ryyzn9

accelerate

🚀 A simple way to launch, train, and use PyTorch models on almost any device and distributed configuration, automatic mixed precision (including fp8), and easy-to-configure FSDP and DeepSpeed support

AI_to_attend_my_online_classes

So, I made an AI to attend my online classes for me. We all now how boring it gets after a while to attend online classes. So, I made an AI to attend them for me. So, lets get started on what the AI needs to do: These are the things I roughed up the AI needs to do Let's see how will the AI get the data from the class? So from the above image I hope you get the basic gist of how the data collection will work. Lets Start with the data collection and processing: Firstly lets start with the installs we need: Now lets get started with the setup like set up the things we need for the whole code: you will know why we need gender or teacher later in this article Now lets get started with the screenshot part: suppose this is my screen so upon running the code This is the screenshot Now lets take this image and use OpenCV to preprocess it for tesseract OCR: So, basically what the above code does is take the screenshoted image and make it negative so that the black turns white and the white turns black and then detect the the text and draw rectangles around it on the original image. Negative image original image with rectangles Now the image is ready for OCR!! The OCR code is: So, what the above code basically does is use OCR on the preprocessed image and paste the output in a text file hence making a document of whats being said in the class. Here is the recognised file from my AIML class in college: AIML Class 20 Apr 2022.pdf Edit description drive.google.com as you can see the OCR is so fast that it scans the sentences more than once even before they finish (I am in the process of fixing this). now lets make each word a element of a list so that iteration through it is easyer: Now that we have a list lets scan through the words and find if the professor has said my name and respond with yes sir/yes man depending on the gender of the teacher: So, now in the above program the thing thats happening is its going through the list containing all words and searching for my name ‘Eeman’ but as it is captions there might be variation on how the teacher says it so its checking for all the possible spellings that spells Eeman. Now once my name is found it will play a pre-recorded yes sir or yes mam sound according to the gender of the teacher. Now lets get to the notes part: One of the things it wanted this to do is summarise everything that has happened in class at the end of the class. First lets make one using GPT-3: So, in the above code its getting the recognised.txt file from my google drive and turning it into a list and changing it into packets of 4001 tokens and passing it though GPT-3. Then the AI is taking the text and summarising it. Now lets try the GPT-Neo method: So, GPT-3 is paid and me poor and free trial ended so lets make an alternative version in GPT-Neo as it is free 🙂 Here is the code its almost same: And with that my notes is part done. How will the AI mute unmute: So, the AI knows when to say Yes sir/Yes mam but it doesn't know how to mute unmute. So the exact co-ordinates of the mute/unmute button in my screen is this: 1223 X 22 The code is pretty simple here it is: Using MacOs so code is so long now my AI can mute/unmute when needed. Now we need my voice because my AI needs to talk in my voice: For that I am going to use Descript. Its an amazing tool used to make a copy of your voice basically and make it say what you want.(Scary but Awesome). Descript creating my voice over dub This process takes time (a lot of time) so you have to wait 🙂 let wait… Still waiting 🙂 This took more time than I expected. Its done!!! Text can be said in my voice using the app in our case we use the API Once its done I have my voice and use it to make my AI say anything I want! Lets get started with the question detection and answering part: lets start with getting recognised.txt file from my drive: lets download the required libraries: Lets make a function named line-maker to divide the lines as elements of a list: now lets use this function: Now lets google the sentences and find answers: for creative questions lets use GPT-Neo for them: Now our AI can answer questions letsss goooooo. And thats the end of this code. I am going to make a sequel of this which I wanna do stuff with 3d and more believable AI. So, Stay tuned. Here is a recording of my AI attending my class it works well gave me attendance 😁: I f

amber-train

Pre-training code for Amber 7B LLM

annotated_deep_learning_paper_implementations

🧑‍🏫 60 Implementations/tutorials of deep learning papers with side-by-side notes 📝; including transformers (original, xl, switch, feedback, vit, ...), optimizers (adam, adabelief, sophia, ...), gans(cyclegan, stylegan2, ...), 🎮 reinforcement learning (ppo, dqn), capsnet, distillation, ... 🧠

arxiv-sanity-lite

arxiv-sanity lite: tag arxiv papers of interest get recommendations of similar papers in a nice UI using SVMs over tfidf feature vectors based on paper abstracts.

