Nikita Katyal

If you're working in data science, data engineering, or simply interested in learning how to design machine learning applications for production, I strongly recommend reading the book 'Designing Machine Learning Systems' written by Chip Huyen. The book offers valuable insights starting from Training data, Feature engineering, Model deployment, Evaluation, Prediction, Monitoring, and Continuous learning in production. Please let me know of any books you've recently read in the ML domain that you'd like to recommend. The next book in my list is 'Designing Data Intensive Applications'.

30

4 ความคิดเห็น

{ MISTRAL Ai & NVIDIA COLLAB } Mistral releases 12B open LLM! 🤯 Mistral NeMo comes as a base and instruct version with a 128k Context window and is multilingual in 9 Language with a new tokenizer ( https://lnkd.in/gjCbz_yz ) TL;DR: 🧠 12B Base and Instruct drop-in-replacement for Mistral 7B 🪟 Supports 128k context window and new tokenizer, Tekken, based on Tiktoken 🌍 Base Model multilingual in English, French, German, Spanish, Italian and more 🔓 Released under Apache 2.0 🏆 Base MMLU 68.0%; Instruct 53.4% MixEval Hard; ⚒️ Instruct versions support function calling 🤯 quantized aware-training for FP8 inference without any performance loss 🤝 Created as a Collaboration between NVIDIA and Mistral AI 🚀 Trained on 3,072 H100 80GB on DGX Cloud 🤗 Available on Hugging Face

1

[arXiv] Deep Learning job interview questions, fully solved & covering a wide range of key AI topics https://lnkd.in/gWkziAH5

8

1 out of 6 people die of cancer worldwide! Check out my visualization on Tableau following the global cancer trend. Link to viz : https://lnkd.in/gVkMTGuV Datasource : World Cancer Research Fund International #datafam #tableau #datavisualisation

181

26 ความคิดเห็น

🔥 Sneak peek of my fireside chat with Chip Huyen for Outerbounds! We explore: 🤖 Chip's journey from writing to AI 🔧 Composable data systems and GPU optimization 🧠 AI's potential in solving complex problems 📚 The future of AI in storytelling ⚡ Writing meets computer science in AI Full conversation in comments below #AIEngineering #Storytelling #MachineLearning

58

1 ความคิดเห็น

Yet another Sentiment Classification Kaggle notebook. What’s the difference? - Why you could have been using the wrong model - Why you could have been using the right model wrong - Example of detective work to pinpoint your model’s failure patterns and how to fix them - Problem solving and Critical thinking in practice - How and where to find answers - “Those who do not learn history are doomed to repeat it” This notebook is perfect for beginners/immediate ML practitioners who aren’t satisfied with “it just werks”.

2

Hello everyone, I recently experimented with the llama-cpp-python package on MacBook with M2-chip. My goal was to benchmark the performance of off-the-shelf models running on GPUs versus the same models running on CPUs. Here are my key observations: - The installation of the llama-cpp-python package has significantly improved, especially on Apple Chips. - Certainly the underlying code written in C/C++ leads to huge efficiency gains in terms of time. - I discovered that llama model objects can be directly used to generate text. For example, a 7B parameter model requiring close to 7 GB could be easily loaded on 16 GB RAM to generate text like "Describe the dog breed Vizsla in a few sentences." - However, more nuanced functions such as create_chat_completion and create_completion cause a momentary spike in memory usage, nearing 32 GB for these 7B parameters so-called "innocuous models", resulting in segmentation errors on my MacBook with 16GB RAM. This led me to troubleshoot the issues on an AWS Linux instance. These observations highlight the stability improvements on Apple Chips, the memory requirements for different model functions, and the possible need for higher RAM capacity when dealing with more complex tasks. #Technology #PerformanceBenchmarking #AppleChips #LLAMACPPPython #ModelGeneration #RAMUsage

7

https://lnkd.in/gnfminyA Link to a housw_sale regression dataset I worked on, please share your thoughts #dataset #ml #machinelearning

1

Create and understand classification models in machine learning - https://lnkd.in/gzY9TUqr Classification means assigning items into categories, or can also be thought of automated decision making. Here we introduce classification models through logistic regression, providing you with a stepping-stone toward more complex and exciting classification methods. Learning objectives In this module, you will: Discover how classification differs from classical regression Build models that can perform classification tasks Explore how to assess and improve classification models Prerequisites Familiarity with machine learning models https://lnkd.in/gzY9TUqr

