ishan
@0xishand
Followers
378
Following
1,001
Media
14
Statuses
404
enterprise ai @nvidia | prev. @brevdev (acq.) @agora_io (acq.), @columbia , @stlouisfed , @sparkcognition
not the metaverse
Joined April 2022
Don't wanna be here?
Send us removal request.
Explore trending content on Musk Viewer
Ricardo
• 336365 Tweets
Mariana
• 292978 Tweets
Arath
• 242453 Tweets
The Next Prince Q1
• 208007 Tweets
#LaCasaDeLosFamososMexico
• 182268 Tweets
Wendy
• 145530 Tweets
#RakshaBandhan
• 117458 Tweets
Pro Win Chang Club
• 93599 Tweets
#LCDFMX2
• 79053 Tweets
#रक्षाबंधन
• 78463 Tweets
Mayito
• 58617 Tweets
連休明け
• 50780 Tweets
滋賀学園
• 44254 Tweets
戸愚呂弟のサングラス
• 32814 Tweets
Sian
• 27653 Tweets
青森山田
• 23932 Tweets
असीम प्रेम
• 16116 Tweets
Carlyn
• 15323 Tweets
#Rakhi
• 14750 Tweets
PRABOWOkerja UNTUKrakyat
• 14107 Tweets
神村学園
• 14096 Tweets
Happy New Week
• 13554 Tweets
KOMPAKbersama BANGUNbangsa
• 13448 Tweets
京都国際
• 13372 Tweets
カミガタ
• 13237 Tweets
リーリエ
• 12708 Tweets
ハッピーセット
• 12430 Tweets
Patricio
• 11224 Tweets
Mildom
• 10849 Tweets
セブン&アイ
• 10454 Tweets
セブン買収
• 10434 Tweets
Pinned Tweet
Wanted to share some work I’ve been doing on Forky: A git-style approach to LLM conversations.
The problem: You’re deep in a chat and suddenly need to explore a tangent. You could open up a new chat and explore…but now you have 2 separate chats
How do you combine them? 🧵
7
5
34
Culmination of months of hard work by the Akash+Brev teams. Big shoutout to the man himself
@tmonty_12
for leading the charge.
Come get yourself some H100s :)
Today, is officially launching support for Akash GPUs.
Brev is on a mission to create the easiest way to develop with the leading open-source AI models and deploy on GPUs in the cloud. Purpose-built for developers who prioritize speed and simplicity.
17
105
483
1
6
44
No way the next 6 months get crazier right?
Exciting news...
🚨 has been acquired by NVIDIA! ❤️🤙
We started Brev with the goal of building the best damn developer experience possible.
We’ve been working closely with NVIDIA since August, and we couldn't be more excited to team up.
I'm
265
90
1K
1
1
35
Time really does fly when you're having fun...
After founding
@tamublockchain
last summer, today is my final day as president of the organization.
Here's a couple things I've learned throughout the process 🧵
4
5
24
Over the last few weeks I've worked through implementing
@karpathy
's micrograd in Golang. Building this in Go was a bit of a challenge due to the lack of operator overloading and easy to use graphviz libraries (thanks
@traviscline
).
Repo and small blog post in the 🧵
6
6
22
Come say hi :)
Akash is heading to NVIDIA GTC next week to connect with AI developers, as well as co-sponsoring a booth with
@brevdev
.
Connect with the team in person:
@_anil_murty_
,
@LeChenghiskhan
, and
@zacharyhorn
from the Overclock Labs team will be attending.
28
131
624
0
1
18
Jupyter Notebooks are a fantastic way to test new models and frameworks, but connecting them to powerful GPUs can be challenging.
With
@agora_io
, you can easily set up a GPU-powered Lab environment and start creating almost instantly.
Let us know if you have any feedback :)
Introducing Agora, a cheap & easy platform for AI development! Starting off with supercharged notebooks, we are revolutionizing building in AI/ML...
