Ali Asaria
@aliasaria
Followers
3,804
Following
1,894
Media
46
Statuses
393
Working on @transformerlab ex CEO of Tulip, . Engineer at heart.
Toronto, Ontario
Joined December 2007
Don't wanna be here?
Send us removal request.
Explore trending content on Musk Viewer
Celtics
• 335019 Tweets
#GranHermano
• 314040 Tweets
#WWERaw
• 160218 Tweets
Juliana
• 151607 Tweets
Tatum
• 146210 Tweets
Ceci Flores
• 99306 Tweets
Dallas
• 80354 Tweets
Mavs
• 72026 Tweets
Kobe
• 60589 Tweets
Kyrie
• 58023 Tweets
Hulk
• 56745 Tweets
Pritchard
• 53919 Tweets
#NBAFinals
• 49946 Tweets
ニンダイ
• 49822 Tweets
Del Moro
• 43152 Tweets
Seth
• 41926 Tweets
NBA Champions
• 40190 Tweets
Finals MVP
• 31085 Tweets
土砂降り
• 29956 Tweets
マイナンバーカード
• 27588 Tweets
Uncle Howdy
• 25417 Tweets
#DifferentHere
• 23050 Tweets
Estevão
• 21715 Tweets
Doncic
• 21010 Tweets
Al Horford
• 18590 Tweets
Jrue
• 18539 Tweets
Banner 18
• 18517 Tweets
携帯契約の本人確認
• 16957 Tweets
運転免許証
• 15399 Tweets
#ファミマから大ボーーーーナス
• 13940 Tweets
券面確認
• 13202 Tweets
FMVP
• 12051 Tweets
Bray
• 11751 Tweets
Derrick White
• 11463 Tweets
Entrepreneurs (like myself) should pay more tax on their gains. It's not fair that the average worker pays a higher tax rate than folks who stand to make tens of millions on a company exit.
Together, we can build a fair AND entrepreneurial society.
24
18
119
A few days ago at a
@betakit
conference I spoke on the on the 🇨🇦 government's proposed changes to the capital gains inclusion rate.
Afterwards, a # of people approached me asking for more clarification on my views. Here we go...
13
38
114
I never thought this would happen: I am using my Linux box with a GPU less and less these days.
Inference and training speed using Apple MLX is blazing on my new Mac M3.
This is a video of Mistral 7b (4bit) getting 30-40 tok/s
@transformerlab
@awnihannun
#mlx
5
8
90
Two days until I announce the new software project I've been working on. Put a lot of my heart into this.
Will launch at Toronto New Demos by
@internetvin
. Thursday night, Jan 11, 7pm.
The event is sold out. Can't wait to share more after the event.
8
2
67
Transformer Lab is what I’ve been working on.
The hope behind Transformer Lab is to make it possible for any person with access to a GPU to do things previously restricted to a small group of Machine Learning specialists.
All Opensource.
8
5
57
LLM Inference demo using
@transformerlab
on my Mac M1 16GB using three different models (1.1B, 2.7B, 7B 4-bit quantized). It's finally possible to do real LLM work on a Mac!
Training coming soon... :)
cc
@awnihannun
4
7
54
Instead of an office space I got a membership at the Royal Ontario Museum. This is my view while I code today.
6
0
48
There is a weird thing these days for me and my Muslim and Arab friends.
Every day we wake up to photos of dead children on our phones. And every night we scroll through more. Literally.
And then we go on with our days because that's what we're supposed to do.
5
4
47
4
0
44
Huge update to
@transformerlab
-- we now support Retrieval-Augmented Generation (RAG) out of the box. It works on MLX, NVIDIA, and CPU backends.
With RAG, you can search your documents / PDFs, tech using LLMs, on your local machine. Give it a try and let me know what you think!
3
3
36
Are you angry about the proposed changes to Canada's tax laws?
It looks like someone is finally doing something about it!
The crowdfunding site below can hopefully make a difference.
10
9
35
I used to run a grocery + beauty + health retailer for part of my career. Here are some of the secrets of pricing I wasn't supposed to share:
1
2
33
A key to being an entrepreneur is to never be satisfied with what you have. Be hungry.
But a key to being *happy* is to be content what you already have. Be accepting.
How do you balance the two?
15
3
29
Today Tulip is announcing the biggest product we’ve ever made.
For the last 4 yrs we’ve been secretly building the world’s best enterprise POS. Big, global retailers will finally have the choice to move to a POS that was built on modern architecture, for the tomorrow’s customer.
