Abhishek Salian

Building

[Abhishek Salian @acvsalian]

@prathoshap Sir, it would be wonderful if you could also include your Vedvaani app, podcasts, and Vedic discourses. It could become a valuable single access point for students to access all your content conveniently in one place.

[prathosh ap @prathoshap]

Namaste X!

This is my first post here - I'm humbled to release my app: VedaVaaNi.

It is an interactive learning platform designed to help you practice Rig Veda and Krishna Yajur Veda chanting with absolute phonetic and prosodic (Chandas) precision.

[Shubham Saboo @Saboo_Shubham_]

Vector databases for AI memory just got replaced by MP4 files.

Store millions of text chunks in MP4 files instead of expensive vector databases with lightning-fast semantic search.

No database needed. 100% Opensource.

[Rishi Jha @rishi_d_jha]

I’m stoked to share our new paper: “Harnessing the Universal Geometry of Embeddings” with @jxmnop, Collin Zhang, and @shmatikov.

We present the first method to translate text embeddings across different spaces without any paired data or encoders.

Here's why we're excited: 🧵👇🏾

[Abhishek Salian @acvsalian]

"Looking to host your own PyPI server for private packages? 🤔 Check out this guide on using GitLab to set it up efficiently! 🚀

https://t.co/XQdRJ6NVKB

#Python #PyPI #DevOps #SoftwareDevelopment

[Jiawei Yang @JiaweiYang118]

Very excited to get this out: “DVT: Denoising Vision Transformers”. We've identified and combated those annoying positional patterns in many ViTs. Our approach denoises them, achieving SOTA results and stunning visualizations! Learn more on our website: https://t.co/RFEiZQx7ZZ

[Martin Bauer @martinmbauer]

Many know Euler's identity, but did you also know that

[Abhishek Salian @acvsalian]

Computational Models for Understanding the Human Brain https://t.co/pAItfHDbhS

[Abhishek Salian @acvsalian]

https://t.co/pXxsG0lym8

[Yam Peleg @Yampeleg]

If you run this on a clean ubuntu 22.04 It will install CUDA + GPU Drivers + CuDNN and everything else you will need to start working.

[Battle-tested as of 5-min ago,
I am updating the GPU stack on some machines]

Enjoy!

---- 𝚒𝚗𝚜𝚝𝚊𝚕𝚕_𝚐𝚙𝚞_𝚞𝚋𝚞𝚗𝚝𝚞_𝟸𝟸𝟶𝟺.𝚜𝚑 ----

#!/𝚋𝚒𝚗/𝚋𝚊𝚜𝚑

# Verify if GPU is CUDA-enabled
𝚕𝚜𝚙𝚌𝚒 | 𝚐𝚛𝚎𝚙 -𝚒 𝚗𝚟𝚒𝚍𝚒𝚊

# Remove previous NVIDIA driver installation
𝚜𝚞𝚍𝚘 𝚊𝚙𝚝-𝚐𝚎𝚝 𝚙𝚞𝚛𝚐𝚎 𝚗𝚟𝚒𝚍𝚒𝚊* -𝚢
𝚜𝚞𝚍𝚘 𝚊𝚙𝚝 𝚛𝚎𝚖𝚘𝚟𝚎 𝚗𝚟𝚒𝚍𝚒𝚊-* -𝚢
𝚜𝚞𝚍𝚘 𝚛𝚖 /𝚎𝚝𝚌/𝚊𝚙𝚝/𝚜𝚘𝚞𝚛𝚌𝚎𝚜.𝚕𝚒𝚜𝚝.𝚍/𝚌𝚞𝚍𝚊* -𝚢
𝚜𝚞𝚍𝚘 𝚊𝚙𝚝-𝚐𝚎𝚝 𝚊𝚞𝚝𝚘𝚛𝚎𝚖𝚘𝚟𝚎 && 𝚜𝚞𝚍𝚘 𝚊𝚙𝚝-𝚐𝚎𝚝 𝚊𝚞𝚝𝚘𝚌𝚕𝚎𝚊𝚗 -𝚢
𝚜𝚞𝚍𝚘 𝚛𝚖 -𝚛𝚏 /𝚞𝚜𝚛/𝚕𝚘𝚌𝚊𝚕/𝚌𝚞𝚍𝚊* -𝚢