Let's dive in 👇
2
13
48
0
0
12
🏛️ -> 🤙
Today Brev is acquiring
@agora_io
A couple months ago, a competitor signed up for Brev. I reached out to see if there are ways we could work together
We signed mNDAs, saw identical roadmaps, started working together, had a clear culture fit.
They hit the next flight to SF 🤙
24
10
206
0
2
13
Fully agree with
@tommyeastman21
here. Greg and the rest of the core team have been nothing but supportive as we’ve built on
@akashnet_
.
Quite on the contrary— I would strongly urge builders to start building on
@akashnet_
and developers in need of compute to check out GPU offerings on or
have had nothing but the best experiences with Greg and team.
@FoundryServices
1
7
51
0
3
12
📈
We are teaming up with NVIDIA to throw a NIM hackathon
@NVIDIAAI
NIMs are an easy way to run AI models anywhere
Free GPUs, beer, and early access to NVIDIA's Llama3 NIM a week before it's released 🤙
Come learn from the brightest and hack on the most cutting edge AI software
9
23
149
0
2
9
Reformatted
@maximelabonne
's guanaco-llama2-1k dataset to match Llama3's prompt formatting here in case anyone wants to use it for a quick ft
2
6
10
Spent yesterday messing around with the new DBRX model. It's a pretty hefty but if you want to run inference, check out this notebook that you can quickly launch on a cluster of 4xA100s. Still figuring out the best ChatML prompt format for it....
0
3
9
I'm extremely excited to see where this club goes under the leadership of
@0xTheSlob
,
@ani_margam
,
@VarunS24
,
@KateAldama7
,
@JFrostHTX
and the rest of the executive team.
Thanks for everything
@tamublockchain
✌️
2
3
8
1. Zk + digital identity
2. Why Sinner beating Djokovic was such a pivotal match for the future of tennis
3. Decompute as an alternative for hyperscalers
4. Pros/cons of an mscs degree
5. Optimal Mario kart builds
What are five topics you can talk about for 30 minutes with zero prep
1. information content of text embeddings
2. architectural mysteries of transformer language models
3. the meta-research process (how & why to pick problems)
4. whether the NBA is rigged
5. shrimp suffering
13
0
83
1
0
6
Biweekly syncs with Anand always helped us ruthlessly prioritize what was important. It played a huge part in how we approached building Agora and Brev.
I highly recommend anyone building in the deAI space to shoot him a DM. You can’t go wrong
I am so proud of this bunch - they manifested this outcome. Proud to say that "I knew them when..." :)
4
2
51
1
2
7
Name a better way to spend a Wednesday than getting your hands dirty with some unreleased NVIDIA software.
The NVIDIA NIM x Brev hack-a-meet-up is Wednesday from 5-10pm
The event is hosted at Solaris AI in the heart of Cerebral Valley, SF
we're already at twice as many attendees as expected, should be a killer time! 🤙🤙🤙
2
4
18
0
1
7
The price of fine-tuning an LLM should be the price of a beer. Thanks
@semafor
for showcasing some of the work we’ve been doing over the last couple months. More to come soon 👀
Blockchain and crypto have been criticized for years as a “solution in search of a problem.”
Now some industry entrepreneurs may have found the problem: A shortage of compute power needed to train AI models.
1
3
16
1
0
7
@ekzhang1
Check out
@marimo_io
. They’ve got a lot of cool notebook features. Also I believe Jupyter has a plug in for collaborative editing that works relatively well
0
0
3
Able to get insane inference stats using tensorrt_llm + triton inference server.
The most painful part of this was the config files. Triton’s auto-gen is decent but I’d love to be able to throw a model in -> get the ensemble configs fully generated…
Anyone know how?
0
0
5
Excited to announce that
@tamublockchain
’s first dev project is feature complete! My team and I will be presenting it on May 1 at the inaugural Texas A&M Blockchain Day.
Whitepaper 🔜👀
2
0
6
abstract away the crypto, comfortable UI
tools like these are what we need to bring web2 companies/devs to web3
kudos to
@agora_io
and
@akashnet_
for this product
1
8
22
0
1
5
Paper deadlines are coming up! If you need a no-hassle deep learning environment powered by the best GPUs in the world, reach out 🙂
If you're an AI developer tinkering with LLMs, you know how difficult it is to get quick and cheap access to a dev environment with high powered GPUs.