Introducing Tulip’s new enterprise Point of Sale (POS) solution! Unlike archaic, monolithic POS systems, Tulip Checkout is created for how modern omnichannel retail operates.
#retail
#retailtech
#retailtransformation
#digitaltransformation
#mPOS
0
1
8
2
4
26
15 years ago, it felt like a founder could pick any industry and build a "software is eating the world" high growth startup in that space.
Now, I find a list of 100 software startups in niches of niches I didn't even know existed.
2
0
22
Even if your answer is wrong, it gets part marks.
I went from the bottom of the class to the top but it also make me lose faith in school.
This lesson ("rewriting answers gets better marks than understanding the material") makes me think about modern LLM benchmarks.
2
0
21
@awnihannun
@transformerlab
This is updated Transformer Lab. The UI got a bit of a refresh today -- will be in next release in a day or two.
3
0
21
@internetvin
I talk to startups about the difference between anime hero stories vs US cartoon hero stories. The big thing I notice is that for US heroes, their skills magically appear. While for anime heroes, they have raw talent but then have to grind for years to master their skill.
1
1
19
Successfully finetuned an LLM on a Macbook Pro, without using a separate GPU. This is a big milestone. Will release the code soon.
0
0
20
If I was the prime minister of Canada, the first unimportant rule I would enforce is the standardization of food spiciness reporting.
It's 2024, none of us should have to say "When you say 'hot' how hot is hot?
4
0
19
There is a term "builders" that we use in the startup community. You know them when you meet them.
True builders care about "monetization" only in as much as it helps them build more.
The world is a more fun place when builders are unleashed.
1
0
16
🇨🇦🤖🚀
Vantage points: Serial entrepreneur
@aliasaria
on the
#CDNtech
"bubble" and how to beat the US
#BKtownhall
1
2
11
1
0
16
Went to get my covid 💉 today. In the tiny room, there was so much 🇨🇦 innovation by people I know.
- The fridge that held the vaccines was by
@jimestill
Danby
- I booked the vaccine appointment on
@medmehealth
by Purya Sarmadi
- The posters showed
@JoinLeague
by
@mserbinis
1
1
16
Exiting news at
@transformerlab
-- we had our first intern join the team and she's great!
Transformer Lab v0.3.7 is out!
Mistral v0.3 and Aya 23 are now available in the gallery. And a Linux AppImage build is available. Check out our release notes on Github to see all the updates.
0
1
2
0
0
16
We are finally ready to show Transformer Lab to the world. I would love if you could try it out. We have big plans!
Today is our first official public release!
Want to dig deeper on LLMs but don't know how to get started? Transformer Lab was created for this.
Fully supports MLX which means you can run models efficiently on Apple macOS... & can also train them in the GUI. Link in profile.
0
0
5
2
1
15
Six years ago, the president of the US instituted a "Muslim Ban" after calling for "a total and complete shutdown of Muslims entering the United States until our country's representatives can figure out what the h*** is going on."
I will never forget.
0
0
14
Data on the average Canadian's perception of the proposed federal budget
10
0
13
The FastChat LLM serving library from
@lmsysorg
(31k Github stars) now natively supports Apple MLX for serving. Docs here . Proud to work on this. FastChat is a good way to run models because it handles prompt formatting and OpenAI SDK.
1
0
13
A big lesson I learned in software development happened during an emergency in our ecommerce shipping department. There was a spike in volume and developers had to redesign everything, while also helping pick-pack orders. It was all hands on deck.
3
1
12
I opensourced BrickCrusher2020 today. It is a BrickBreaker clone for
@playdate
by me, the original creator of BrickBreaker for BlackBerry.
I learned the Lua programming language to make it. Lots of fun. Enjoy!
0
0
12
To this day, that warehouse still plays the super mario coin sound every time you grab the right item. 🪙🎶
I try to remind myself of those days every time I work on sofware. We make better things if we find time to stand in the shoes of the real end user as we build.
0
0
12
One of the greatest gifts in life is being in love with your work.
My first job was teaching kids science. We'd spend summer days catching crawfish and counting their legs. I was exhausted every day. I would have done the work for free.
1
0
12
A long time ago, a friend of mine was in charge of setting up the first datacenter in India for one of the big 3 cloud providers. One day the mob came and left a note: we will burn down your building unless you pay us protection $.
2
8
11
First time founder: "This will be so fun!"
Second time founder: "This will be a huge success if I avoid the two things that I didn't plan for last time, retain more equity, and retain controll."
Third time founder: "This will be painful no matter what I do"
0
0
11
They were a public company. They ignored the message. But more letters and messages came until it was impossible to ignore.