1

I just completed the first chapter of this amazing book Designing Machine Learning Systems by Chip Huyen and I realised how much ground work has went through to just create the first chapter and I am yet go further. She has explained so well the difference between ML in academics and ML in Production. I won’t go into the details, for that you need to read the book. But I remember a recent incidence where we got a project to summarise tables using GenAI and I, naively, suggested that we can just converted all rows into text and stored it in vector database and from there we can created a RAG based Chatbot. It is simple and I showed the results using a sample table and the results were also good. However, an experienced colleague from engineering team said, “Do you know the size of dataset, and do you know the cost of vector db to store it”. The table size was huge and to store it in vector db will also be huge. So, we as a Data Scientist not only need to deliver the results but do need to take care of lot of factors and this book pin points all. I would recommend all the budding Data Scientist to read this book, it will help you a lot in your journey as a Data Scientist.

108

8 ความคิดเห็น

Exciting Research Alert: Revolutionizing Dense Passage Retrieval with Entailment Tuning! The good folks at HKUST have developed a novel approach that significantly improves information retrieval by leveraging natural language inference. The entailment tuning approach consists of several key steps to enhance dense passage retrieval performance. Data Preparation - Convert questions into existence claims using rule-based transformations. - Combine retrieval data with NLI data from SNLI and MNLI datasets. - Unify the format of both data types using a consistent prompting framework. Entailment Tuning Process - Initialize the model using pre-trained language models like BERT or RoBERTa. - Apply aggressive masking (β=0.8) specifically to the hypothesis components while preserving premise information. - Train the model to predict the masked hypothesis tokens from the premise content. - Run the training for 10 epochs using 8 GPUs, taking approximately 1.5-3.5 hours. Training Arguments for Entailment Tuning (Yes! They Shared Them) - Use a learning rate of 2e-5 with 100 warmup steps. - Set batch size to 128. - Apply weight decay of 0.01. - Utilize the Adam optimizer with beta values (0.9, 0.999). - Maintain maximum gradient norm at 1.0. Deployment - Index passages using FAISS for efficient retrieval. - Shard vector store across multiple GPUs. - Enable sub-millisecond retrieval of the top-100 passages per query. Integration with Existing Systems - Insert entailment tuning between pre-training and fine-tuning stages. - Maintain compatibility with current dense retrieval methods. - Preserve existing contrastive learning approaches during fine-tuning. Simple, intuitive, and effective! This advancement significantly improves the quality of retrieved passages for question-answering systems and retrieval-augmented generation tasks.

14

If you're eager to dive deep into understanding transformers, there's no better way than getting your hands dirty and building one from scratch. In this repository, you'll find a comprehensive breakdown of the transformer model's inner workings, including: 1. Encoder-Decoder Implementation: Dive into the translation project from scratch, unraveling the intricacies of both the encoder and decoder components. 2.Positional Encoding Demystified: Understand positional encoding from scratch, a crucial aspect of transformer architecture. 3.Attention Mechanism Dissected: Explore the attention mechanism from scratch, which forms the backbone of the transformer's power. 4. GPT2 Architecture Unveiled: Delve into building the GPT2 architecture from scratch in PyTorch, gaining insights into its structure and functionality. #transformer #DL #AI #ML #genai #GPT #GPT2 #jobseeker https://lnkd.in/g-MGEyTi

4

1 ความคิดเห็น

4-Stage Recommender Systems : Retrieval -> Filtering -> Scoring/Ranking -> Ordering 1. Retrieval: Gather a broad set of potential items using techniques like collaborative filtering or content-based filtering. 2. Filtering: Narrow down the set based on user preferences, browsing history, or contextual factors like time and location. 3. Scoring: Use machine learning models, such as neural networks or gradient boosting machines etc., to predict and assign relevance scores to each filtered item. 4. Ordering: Arrange the ranked items in a sequence that maximizes user engagement or satisfaction, considering factors like diversity, novelty, and business goals. This ordered list is then presented to the user as personalized recommendations.

5