This is what we've been building
@agora_io
. See how easy it is to get access to 2 A100 GPUs😎!
3
5
13
0
1
5
Come say hi :)
come join us!
An exciting lineup of folks this week -- speakers, come find
@sarahchieng
and
@0xishand
upstairs for an interview :)
0
6
41
1
0
5
I want thank
@athreesh
,
@sreethereum
,
@iamjakelevy
and the rest of
@txblockchain
for all their help this year. Working with their team was always a pleasure and I look forward to future collaborations!
1
1
5
Curious to see Amazon’s onboarding process and how they’ll deal with payments, user wallets, etc
Amazon is building an NFT marketplace:
Is this the moment we've all been waiting for? 🧵👇
(1/16)
388
516
2K
0
0
5
If you’d like to access nvidia a100s without paying a 7 figure cloud bill, please reach out :)
We've built a Fine-tuning -> Inference Pipeline on
@akashnet_
!
Take an open-source model like LLaMa-2, fine-tune it on your dataset, and deploy it for inference (all powered by decentralized compute)🚀
Watch
@tmonty_12
present a demo👇
1
10
35
1
3
5
So let me get this straight. Instead of creating a proposal and requesting additional funding, you decided to jump on X and complain that you’re not being pulled long for the ride now that Akash is receiving mainstream attention
@gregosuri
@Petar26995863
@SpheronFDN
@akashnet_
a) Liquidity support. If a 5 AKT deposit per deployment was too high, what prevented you from submitting an on-chain proposal to lower the deposit amount? - We did raise the concern for this & we also opened the discussion; then you asked to raise a proposal to lower it, which
1
0
1
0
0
4
@AgustinLebron3
At the ATP level, dual forehands wouldn’t work due to how fast the game is played. Take return of serve - if you had 2 forehands, your non-dominant side would be a large weakness as you dont have time to move your racquet up.
1
0
4
What he said 🫡
Wrote this piece with my guys
@0xishand
@tmonty_12
! We cover decentralized GPU computing and how it can revolutionize AI/ML - from its potential to current limitations.
If you have a couple min, take a read and lmk what you think! Would love to hear your bear/bull case
4
1
14
0
1
4
5. Rely on your team
During the fall, I kept overloading myself with tasks thinking that as the founder I had to set an example and do it all. But over time, I realized that delegating helped me grow as a leader and helped the executive team grow as professionals!
1
1
3
Huge milestone here for deCompute
A large cluster of NVIDIA A100s from
@FoundryServices
is live now, and available for anyone to deploy.
The
#AISupercloud
is gaining momentum.
→
6
46
239
0
0
4
Great to see teams working on compute as a commodity!
Introducing Prime Intellect Compute: The Compute Exchange
We're excited to announce that our platform for aggregating and orchestrating global GPU resources is now fully public.
Our mission is to make instant compute and intelligence accessible to all.
6
27
156
0
2
4
🫡
Asked for a feature on Sat morning, feature delivered Sat night.
No rest for the best.
Cheers
@agora_io
, constantly adding value to
@akashnet_
3
4
18
0
1
4
Something that we explored early on at
@agora_io
was a way to allocate GPU resources based on provider payouts. How are you planning to approach this or will it be a manual process to start?
Each network is going to have their own liquidity mining program to compete for GPUs, meaning the opportunity for providing resources to these networks early on I think is going to be pretty lucrative.
Going to ape a high-end gpu before everyone else in crypto starts doing so
2
0
14
2
0
4
If you didn’t switch to ollama today during the outage you’re probably ngmi
0
1
4
The lectures worked for students who were deeply passionate but did not captivate someone with a high-level interest in the space.
This past semester, we made a pivot and tested out our first developer cohort.
1
1
3
Let me know what other HF tutorial notebooks you want to see as 1-click deploys on Brev!
