Abubakar Abid
@abidlabs
Followers
13K
Following
26K
Media
865
Statuses
15K
Founder @Gradio, acq'd by @HuggingFace. Hind Rajab was 5 yrs old when she — and 18,500 other children — were killed by Israeli forces. PLEASE don't be silent ❤️
Stanford, CA
Joined May 2013
If you think that somehow the presence of some Hamas members justifies the killing of children in Rafah. Ask yourself: . If 2 Hamas members were hiding in a daycare in Israel or in the US, would you support bombing the daycare? . Why are Palestinian lives worth less?.
3K
6K
26K
I've worked with many aspiring PhD students (esp. international) who ask what a statement of purpose should look like. I'm sharing the SoP that helped me get into PhD programs at MIT, Stanford, and Georgia Tech -- hope it helps anyone applying this cycle!.
82
1K
5K
I'm shocked how hard it is to generate text about Muslims from GPT-3 that has nothing to do with violence. or being killed.
152
2K
5K
In 2019, I started remotely mentoring an undergrad CS student in Turkey (friend of friend) on an ML research project . He got admitted to a CS PhD program at Georgia Tech, and is now a student there. I'm planning on mentoring more students like him. Would anyone like to join me?.
166
123
1K
🥳 LLMs aren't limited to text anymore! . Introducing 𝚐𝚛𝚊𝚍𝚒𝚘-𝚝𝚘𝚘𝚕𝚜, a new open-source library that lets you combine an LLM agent with any Gradio app (for image generation, document analysis, etc.) . 💫 GitHub Repo:
32
198
937
Many computer vision classes teach theory and algorithms 🙂 . BUT they don't cover how to build an ML application end-to-end ☹️. I've put together a practical tutorial on how to build an image classifier, from datasets to deployment!. Check it out:
13
212
925
Trying out the new 𝚝𝚛𝚊𝚗𝚜𝚏𝚘𝚛𝚖𝚎𝚛𝚜.𝚝𝚘𝚘𝚕𝚜 library. Here's Hello World of our times (summarize and chat with a PDF) in 6 lines of code including import statements 🤯. Best part is it's all open source, from start to finish 🤗
19
143
810
Generating human-like text seems to require ~176B parameters. But generating human-like art requires 1/200 of the parameters (890M). Why?.
193
78
800
Embraced by Hugging Face: the Inside Story of Our Startup’s Acquisition. In late 2021, our team of five engineers, scattered around the globe, signed the papers to shut down our startup, Gradio. For many founders, this would have been a moment of sadness or even bitter
49
122
726
Thinking about teaching a class on computer vision. Most machine learning courses dive into algos, but I've been thinking instead of covering the "stack" for modern ML:. - Preparing the right datasets.- Finetuning models.- Deploying demos/apps. Would anyone find this useful?.
55
38
715
The way we teach ML doesn't consider real-world reliability. We teach:.1. pick static dataset (MNIST).2. split into train/test.3. train until high test acc.4. move on. Instead of moving on, we should:.5. deploy the model.6. get users to break models.7. adapt dataset to fix issues.
25
95
633
1/2 Why do so many machine learning models from academia & industry never get deployed as real applications?. One reason is all of the different libraries and languages that are need to deploy models:
12
108
589
A recent article in the MIT Technology Review reported that 100s of machine learning models had been built to help doctors diagnose COVID-19, BUT . "none of them helped". This is a serious failure of our field, and we need to break down the reasons why 🧵.
13
167
519
This is INSANE. I'm speaking in a mix of Urdu and English. Without knowing that, Whisper + Bloom + TTS responds back coherently 🤯. And you can try it right now (credit: @yvrjsharma) .
6
76
499
🧑🍳 Introducing the 𝑮𝑷𝑻 𝑩𝒂𝒌𝒆𝒓. A completely open-source implementation of OpenAI's custom GPTs on 🤗 Spaces. You own the weights, you own the code, so you can deploy your GPT on ANY platform at any time 🔥. Powered by the remarkably good Zephyr-β 7B 🪁
14
111
475
Ever wonder what the best open-source LLM is? . Introducing the "Open LLM Leaderboard": a place to see all open-source LLMs on 🤗 evaluated on 4 separate benchmarks! . You can even submit your own LLMs to get them evaluated & added to the leaderboard 💻.