# System update
𝚜𝚞𝚍𝚘 𝚊𝚙𝚝-𝚐𝚎𝚝 𝚞𝚙𝚍𝚊𝚝𝚎 -𝚢
𝚜𝚞𝚍𝚘 𝚊𝚙𝚝-𝚐𝚎𝚝 𝚞𝚙𝚐𝚛𝚊𝚍𝚎 -𝚢

# Install essential packages
𝚜𝚞𝚍𝚘 𝚊𝚙𝚝-𝚐𝚎𝚝 𝚒𝚗𝚜𝚝𝚊𝚕𝚕 𝚐++ 𝚏𝚛𝚎𝚎𝚐𝚕𝚞𝚝𝟹-𝚍𝚎𝚟 𝚋𝚞𝚒𝚕𝚍-𝚎𝚜𝚜𝚎𝚗𝚝𝚒𝚊𝚕 𝚕𝚒𝚋𝚡𝟷𝟷-𝚍𝚎𝚟 -𝚢
𝚜𝚞𝚍𝚘 𝚊𝚙𝚝-𝚐𝚎𝚝 𝚒𝚗𝚜𝚝𝚊𝚕𝚕 𝚕𝚒𝚋𝚡𝚖𝚞-𝚍𝚎𝚟 𝚕𝚒𝚋𝚡𝚒-𝚍𝚎𝚟 𝚕𝚒𝚋𝚐𝚕𝚞𝟷-𝚖𝚎𝚜𝚊 𝚕𝚒𝚋𝚐𝚕𝚞𝟷-𝚖𝚎𝚜𝚊-𝚍𝚎𝚟 -𝚢