How does a publicly traded, well respected US company pay off the mob and file their taxes?
Answer: you ask for fake taxi receipts.
The mob agreed.
0
2
10
Every day I log into my email and delete all the msgs from B2B marketers who use tools to find my address and msg me.
Every day the BDR team at my co uses the *same* tools to find client addresses to do the same to others who also delete them.
B2B marketing needs to change.
3
1
10
You want to be productive with LLMs on an Apple Mac. Why buy a second computer with a GPU when you already have a Mac? Ok so first... what model of a Mac should you buy?
1
0
8
After starting my first company, I got to meet and learn about so many stories of people with enormous wealth. The stories were an adjustment for me. For example...
1
0
9
How fast can you run good open source LLMs on consumer hardware? We've know that
@groq
can output at 450+ tok/s, and other commercial API providers are usually around 100.
Can we get to 200+ on our local computer?
Let's do a test... ( using
@transformerlab
)
2
3
8
Training LLM models (3b params) on my home RTX3090 is costing $0.19 per hour in my energy bill. 😯
1
1
9
"Everyone (including software engineers, small startups, academia, indie developers,...) should train, optimize and run their own models. Time for AI builders, not AI users!"
@ClementDelangue
YES!
Great first three days in the Bay Area, so much energy here! My current messages to the ecosystem:
- Everyone (including software engineers, small startups, academia, indie developers,...) should train, optimize and run their own models. Time for AI builders, not AI users!
-…
14
26
280
1
0
9
Technology is changing the retail world so fast.
Yet we still have major retailers running point of sale systems that look like 👇. How can this be?!
The era of the ugly POS is over. At
@TulipRetail
we won't rest until they're all in the 🗑️. Big announcements coming this year!
0
1
9
I was on the
#Supermanagers
podcast, where
@aydin
and I chatted about team alignment and trusting by default.
Let me know what you think!
1
1
7
One time a friend of mine worked at Google and had to present to Larry Page. Larry had just come back from vacation in Africa and btw complained that Google maps weren’t great there. The maps person said oh that’s because the satellite data has poor coverage in Africa. 1/2
1
0
8
We're about to launch a massive update to
@transformerlab
that will make it vastly easier for almost anyone with a computer to do deep work with Large Language Models. Stay tuned -- will use the next few days to test everything.
0
1
8
@bentlegen
@funkaoshi
@hyfen
@davidcrow
@jevon
@jaygoldman
@elucid_ont
@jaco_says
I miss those days! There was an energy that is missing now.
0
0
6
It's impossible to estimate delivery dates when creating.
Over the last three days, I was working against a list of three coding tasks that were supposed to take the same amount of work. One of them took 2.5 days, the other two took 20 minutes combined.
0
0
8
Interested in playing with LLMs and running experiments on your own? We built Transformer Lab as an open source platform for folks who want to learn about how LLMs work and do advanced things with them including training, RAG, etc.
Learn more here:
1
0
7
Happy tomorrow birthday
@photomatt
. Thank you for inspiring me to learn the tech stack I used for work, bc of yr opensource. Thank you 4 investing in both my startups. But mostly 🙏🏽 for inspiring me thru stories I 👂🏽 abt u from others — you radiate kindness in every interaction.
0
0
7
But inherent (or often explicit) in their arguments is that the western world is inherently superior, while the rest of the world is evil and untrustworthy.
See for example
Many of us don't see the world this simply.
1
0
7
Some of the best product leadership advice I've ever received came from a single conversation with
@fahdananta
. Highly recommend.
A handful of spots left for this AMA session
I hope we get to talk about hip hop and basketball
1
0
9
1
0
7
So in the middle of the 30 min meeting Larry picked up his phone and purchased a multimillion dollar satellite to fly over Africa while everyone sat and listened. Then he put down his phone and told them to continue the presentation.
1
0
6
1/ On Pi Day I wanted to share my (true) 𝝅 story at
@UWaterloo
. When I was 18 years old and in highschool, I worked on the university campus. Somehow, I developed an unhealthy obsession with Pi, which eventually led me to commit academic fraud. Here is the story.
2
0
7
@awnihannun
I am seeing 43.5 tok/s on M3 Max 96GB (and great quality responses) on MLX 4-bit Llama 3 8B
2
0
6
Somehow we've all just accepted that a 2GB blob of numbers can speak. Sometimes I have to step back just to absorb how insane this is.
I understand how it works. But I don't. This is magic.