17
107
437
We're building a new Decentralized Research team at @HuggingFace, to work closely with the ML community at large. When I did my PhD, most folks just worked within their labs. Now ML research happens everywhere, from Discord servers to huge cross-institutional collaborations.
10
50
426
How to deploy a machine learning model. Old way: .- read a bunch of blogs on serializing & dockerizing models.- write a Flask app to serve.- google how to create a simple front end.- get an EC2 instance to host everything. New way: <1 line of code with @GradioML python library
10
74
386
Slides for my @CVPR 2022 talk: . "Papers and Code Aren't Enough: Why Demos are Critical to ML Research and How to Build Them". Thank you to @humphrey_shi and #CVPR2022 for organizing!
3
54
363
Hugging Face just raised a 235M / 4.5B series D!. Thinking about writing a personal blog post about the last 2 years — how exactly we got acquired, how it almost didn’t happen, life afterwards. Would anyone be interested in reading about that?
41
12
353
This is incredible!! One of the coolest ways to use Stable Diffusion / Dall-E is now open-source🤩🤩.
I turned stablediffusion-infinity (#outpainting with #stablediffusion on an infinite canvas) into a web app using @pyscript_dev and @Gradio: PyScript for the frontend and Gradio for the backend. Source code: Also available on Colab
2
61
348
Props to the authors of "Transformers Can Do Bayesian Inference" for releasing a demo of their model along with the code!. Great for accessibility and reproducibility in machine learning 👏. Paper: Demo:
0
83
323
"How to Read and Write a Paper". The brilliant @mohsaied giving a talk to all of our machine learning fellows @fatimafellowshp that I wish I had heard 5 years ago
8
44
321
𝚐𝚛𝚊𝚍𝚒𝚘.𝙸𝚗𝚝𝚎𝚛𝚏𝚊𝚌𝚎.𝚏𝚛𝚘𝚖_𝚙𝚒𝚙𝚎𝚕𝚒𝚗𝚎(. ). A new way to demo your machine learning models in 1 line of code, specifically designed for🤗 transformers!. An example:
2
53
321
Cool tutorial by @_bhaveshbhatt on how to build your own ChatGPT using OpenAI's GPT-3.5 (so you don't have to wait in line anymore!).
7
60
326
Hugging Face Spaces is becoming the "app store" for machine learning apps. Here are some of my favorite @Gradio apps on .@huggingface spaces from this past week:.
1
39
322
Machine learning creators are building hundreds of apps on the "ML app store": Here are 5 of the coolest machine learning demos that you can try in your browser 🪄.
3
55
312
I’m excited to announce the start of the Fatima Fellowship program this March! . If you’re a student anywhere in the world interested in academia and looking to be mentored by amazing researchers on a project to strengthen your PhD application, sign up:
15
155
303
Try out the Blazing Fast LCM Lora SD 1.5 in your browser. It'll generate images as fast as you can type .
20
60
306
Hugging Face Spaces is the "app store" for machine learning apps. Here are some of my favorite @Gradio apps on @huggingface spaces from this past week:.
2
50
292
When I search for "machine learning interview questions", I almost completely find stuff like this, none of which really test for ability to do ML:. What's a good source of modern ML interview questions that you would use if you were hiring an engineer for applied ML today?
8
46
288
If you teach a machine learning class with final projects, here is a cool way to run it:. 1. Set up a 🤗 org for your class.2. Students create @Gradio demos for their models & upload to org Spaces 🚀.3. Present demos directly from Spaces & keep record for next semester👩💻. 💯 free.