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:
🤙🤙🤙
0
6
31
0
1
3
@petergyang
@ollama
I would say so. I like using this UI . Have gotten fantastic inference speeds on my M2
0
0
3
3. Don't be afraid to adapt
To account for the knowledge disparity, we started off with a lecture based approach (shout out
@_eshabora
for letting us use those amazing slide decks).
This worked for a bit until we got feedback that meetings felt like another class.
1
1
3
2. Building a community
Student blockchain clubs are not just about the technology or projects - they're about the people. Throughout the year, we tried to hold as many social events as possible to engage the entire Texas A&M community.
1
1
3
@casper_hansen_
@migtissera
@andrew_n_carr
@EMostaque
Agreed - it all comes down to a lack of tooling. Take
@akashnet_
for ex: there’s currently clusters of A100s sitting idle due to crypto/wallet friction. My friends and I built
@agora_io
to allow engineers access to these gpus without any blockchain knowledge needed
2
0
3
Absolutely legend
I really wanna take a moment to thank the man upstairs... Alan Butler
By far, the best part of the founder journey is getting really close to special people along the way. I met Alan as an investor and advisor, but he quickly became a mentor and close friend.
He taught me a
6
3
116
0
0
3
Common StarkNet W for calling this Madara
🌟 Announcing Madara, a cutting-edge experiment: a Rust🦀-powered Starknet 🐺 sequencer using Substrate! 🚀 Early benchmarks reveal a stunning 76 TPS for ERC20 transfers! 📊 Check out the demo video & explore Madara's potential!
34
98
417
0
0
3
Peak Theim was a monster. Looking forward to what’s next 🫡
OFFICIAL: Dominic Thiem announces his retirement at the end of the season.
Very emotional video.
"I'm gonna finish my career at the end of this season (...) My wrist is not at the way it should be and how I want it. I was thinking about this decision for a very long time"
165
469
4K
1
0
3
4. Mistakes are inevitable
Leading a student organization is not always smooth sailing. We spent our first semester being an unofficial club, recruiting completely through word of mouth, and not being able to book rooms...all because I submitted the application 2 hours late.
1
1
3
@solanobahn
@cloudmosio
Check out
@agora_io
- you can deploy on Akash without worrying about crypto. We have an internal routing system that matches you with high quality providers!
0
0
3
Staying motivated and driven throughout each week is hard.
Here's a methodology to make sure your dicipline stays bulletproof
(0/14)
1
1
3
We integrating this
@athreesh
@tmonty_12
?
Announcing: BlueSea Frontier Compute Cluster
The Biden admin's AI Executive Order and the EU’s AI Act aim to centralize control under guise of safety
The solution is at sea.
BSFCC's are not just barge based compute platforms, but sovereign nation states
405
625
3K
0
0
3
@casper_hansen_
@migtissera
@andrew_n_carr
@EMostaque
@akashnet_
@agora_io
There’s teams like
@BanyanComputer
working to make decentralized storage a viable alternative for enterprises who don’t want to deal with the blockchain itself. Check them out!
0
0
2
Insane how fast this space moves. These proving stats were unheard of just 1 year ago
Very interesting development on AI model output verifiability: zkLLM
zkLLM enables the verification of outputs
- in < 15 minutes
- up to 13B parameters models
- achieving a proof size of less than 200kb
22
18
137
1
0
2
@AgustinLebron3
Going from a semi western fh ground -> backhand is not as much of a change as you would think bc of non dominant hand support.
The act of fully switching hands takes longer than a grip switch which means one side can be rushed during a point -> a weaker shot
1
0
2
@ai
@brevdev
@agora_io
has built a colab style sandbox environment on top of
@akashnet_
. Stay tuned for our bittensor integration 👀
1
0
2
0
0
2
Whether it was a Mario Kart tournament or a study hall, it was amazing seeing people from different fields come together to learn and collaborate. Texas A&M has an incredibly diverse student body and I was incredibly lucky to meet so many talented people through the club.