# Add PPA repository for NVIDIA drivers
𝚜𝚞𝚍𝚘 𝚊𝚍𝚍-𝚊𝚙𝚝-𝚛𝚎𝚙𝚘𝚜𝚒𝚝𝚘𝚛𝚢 𝚙𝚙𝚊:𝚐𝚛𝚊𝚙𝚑𝚒𝚌𝚜-𝚍𝚛𝚒𝚟𝚎𝚛𝚜/𝚙𝚙𝚊 -𝚢
𝚜𝚞𝚍𝚘 𝚊𝚙𝚝 𝚞𝚙𝚍𝚊𝚝𝚎 -𝚢

# Install NVIDIA driver and dependencies
𝚜𝚞𝚍𝚘 𝚊𝚙𝚝-𝚐𝚎𝚝 𝚒𝚗𝚜𝚝𝚊𝚕𝚕 -𝚢 𝚌𝚞𝚍𝚊-𝚍𝚛𝚒𝚟𝚎𝚛𝚜

# Download and set up CUDA repository
𝚠𝚐𝚎𝚝 𝚑𝚝𝚝𝚙𝚜://𝚍𝚎𝚟𝚎𝚕𝚘𝚙𝚎𝚛.𝚍𝚘𝚠𝚗𝚕𝚘𝚊𝚍.𝚗𝚟𝚒𝚍𝚒𝚊.𝚌𝚘𝚖/𝚌𝚘𝚖𝚙𝚞𝚝𝚎/𝚌𝚞𝚍𝚊/𝚛𝚎𝚙𝚘𝚜/𝚞𝚋𝚞𝚗𝚝𝚞𝟸𝟸𝟶𝟺/𝚡𝟾𝟼_𝟼𝟺/𝚌𝚞𝚍𝚊-𝚞𝚋𝚞𝚗𝚝𝚞𝟸𝟸𝟶𝟺.𝚙𝚒𝚗
𝚜𝚞𝚍𝚘 𝚖𝚟 𝚌𝚞𝚍𝚊-𝚞𝚋𝚞𝚗𝚝𝚞𝟸𝟸𝟶𝟺.𝚙𝚒𝚗 /𝚎𝚝𝚌/𝚊𝚙𝚝/𝚙𝚛𝚎𝚏𝚎𝚛𝚎𝚗𝚌𝚎𝚜.𝚍/𝚌𝚞𝚍𝚊-𝚛𝚎𝚙𝚘𝚜𝚒𝚝𝚘𝚛𝚢-𝚙𝚒𝚗-𝟼𝟶𝟶
𝚠𝚐𝚎𝚝 𝚑𝚝𝚝𝚙𝚜://𝚍𝚎𝚟𝚎𝚕𝚘𝚙𝚎𝚛.𝚍𝚘𝚠𝚗𝚕𝚘𝚊𝚍.𝚗𝚟𝚒𝚍𝚒𝚊.𝚌𝚘𝚖/𝚌𝚘𝚖𝚙𝚞𝚝𝚎/𝚌𝚞𝚍𝚊/𝟷𝟸.𝟹.𝟷/𝚕𝚘𝚌𝚊𝚕_𝚒𝚗𝚜𝚝𝚊𝚕𝚕𝚎𝚛𝚜/𝚌𝚞𝚍𝚊-𝚛𝚎𝚙𝚘-𝚞𝚋𝚞𝚗𝚝𝚞𝟸𝟸𝟶𝟺-𝟷𝟸-𝟹-𝚕𝚘𝚌𝚊𝚕_𝟷𝟸.𝟹.𝟷-𝟻𝟺𝟻.𝟸𝟹.𝟶𝟾-𝟷_𝚊𝚖𝚍𝟼𝟺.𝚍𝚎𝚋
𝚜𝚞𝚍𝚘 𝚍𝚙𝚔𝚐 -𝚒 𝚌𝚞𝚍𝚊-𝚛𝚎𝚙𝚘-𝚞𝚋𝚞𝚗𝚝𝚞𝟸𝟸𝟶𝟺-𝟷𝟸-𝟹-𝚕𝚘𝚌𝚊𝚕_𝟷𝟸.𝟹.𝟷-𝟻𝟺𝟻.𝟸𝟹.𝟶𝟾-𝟷_𝚊𝚖𝚍𝟼𝟺.𝚍𝚎𝚋
𝚜𝚞𝚍𝚘 𝚌𝚙 /𝚟𝚊𝚛/𝚌𝚞𝚍𝚊-𝚛𝚎𝚙𝚘-𝚞𝚋𝚞𝚗𝚝𝚞𝟸𝟸𝟶𝟺-𝟷𝟸-𝟹-𝚕𝚘𝚌𝚊𝚕/𝚌𝚞𝚍𝚊-*-𝚔𝚎𝚢𝚛𝚒𝚗𝚐.𝚐𝚙𝚐 /𝚞𝚜𝚛/𝚜𝚑𝚊𝚛𝚎/𝚔𝚎𝚢𝚛𝚒𝚗𝚐𝚜/
𝚜𝚞𝚍𝚘 𝚊𝚙𝚝-𝚐𝚎𝚝 𝚞𝚙𝚍𝚊𝚝𝚎 -𝚢

# Install CUDA Toolkit 12.3
𝚜𝚞𝚍𝚘 𝚊𝚙𝚝-𝚐𝚎𝚝 -𝚢 𝚒𝚗𝚜𝚝𝚊𝚕𝚕 𝚌𝚞𝚍𝚊-𝚝𝚘𝚘𝚕𝚔𝚒𝚝-𝟷𝟸-𝟹

# Set up paths for CUDA
𝚎𝚌𝚑𝚘 '𝚎𝚡𝚙𝚘𝚛𝚝 𝙿𝙰𝚃𝙷=/𝚞𝚜𝚛/𝚕𝚘𝚌𝚊𝚕/𝚌𝚞𝚍𝚊-𝟷𝟸.𝟹/𝚋𝚒𝚗:$𝙿𝙰𝚃𝙷' >> ~/.𝚋𝚊𝚜𝚑𝚛𝚌
𝚎𝚌𝚑𝚘 '𝚎𝚡𝚙𝚘𝚛𝚝 𝙻𝙳_𝙻𝙸𝙱𝚁𝙰𝚁𝚈_𝙿𝙰𝚃𝙷=/𝚞𝚜𝚛/𝚕𝚘𝚌𝚊𝚕/𝚌𝚞𝚍𝚊-𝟷𝟸.𝟹/𝚕𝚒𝚋𝟼𝟺:$𝙻𝙳_𝙻𝙸𝙱𝚁𝙰𝚁𝚈_𝙿𝙰𝚃𝙷' >> ~/.𝚋𝚊𝚜𝚑𝚛𝚌
𝚜𝚘𝚞𝚛𝚌𝚎 ~/.𝚋𝚊𝚜𝚑𝚛𝚌
𝚜𝚞𝚍𝚘 𝚕𝚍𝚌𝚘𝚗𝚏𝚒𝚐