4
36
284
This isn't just a problem with GPT-3. Even GPT-2 suffers from the same bias issues, based on my experiments. You can try it yourself here:
13
35
254
If you visit @PyTorch's website, you'll notice that the website for the models now have a button linking to the model's @Gradio demo on @HuggingFace Spaces!🎉
7
34
264
I expect a lot more machine learning blog posts to follow this pattern in the future ⤵️. 1⃣ Train a model & deploy it as a demo on 🤗 Spaces.2⃣ Write a blog post in Markdown.3⃣ Embed your interactive model using 2 lines of code. Here's an example I'm writing:
5
40
278
In PyTorch, you're probably used to writing your model inference code with 𝚝𝚘𝚛𝚌𝚑.𝚗𝚘_𝚐𝚛𝚊𝚍. BUT there's a better alternative in recent versions of PyTorch. A quick 🧵
6
28
273
Once open-source catches up, it moves much faster than closed APIs
MiniGPT-4 is pretty astonishing: an AI chatbot you can use to ask questions about an image (a feature that's been promised but not yet shipped by GPT-4), building on top of the Vicuna-13B LLM (derived from LLaMA) and BLIP-2 vision-language model
7
25
261
Thanks to the more than 3000 people who have starred the Gradio GitHub repo ✨ 💫. If you haven’t already done so, consider giving us a quick star to help our free, open-source library get more visible!
4
42
244
Serverless machine learning apps . We just released a new version of Gradio. I'm pretty confident that now Gradio is the easiest way to build ML apps that run entirely in your browser (no backend server needed at all!)
6
36
241
I'm so excited to share @Gradio 2.0! We rewrote much of the library & here are some cool things you can do now:. 🤗 Load ~10,000 @huggingface Hub models with intuitive GUIs with 1 line of code.🔗 Link models in Parallel & Series to make more complex workflows. 👇 Examples.
8
55
231
Zero-shot image classification entirely in the browser (no server) in 17 lines of code 🔥
2
37
221
Getting started with a machine learning task? Quickly ramp up to speed with Hugging Face's brand-new Task Pages for 15 common ML problems🚀🚀. Contains example code, interactive widgets, descriptions. What else should we add?
7
50
214
I actually think this LlaVA model might be better than GPT4V 😯.
11
31
208
I weirdly really enjoy just refactoring Python code to make it more readable 😊. BEFORE -> AFTER:
11
13
205
Thinking about having machine learning demo nights, where people demo models that they’ve built and get feedback from the community. Would anyone be interested in joining / presenting?.
25
8
198
🔥 End-to-end machine learning tutorial: Build your own Google Lens Landmark Detection Application .
0
36
198
Just discovered this cool demo being used by a machine learning class at the @SaintLouisU to teach the K nearest neighbors algorithm on MNIST . Demo:
3
44
192
If you use @weights_biases, why not embed a LIVE DEMO of your machine learning model from 🤗 Spaces in your W&B dashboard to let your entire team play around with your model? 🪄. Here's an example. ,. Spaces: W&B:
6
20
186
and it's gradio!.
7
20
173
Excited to share that I'll be teaching a course on computer vision with co:rise! 🧑💻. We’ll “reverse-engineer” and build applications like: Face ID, self-driving, and image generation. Focusing on practical tools like @Gradio and @Huggingface 🤗. More info:
3
25
176
In my PhD research back in 2016, I was working on ways to use machine learning in medical applications. Five years later, I'm excited to share for the first time some ways @Gradio is being used to test biomedical ML models in institutions across the US. Will be sharing more info!
2
18
167
🔥🔥🔥 @BigscienceW's first model (T0pp) is out! . Highlights: 1/16th the size of GPT-3 but outperforms GPT-3 when prompted correctly. In a single line of code, create a demo around it using @Gradio or call the🤗 API:
5
36
168
In the 18 months after @Gradio was acquired by @HuggingFace, our users have grown 22x. We're just getting started, there's so much more to do in order to democratize access to good machine learning across all platforms 🚀🚀
9
15
170
Now that @huggingface Spaces is officially public 🎉and people have made 100s of Spaces already,. Let me share 5 tricks you might not know about a🤗 Space running on @Gradio. .