DCFormer

deepLense

Deep Regression Techniques for Decoding Dark Matter with Strong Gravitational Lensing Description Strong gravitational lensing is a promising probe of the substructure of dark matter to better understand its underlying nature. Deep learning methods have the potential to accurately identify images containing substructure, and differentiate WIMP part

Galaxy_Zoo-classification-CNN-Transformer

Classify the morphologies of distant galaxies with cnn-vit

halutmatmul_for_windows

Stella Nera is the first Maddness accelerator achieving 15x higher area efficiency (GMAC/s/mm^2) and 25x higher energy efficiency (TMAC/s/W) than direct MatMul accelerators in the same technology

Open-Llama

The complete training code of the open-source high-performance Llama model, including the full process from pre-training to RLHF.

ryyzn9/DCFormer

ryyzn9/accelerate

🚀 A simple way to launch, train, and use PyTorch models on almost any device and distributed configuration, automatic mixed precision (including fp8), and easy-to-configure FSDP and DeepSpeed support

ryyzn9/halutmatmul_for_windows

Stella Nera is the first Maddness accelerator achieving 15x higher area efficiency (GMAC/s/mm^2) and 25x higher energy efficiency (TMAC/s/W) than direct MatMul accelerators in the same technology

ryyzn9/Open-Llama

The complete training code of the open-source high-performance Llama model, including the full process from pre-training to RLHF.

ryyzn9/AtLas

ryyzn9/flash-attention

Fast and memory-efficient exact attention

ryyzn9/flash-linear-attention

Efficient implementations of state-of-the-art linear attention models in Pytorch and Triton

ryyzn9/gpt-neox

An implementation of model parallel autoregressive transformers on GPUs, based on the Megatron and DeepSpeed libraries

ryyzn9/grok-1

Grok open release

ryyzn9/lightning-attention

Lightning Attention-2: A Free Lunch for Handling Unlimited Sequence Lengths in Large Language Models

ryyzn9/linear_open_lm

A repository for research on medium sized language models.

ryyzn9/llama3

The official Meta Llama 3 GitHub site

ryyzn9/llamafile

Distribute and run LLMs with a single file.

ryyzn9/LLM-Agents-Papers

A repo lists papers related to LLM based agent

ryyzn9/llm-foundry

LLM training code for Databricks foundation models

ryyzn9/LLMTest_NeedleInAHaystack

Doing simple retrieval from LLM models at various context lengths to measure accuracy

ryyzn9/MetaGPT

🌟 The Multi-Agent Framework: First AI Software Company, Towards Natural Language Programming

ryyzn9/minLlama3

a quick & complete guide to Llama 3's architecture

ryyzn9/nanoGPT-TK

The simplest, fastest repository for training/finetuning medium-sized GPTs. Now, with kittens!

ryyzn9/ollama

Get up and running with Llama 3, Mistral, Gemma, and other large language models.

ryyzn9/othello_mamba

Evaluating the Mamba architecture on the Othello game

ryyzn9/pykan

Kolmogorov Arnold Networks

ryyzn9/pythia

The hub for EleutherAI's work on interpretability and learning dynamics

ryyzn9/retnet_transformer

ryyzn9/ThunderKittens

Tile primitives for speedy kernels

ryyzn9/tiny-gpu

A minimal GPU design in Verilog to learn how GPUs work from the ground up

ryyzn9/torchscale

Foundation Architecture for (M)LLMs

ryyzn9/unilm

Large-scale Self-supervised Pre-training Across Tasks, Languages, and Modalities

ryyzn9/vllm

A high-throughput and memory-efficient inference and serving engine for LLMs

ryyzn9/X_net

a new transformer architecture