# Install cuDNN v11.3
𝚠𝚐𝚎𝚝 𝚑𝚝𝚝𝚙𝚜://𝚍𝚎𝚟𝚎𝚕𝚘𝚙𝚎𝚛.𝚗𝚟𝚒𝚍𝚒𝚊.𝚌𝚘𝚖/𝚌𝚘𝚖𝚙𝚞𝚝𝚎/𝚖𝚊𝚌𝚑𝚒𝚗𝚎-𝚕𝚎𝚊𝚛𝚗𝚒𝚗𝚐/𝚌𝚞𝚍𝚗𝚗/𝚜𝚎𝚌𝚞𝚛𝚎/𝟾.𝟸.𝟷.𝟹𝟸/𝟷𝟷.𝟹_𝟶𝟼𝟶𝟽𝟸𝟶𝟸𝟷/𝚌𝚞𝚍𝚗𝚗-𝟷𝟷.𝟹-𝚕𝚒𝚗𝚞𝚡-𝚡𝟼𝟺-𝚟𝟾.𝟸.𝟷.𝟹𝟸.𝚝𝚐𝚣
𝚝𝚊𝚛 -𝚡𝚣𝚟𝚏 "𝚌𝚞𝚍𝚗𝚗-𝟷𝟷.𝟹-𝚕𝚒𝚗𝚞𝚡-𝚡𝟼𝟺-𝚟𝟾.𝟸.𝟷.𝟹𝟸.𝚝𝚐𝚣"

# Copy cuDNN files to CUDA toolkit directory
𝚜𝚞𝚍𝚘 𝚌𝚙 -𝙿 𝚌𝚞𝚍𝚊/𝚒𝚗𝚌𝚕𝚞𝚍𝚎/𝚌𝚞𝚍𝚗𝚗.𝚑 /𝚞𝚜𝚛/𝚕𝚘𝚌𝚊𝚕/𝚌𝚞𝚍𝚊-𝟷𝟸.𝟹/𝚒𝚗𝚌𝚕𝚞𝚍𝚎
𝚜𝚞𝚍𝚘 𝚌𝚙 -𝙿 𝚌𝚞𝚍𝚊/𝚕𝚒𝚋𝟼𝟺/𝚕𝚒𝚋𝚌𝚞𝚍𝚗𝚗* /𝚞𝚜𝚛/𝚕𝚘𝚌𝚊𝚕/𝚌𝚞𝚍𝚊-𝟷𝟸.𝟹/𝚕𝚒𝚋𝟼𝟺/
𝚜𝚞𝚍𝚘 𝚌𝚑𝚖𝚘𝚍 𝚊+𝚛 /𝚞𝚜𝚛/𝚕𝚘𝚌𝚊𝚕/𝚌𝚞𝚍𝚊-𝟷𝟸.𝟹/𝚕𝚒𝚋𝟼𝟺/𝚕𝚒𝚋𝚌𝚞𝚍𝚗𝚗*

# Install nvtop for monitoring
𝚜𝚞𝚍𝚘 𝚊𝚙𝚝 𝚒𝚗𝚜𝚝𝚊𝚕𝚕 𝚗𝚟𝚝𝚘𝚙 -𝚢

# Verify the installation
𝚗𝚟𝚒𝚍𝚒𝚊-𝚜𝚖𝚒
𝚗𝚟𝚌𝚌 -𝚅

[Martin Bauer @martinmbauer]

About plagiarism: Plagiarism isn't really the problem, it's a symptom.

The problem is that there're people going into a PhD primarily not because they want to do original research but because they want the PhD title.

1/4

[Abhishek Salian @acvsalian]

@martinmbauer Well said

[Aether @aether_coin]

Looking at the binary code on a CD

[Frank Nielsen @FrnkNlsn]

Browse over 500 cards related to machine learning, divergences, and geonetry.

https://t.co/7DkExOZZPi

[Vaibhav (VB) Srivastav @reach_vb]

Want high-quality Audio embeddings? CLAP! 👏

We support the latest general, music and speech CLAP models in Transformers! Use it for Text-to-Speech/ Text-to-Music training and more.