1
1
2
0
0
2
There were times where I miscommunicated with execs, failed to meet personal goals, and didn't send emails on time.
But it's all a part of the process. It's important to reflect on these mistakes and learn from them, but it's equally as important to keep pushing forward.
1
1
2
Over time, this persistence paid off and we started seeing an uptick in attendance at our events and socials. It was a valuable lesson and I'm glad we learned it early.
1
1
2
Spot on here
@tmonty_12
. I’ve ran into some similar realizations as well this semester. Building a community is super important for the long run!
After the first couple of meetings for
@cuseblockchain
this semester, I have had some big realizations about running a blockchain club at SU👇
1
0
8
0
0
2
@camdoody
Anish already said it but can confirm
@tnadevedant
and
@armelwtalla
are stone cold killers
1
0
2
@casper_hansen_
@migtissera
@andrew_n_carr
@EMostaque
@akashnet_
@agora_io
Valid concerns.
1. Availability for 8x clusters aren’t the best right now but will improve as the provider incentive program goes live. These providers are T2+
data centers or better
2. The providers that offer A100 clusters rn are physically connected
1
0
2
@Yuchenj_UW
Congrats
@Yuchenj_UW
,
@zjasper666
and the rest of the Hyperbolic team! Looking forward to seeing the progress!
1
0
2
The results were astounding! We had a huge influx of people apply looking to gain real world experience.
Moving forward, the club will transition to a project-based approach where students can get their hands dirty and dive deep into the various blockchain rabbit holes.
1
1
2
@casper_hansen_
@migtissera
@andrew_n_carr
@EMostaque
@akashnet_
@agora_io
Provider quality definitely needs to be improved. We currently route you to providers that have the best combination of price/speed and that will only get better as more providers join akash
0
0
2
@sreethereum
@PatniShubh
@JulianSaks
@txblockchain
@ShantikiranC
Yep exactly. Agora aggregates GPUs across different providers. Atm we do not have our own GPUs
0
0
1
350????
We are thrilled to be a launch partner for Meta Llama 3.
Experience Llama 3 now at up to 350 tokens per second for Llama 3 8B and up to 150 tokens per second for Llama 3 70B, running in full FP16 precision on the Together API! 🤯
28
57
394
0
0
2
The most intuitive metric is Accuracy. This is a good metric for balanced data but not when a dataset is imbalanced (as most real world data tends to be).
(2/X)
1
1
2
These guys just don’t miss
We’ve been cooking with recently, big launch coming soon 👀
0
0
8
0
0
2
@Old_Samster
Yep - Agora is cloud agnostic and supports decompute as well as the traditional cloud providers. We’re bullish on the value decompute provides and built our stack to natively support these networks as they emerge/scale.
0
0
2
We’re SO back
@traviscline
WorldSim is back, and better than ever!
Check it out @
Join our Discord to join our community:
Tons of bugs fixed, credits and payments systems working - here are all the new features that are active in the latest version:
16
49
279
0
0
2
Instead, it was a sign that we needed to improve our marketing and outreach!
We started experimenting with different mediums and rebuilt our entire brand kit. We created feedback forms for students and made sure to follow up with any questions and concerns.
1
1
2
@Old_Samster
That’s why we’re building this MLOps platform at
@agora_io
. MLEs shouldn’t have to compromise on their existing tools/stack to access the raw underlying compute that these networks provide 🫡
0
0
2
Big Go guy huh
@traviscline
?
LLMs, langchain and LangchainGo. It was great to learn properly about them, thanks
@traviscline
! 😊
#golang
#gopherconEU
@gopherconeu
0
6
37
1
0
2
@migtissera
@casper_hansen_
@andrew_n_carr
@EMostaque
@akashnet_
@agora_io
Can you open your DMs
@migtissera
- would love to discuss more.
1
0
2
Can we please settle this, is it pronounced C-L-I or “klee”?
6
0
3
0
0
1
Precision and Recall have an inverse relationship (hint: how would I decrease FP and what would that do to FN) and it's important to report both or use a metric that combines both...enter the F1 score
(9/X)
1
0
1