0
1
6
"To keep the blood cell supply [in the human body] steady, about a billion meters of DNA must be copied every minute."
I never thought about the scale of DNA replication in the human body like that. Wow.
0
0
5
Ever considered starting your own startup? This is a great article about what it takes, from one of the best in Canada
@byosko
2
1
5
The server at the new Vietnamese coffee place said there is 20% more caffeine but it feels like 300%.
😵💫
Anyway it was good.
1
0
6
@mrkrishsatya
@robjama
Me neither and all I said was “there’s some good people on both sides.”
1
0
6
One of the core arguments and fears you can see in western AI warners is that AGI is on the horizon and it will be the most powerful weapon ever. They say "it must not get in the hands of the CCP".
3
0
6
Reference to article that explains unique thing about M1 vs all other Apple Silicon, regarding bfloat16:
0
0
2
The top students wouldn't try to understand the materials like me, a noob. They would just re-write the past exam questions over and over (and over). Their advice: when you get to the exam, first rewrite the answer to the most similar question you've seen before. Then tweak.
1
0
6
Tap to Pay on iPhone now available in Tulip!
A big announcement from
@TulipRetail
on the path to unlocking the future of stores for the world's best retailers. Thanks to all the work from the Tulip R&D team to make this happen. & Thanks to
@Apple
and
@Adyen
for all the support.
Tulip is pleased to announce that their Point of Sale now supports Tap to Pay on iPhone. This was made possible due to Tulip’s partnership with
@Adyen
, the global financial technology platform of choice for leading businesses.
Read the full release:
0
0
2
0
0
5
Absolutely gut-wrenching. As a parent, hearing these stories and seeing the pictures every day breaks us over and over.
We in Canada and across the globe could have done more to stop this horrendous tragedy. But instead, helplessly, we watch with tears in our eyes.
This is a gut-wrenching testimony from
@SeemaJilani
. I will never get the depth of dehumanization of Palestinians people possess that they can hear this these stories ¬ be moved to call for an end to this barbarity.
This the United States’ genocide as much as it is Israel’s.
4
174
343
0
0
5
👀 "M4 versions of the Mac desktops could support as much as 512GB Unified Memory"
2
0
5
@jrodgers
The 50% break on capital gains exists not for rewarding risk, but because the wealth class has lobbying power. This tax break is the biggest way wealthy people avoid paying taxes. And it's confusing enough that most non-wealthy people didn't understand how unfair it is.
2
0
5
@fahdananta
I like to embrace both sides of the inner self. There is a part of that is kind but also vulnerable. And there is a beast that will never be held back. The first side lets us connect with the other, the second part lets us tear down barriers. Too much either way is unhealthy.
0
0
5
4 our clients, this is *the* most important piece of tech that powers stores. & for the longest ⏲️, existing solutions have held back retailers more than anything.
Nothing is more important to Tulip than this new product. Everything we've done so far was to get us to this point.
0
0
5
We mistakenly idolize people and companies when everything is working out in their favor.
But the biggest test of greatness is how people behave and treat others when things are tough.
To every CEO that is cutting staff: now is your chance to prove what kind of leader you are.
0
1
5
I wrote an article about Shopify's Secret Master plan (and how it is affected by the recent divestments in logistics) on LinkedIn. Take a look here:
1
0
4
@trillhause_
@transformerlab
@awnihannun
Macbook Pro M3 Max chip with 14‑core CPU, 30‑core GPU and 16‑core Neural Engine
96GB unified memory
0
0
4
We can run using vLLM to get a bit faster. Here is Vicuna 7b running using vLLM.
Now we get 55 tok/s
1
0
4
@internetvin
Scarborough is special. And big. Let's do a food tour one time -- happy to show you some of my fav places.
1
0
4
Check out the speed of TinyLlama 1.1B running on an RTX3090 using vLLM as the inference engine.
We get 250 tok/s and it's 🚀🚀🚀
1
0
4
I dreamed of creating a tool that would let anyone do powerful things using python.
I spent 99% of my time reliably installing python.
2
0
4
Retweet, comment or thumbs up if you think Canada has the opportunity to be the leader in investing in the 99% as a path to being a global innovation leader.
1
0
4
When I was in uni, I got really bad exam marks.
So I joined a study group and they showed me how to boost my avg from 60% to 90%+ . But the secret was depressing.
1
1
4
We can get faster if we use a MUCH smaller model. Here we run TinyLlama (1.1b parameters) using Hugging Face Inference.
We get 78 tok/s and it looks pretty fast!
But we can *still* do better...
1
0
4