3
30
166
By now, you've probably heard about Hugging Face Spaces, a *free* way to host @Gradio demos, like this cool depth perception model created by @ak92501 a few hours ago. So how can you get started? Here's a little tweetorial on hosting your first model on @huggingface spaces 🧵
3
35
157
Suppose you have many ML models. How do you compare them for failure points, biases, etc.? Instead of testing them one by one, try:. gradio.mix.Parallel(. ). To compare them easily! You can even compare your own models with pretrained @huggingface models. DEMO!
5
30
148
"Building Generative AI Applications with Gradio". A brand new mini-course covering image generation, LLMs, and more 🥳. Taught by @AndrewYNg and @multimodalart 😎. Take it for free at:
2
38
148
Really stoked to share Gradio's new "Use via API" page. 1⃣ Build a @Gradio app (or find one on Spaces).2⃣ Click on the "Use via API" link in the footer.3⃣ See the expected payload and try it out immediately.4⃣ View handy code snippets in Python or JS. Embed ML everywhere!
4
24
149
BIG update for @gradio and @huggingface Spaces 🎉 . If you've tried a popular Space, you know you can get stuck in a long queue. The most popular Spaces now cache examples so that you can see the results on the examples without waiting in line 🤗. Try:
2
18
145
What this means:. 1. Build a GUI for your machine learning model with our open-source Python library, @Gradio.2. Host it on Hugging Face Spaces for free.3. Public or private repos are both free -- meaning you can share it within your org or with the world . 🚀🥳🔥🤯🛠️🥳😍🚀.
TODAY'S A BIG DAY. Spaces are now publicly available. Build, host, and share your ML apps on @huggingface in just a few minutes. There's no limit to what you can build. Be creative, and share what you make with the community. 🙏 @streamlit and @gradio .
2
30
141
TMR: Text-to-Motion Retrieval Using Contrastive 3D Human Motion Synthesis. ArXiv: Demo:
3
36
134
UniFormer is a great model to keep in mind when you need to segment any kind of image: .
1
21
135
Find a sentiment analysis model in @huggingface, create a @gradio app using Codex and test it out all in 30 seconds. Challenge accepted. ⚡️⚡️
2
19
131
We admitted 65 aspiring PhD students into the inaugural set of Fatima Fellows, and we have matched 38 to mentors! . Some of our research projects:.- debiasing language models.- sampling non-convex spaces.- interpretable robotics.- lifelong machine learning.
1
15
135
Even though 11% of the world speaks Urdu or Hindi, together the two languages only represent about 1.7% of the data that language models are trained on (that's from datasets that are DESIGNED to be multilingual). Who's working on fixing that?
15
33
125
Our small but impactful @Gradio team at @HuggingFace 🤗. We're looking to add someone to our team who can help more developers learn and use Gradio. If you're interested in blogging, developer education, and building cool machine learning demos, DM me!
4
14
130
NEW: Native Embeddings for @Gradio demos🎉. Any Gradio demo on @HuggingFace Spaces can be easily embedded into your blog/portfolio/website 🤗. Comes with a slick UI and no nasty <iframe>s:
6
27
134
Releasing Gradio 2.7.5 🎉 🎉 . ⚡️ We've migrated the entire backend from Flask to @FastAPI! .✍️ You can now write your own custom callbacks for the FLAG button. AND. it includes commits from the newest member of the @Gradio team: Omer Faruk (@farukozderim) 🎊
2
15
130
So excited to announce that @Gradio 2.1 includes Shap explanations out of the box! . Basically, if you have ANY classification/regression model, you can now understand the WHY behind your predictions. All it takes is 1 parameter: interpretation="shap" . Examples & more details👇.
5
27
124
I don't think I've every seen such a complex Gradio app. This is incredible 🤩.
Gather round my #aiArtists , and let me show you something: .This is called #stableDiffusion webui .Not to be confused with #dreamStudio, this runs on your machine/python and is even accessible remotely via ipad/iphone! .Features in 🧵 (you don't want to miss this)
1
6
124