What is CLAP?

CLAP (Contrastive Language-Audio Pretraining) is a neural network trained on various (audio, text) pairs. It uses a SWINTransformer to get audio features from a log-Mel spectrogram input, and a RoBERTa model to get text features.

Combined CLAP gives you high-quality audio embeddings blazingly fast! ⚡️

How do you use CLAP?

from transformers import ClapModel, ClapProcessor

# initialise the model
model = ClapModel.from_pretrained(
"laion/larger_clap_music").to("cuda:0")

processor = ClapProcessor.from_pretrained(
"laion/larger_clap_music")

# process the inputs
inputs = processor(
audios=<PASS YOUR AUDIO SAMPLE HERE>,
return_tensors="pt").to("cuda:0")

# generate embeddings
audio_embed = model.get_audio_features(**inputs)

What would you like to embed first? :)

P.S. Remember to check out other CLAP checkpoints in the @laion_ai org on the hub!

That's it! 🤗

[World of Engineering @engineers_feed]

The difference in quality of each teapot

[Stefano Gogioso @StefanoGogioso]

Spaces of quantum states

It is well known that the (pure) states of a qubit have the geometrical structure of a sphere, a representation typically known as the Bloch sphere. For quantum systems of higher dimensions, the state space is significantly harder to visualise geometrically, but we can nonetheless gain some intuition by describing it topologically, i.e. in some continuously deformed way that might be easier to grok.

One way to understand the topology of the space of states for a quantum system is to fix a reference basis (often known as the "computational" basis) and to separate the information of a quantum state into two distinct components:

- the classical probability distribution that arises by measuring the state in the reference basis
- the quantum phases that are lost in the measurement process

The space of classical probability distributions is easy to describe, both geometrically and topologically: for an n-dimensional system, this is the (n-1)-dimensional simplex. It is a line for n=2, a triangle for n=3, a tetrahedron for n=4, and so on.

The space of quantum phases for a given basis is also easy to describe, again both geometrically and topologically: it is always a torus, and its dimensionality depends on the number of non-zero components in the probability distribution associated to the state. It is: the 0-torus (a point) for probability distribution fully concentrated on a single outcome; the 1-torus (topologically, a circle) for probability distribution with two possible outcomes; the 2-torus (topologically, a doughnut) for probability distributions with three possible outcomes; and so on.

To obtain a topological description of the full space, we have to glue the two parts together. For an n-dim quantum system, we start with the (n-1)-dim simplex, describing the possible probability distributions that can arise from measurements in the reference basis. Then we glue a torus of the appropriate dimensionality onto each point of the simplex: a point on the vertices, a circle on each point in the interior of the edges, a doughnut on each point in the interior of the triangular faces, a 3-torus on each point in the interior of a tetrahedron, and so on.

Et voilà! The space of quantum states, obtained as the gluing of quantum phases over classical probability distributions.

PS: Points at the same location in tori at nearby points of a simplex are themselves nearby (topologically close). Passing from the interior to the boundary of any face (e.g. from the interior of a triangle to the interior of an edge) has the effect of collapsing one of the circular dimensions of the torus to a point, reducing its dimensionality by 1. (There was no space for this last point in today's sketch, so I'll show the overall effect in a future visualisation.)

[Valeriy M., PhD, MBA, CQF @predict_addict]

Yet another day, another article 'predicting' gold price using fit_predict ML.

No benchmarking, wrong metrics, all sorts of other issues.

When someone thinks they can predict the prices of stocks/commodities using fit_predict, some might offer London Bridge to sell to them.

But for folks who want to understand why this is so wrong, I have written an article, 'Avoiding the Forecasting Pitfalls: 10 Red Flags in Hiring Data Scientists,' and you can easily count how many forecasting red flags are there for yourself.

https://t.co/fdxCrD02S8

[Massimo @Rainmaker1973]

When air is pumped into sand, it turns sand into a liquid-like substance

[📹 Mark Rober]

[DAIR.AI @dair_ai]

Top ML Papers of the Week (Oct 30 - Nov 5):

- LLMs for Chip Design
- Battle of the Backbones
- Next Generation AlphaFold
- Symmetry in Machine Learning
- Enhancing LLMs by Emotion Stimuli
- Efficient Context Window Extension of LLMs
...