Today is being acquired by NVIDIA!
Brev’s goal is to build the easiest way for AI/ML developers to use a GPU.
Teaming up with
@nvidia
means being able to deliver on that mission by pairing the most powerful hardware with the industry leading software.
Hugging Face has a great notebook using LoRA to augment diffusion models using 🤗 Transformers and 🤗 Diffusers
1-click deploy to run it on a GPU w/ everything set up:
🤙🤙🤙
🚀 Pause & resume your Lambda Cloud GPU instances effortlessly with 's new lambda integration 💡
✅ Persistent volumes for seamless start/stop
🔋 Auto-stop on idle or job completion
💰 Scale to CPU when GPU not in use & save more!
OUTSIDE LLMs
Be amidst the best of the best at the biggest Gen AI Music Hackathon in 🇺🇸
8/4 ~ 8/6
@OutsideLLMs
The official hackathon for
@sfoutsidelands
in partnership with
@EMPIRE
(and more that we can't wait to reveal)
Applications open:
🚨Introducing Brev 2.0🚨
• Cheap, actually available GPUs — $2/hr A100s 🤑
• The simplest GPU provisioner 🤙
• Open-source tool to set up python & CUDA once and for all 😅
We hope this is the easiest way to fine-tune anything.
There are 3 main changes. Here’s why & what 👇
🚨Introducing Brev 2.3🚨
Getting a GPU can fail because of no availability, even if other regions in the same cloud have healthy supply
Brev's Baby Provisioner™️ automatically searches every region in the cloud to find the available GPU
Big provisioner 🔜
let it rip 🤙
🚨 Major Release 🚨
Reduce your dev cloud bill by up to 94%
You can install the Brev CLI on ANY cloud dev environment for instant optimizations
The best part? You don't even need a Brev account (but we'd love for you to make one)
Day 4⃣: Fine-tune Phi-2!
2 guides to fine-tune Microsoft's 2.7B Phi-2, which has a nearly SOTA performance among under 13B-parameter models
Got this one out ASAP after hearing the requests! Hope it's helpful! 🎅
link to the guides below // Day 5 tomorrow 🎁
🚨Introducing Brev 2.1🚨
• Credits for super transparent billing
• More NVIDIA A100s available
• too many stability upgrades to count😅
It’s pretty much the 2.0 that can support all the growth we’ve seen in the last few months
Let it rip 🤙
Requesting feedback for upcoming university tours….
What AI workshops / topics are you interested in?
Also I made a new IG @ harpercarrollai
Hope to see you there!
v0.6.80:
The week of improvements:
- Clone teammate’s Workspaces without Git Repositories
- Faster log in by adding `--skip-browser` to `brev login`, no need to hit enter
Curious about our bug fixes? Read about them here:
v0.6.71:
Big things this week:
• Setup scripts for reproducible dev envs (w/o committing to repo!!)
• Auto-install vscode extensions
• "brev shell" to jump to a terminal in any workspace
+ a ton of smaller improvements
🚨Brev Launch Week Day 1🚨
👉 Scalable Storage 👈
Has this happened to you?
It's so painful to run out of storage while working. You can now seamlessly scale up your hard drive with Brev without having to set anything up again.
🗓️ v0.6.89
Brev’s new update lets you modify or add new Git URLs to existing dev environments. This supports workflows such as starting from an empty dev environment and building it out as you go.
ALSO squashed some general stability bugs. Read here —> []
v0.6.6 has been released and it got a whole bunch lighter 🤙
We're creating cloud computers for development that are indiscernible from local development, and today's release is the largest step we've made towards that direction!
Just posted a video about one of my favorite tools for working with
@ollama
at the moment:
@brevdev
. It's an amazing way to find available GPU's on new instances. Spin them up and shut them down when done. Check out the video here: It will go live soon
Why is every hackathon just using OpenAI APIs?
We're hosting a hackathon this weekend
@brevdev
,
@metaphorsystems
, and
@TheCeloCenter
The only rule is that you train your own model. We're offering free GPUs, food & drink. So come and let it rip 🤘
🌱Sponsor: Brev🌱
Brev is a dev environment to get your hackathon project deployed & ready for scale in < 5 mins. Spend the hackathon building, not messing with infrastructure and configs. Book 1-on-1 time now
or check out !
New Release: 📢Public Ports📢
One-click to expose a port on your dev environment & share what you're running with the world
Here's a guide to get a stable diffusion API while you make dev changes!
What's Q*?! Reinforcement Learning, Q-Learning, Model-Based vs. Model-Free Methods, & Temporal Difference Learning
This Brev Concepts video covers the basics, is accessible to all backgrounds, and includes a little math for those interested.
YouTube link & timestamps below 🥳👇
🚨 Launch Week Day 3🚨
✨ 𝕁𝕦𝕡𝕪𝕥𝕖𝕣 ℕ𝕠𝕥𝕖𝕓𝕠𝕠𝕜𝕤✨
Access any instance with a Jupyter Notebook right from your browser 🤙
Pro-tip: use Web Tunnels from yesterday’s feature launch to securely share your notebook session with a friend 🔥
The
@BrevDev
beta is finally out!! If you don’t wanna waste time with infra and environments, lemme onboard you (link in bio). Trade me 20 min for a few hours :)
Chat with your PDF using all open-source tools
This simple guide 👇 uses OCR + MistralLite + prompt engineering to create a chatbot for your PDFs
MistralLite, Amazon's variation of Mistral, allows for 32K context tokens, which can fit about 48 pages / 24,000 words.
