) for DeepSeek R1Because PCI-E 5x16 (~60GB/s) is close to dual channel DDR5 bandwidth, this is the cheapest method to run huge models. Code: https://github.com/BlinkDL/fast.c
125.
Run Deepseek from fast NVMe drives(github.com)
5 points by ironbound 2y ago|3 comments
126.
by ironbound 2y ago
Daily summary of the stock portfolio would be welcome
127.
by ironbound 2y ago
self reinforcing without people (rlhf) is wild
128.
by ironbound 2y ago
Looks good for hobby projects, can flash from a phone via OTG USB.
129.
by ironbound 2y ago
people get lazy
130.
by ironbound 2y ago
'Being ignorant is not so much a shame, as being unwilling to learn.' Back in the day, you could install windows XP and be infested in minutes https://en.wikipedia.org/wiki/Blaster_%28computer_worm%29
131.
by ironbound 2y ago
Next you'll be saying "I dont need an immune system..." Fun fact an attacker only needs to steal credentials from the home directory to jump into a companies AWS account where all the juicy customer data lives, so there are r
132.
by ironbound 2y ago
I've had sucess with GaLore: Memory-Efficient LLM Training by Gradient Low-Rank Projection https://arxiv.org/abs/2403.03507
133.
by ironbound 3y ago
University of Minnesota researchers tried this and got banned! [0] https://www.theverge.com/2021/4/30/22410164/linux-kernel-uni...
134.
Schedule-Free Learning – A New Way to Train(github.com)
131 points by ironbound 3y ago|43 comments
135.
by ironbound 3y ago
Is there a reason to uses this over something like bazel build?
136.
by ironbound 3y ago
Sounds like you havent seen Wafer-scale integration computing, Tesla has one and comercial companies like cerebras will sell you a cabnet without the miles of fiber networking. https://www.cerebras.net/andromeda/
137.
by ironbound 3y ago
Looks like they are pushing data center cpu's not cards https://www.anandtech.com/show/18721/ces-2023-amd-instinct-m...
138.
by ironbound 3y ago
Sure till Nvidia's lunch is eaten by hardware AI companies https://www.cerebras.net/andromeda/ https://tenstorrent.com/grayskull/
139.
by ironbound 3y ago
Best start to a video this year, really wish more people would use a hook to get people engaged in learning.
140.
by ironbound 3y ago
Nope and I'd really like to see a yearly benchmark to create some sort of feature rating!
141.
by ironbound 3y ago
Try it out https://huggingface.co/spaces/CarperAI/StableVicuna
142.
by ironbound 3y ago
This was a good talk on routing https://theamphour.com/626-intelligent-routing-with-sergiy-n...
143.
by ironbound 3y ago
PSA: tldr is a magic word in LLM's
144.
LiveOverflow Looks at Prompting(youtube.com)
1 points by ironbound 3y ago|1 comments
145.
by ironbound 3y ago
you'll be limited right away by the context lenght of 4096 tokens
146.
by ironbound 3y ago
Could check this out https://github.com/hexops/mach-gpu
147.
by ironbound 3y ago
Re-using SSL is annoying..
148.
by ironbound 3y ago
https://huggingface.co/OpenAssistant/oasst-sft-4-pythia-12b-...
149.
OpenAssistant – Alignment Livestream(youtube.com)
2 points by ironbound 4y ago|0 comments
150.
by ironbound 10y ago
Wish this was covered more.. but here goes! Abusing a function with requests(Ddos/fail auth), to the point it's supporting third party service blocked said functions APIkeys.. would be an interesting attack vector.

More ›