P.S.
I tried fine-tuning Mixtral on an 8xA10, not A100, to save costs but it didn't work
I want to make sure there's a cost-effective guide out first, so I applied QLoRA and got it working on a smaller GPU -- guide out tomorrow
Full Mixtral 8xA100 hopefully Friday 🤙
Outside Lands is Silicon Valley’s hottest music festival.
But something huge just happened.
Artists, musicians, and engineers from all over the world are here to show what’s possible at the cutting edge of AI and music
Live tweeting the demos from the
@OutsideLLMs
hackathon🧵:
Wanna deploy your Mistral 7B model once you finish fine-tuning it?
I made a super-simple guide to deploy any fine-tuned model on
@replicate
🔥
Video tomorrow 🤙 Notebook below 👇
🦙LET'S FINETUNE LLAMAS🦙
Even with Llama2 (and CodeLlama 🔜 ) NOT ENOUGH PEOPLE ARE FINE-TUNING
Saturday in SF 9am-5pm
Come learn from
@hwchase17
@WilliamBryk
@MichaelRoyzen
@samlhuillier_
and more
Bring your weird data & leave with a model 🤙
v0.6.75:
• Type in "brev shell" to get a terminal for any of your workspaces (ssh wrapper without config)
• Performance 😎 We're faster (32 - 51% to be exact-ish)
v0.6.79:
- You can now add environment variables that are scoped to a workspace
- Added new editor options (Vim, Emacs, Atom) during the CLI onboarding
You can read more every week on our changelog 👇
DBRX is the newest MoE model from
@databricks
that's outperforming GPT-3.5, 2x faster than Llama-70b
We spent the day messing around with it
Here's a notebook that uses vllm and
@gradio
to run interactive inference. Check it out, link below 😊🤙🔥
@ndzfs
@teeyehnee
Hey thanks for the shoutout!
Here's a guide to fine-tune llama2:
You might also like Mistral 7B, it's the latest model and is performing really well. Here's a guide for that one:
Someone in our Discord asked when it's better to fine-tune an LLM vs. use RAG
The answer? It depends. Are you trying to emulate data or reference it?
For example, fine-tuning a model on your user data is useful if you want to generate fake user data. In contrast, if you want
Episode 8: Fine-tuning R&D
We learn about fine-tuning and plan a new bounty for our web UI.
Watch us stumble through an excellent tutorial from
@brevdev
in which we fine-tune Mistral 7B on a custom dataset with a story about the e/acc rebels who overthrow the Worldcoin empire.
Just launched a ControlNet template & guide with
@brevdev
!
ControlNet is great for making incredible variations of your logos or QR codes - here is a blend of our logo with The Endless Summer image.
Guide includes an instructional video...ENJOY
Shout out
@CeloOrg
@brevdev
@roboflow
for making our AI & Music hackathon for
@sfoutsidelands
part of the Endless AI Summer!!
Someone needs to make some AI surf music videos to go with the theme !!
🤖🎶🌊🏄♂️
🚨 Launch Week Day 2 🚨
✨𝕎𝕖𝕓 𝕋𝕦𝕟𝕟𝕖𝕝𝕤✨
🔐Get a public URL to a service you're running on localhost, but w/ granular permissions so it's not just open to the public internet
To set up the tunnel, just enter the port and who to share it with🎉
Mistral 7B outperforms Llama 13B on all tested benchmarks
Made a guide showing how to fine-tune it cost-effectively using QLoRA
Jupyter notebook linked below 🤙
“we look at things like , for example. They've really built a serverless platform that's just really, really focused on making that whole life cycle easy for developers which is something that we haven't seen as much from the large cloud providers”
🚨Brev Launch Week Day 4🚨
The video says it all 🥳
Automatic storage/volume retirement is now live. If a large machine is not being utilized, we get rid of the unused storage, reducing costs by at least ~50%, and way more for more expensive instances 🤙
It's almost summer and we've been shipping. Time for launch week!
🚨Brev Launch Week Day 1🚨
👉 GCP Integration support 👈
Brev now offers instances from GCP meaning more A100 availability, but also, connect your own GCP account to use Brev with your GCP ✨and✨ AWS credits!
v0.6.86:
🛑 Autostop for Remote Instances
A cool feature we’ve been improving each week is Brev’s Autostop. This can help engineers working on remote development save big bucks on their cloud bill by stopping instances automatically after 3 hours of inactivity!
🚨 Launch Week Day 4🚨
🌿 𝔹𝕣𝕖𝕧 𝔹𝕚𝕠𝕞𝕖 🌿
Get free GPUs for creating templates!
We make templates for all the new models like Falcon, Dreambooth, etc. If you make one, every 5x it's used, you'll get a free hour of GPU!💰
🚀 ”brev open [workspace name or id]” to automatically open vscode directly to your project folder
🔒 Auth has been secured so you no longer need to login daily