Li Yin

🚀 Looking to streamline fine-tuning for models like Llama 3, Mistral, and Mixtral? Our latest blog shows how Anyscale simplifies LLMs fine-tuning with Ray: 🔑 Key takeaways: • Scalable fine-tuning with LLM-Forge • Model Deployment with RayLLM Read more: https://lnkd.in/gQwD4A2U

35

Learn how you can convert a PyTorch model, trained on the web, to a mobile-optimized format. Madhur Zanwar describes the process using a few lines of code. He also discusses his journey and how one approach didn't yield satisfactory results while trying to run it in a React Native app. #machinelearning #pytorch #computervision #edgeml #reactnative #objectdetection #yolov5

14

🚀 Check out the latest Medium post by our student Tianyunxi Yin: "Quantization Fundamentals for Model Compression". Explore the essentials of quantization and learn how it helps in compressing machine learning models for better efficiency without compromising performance. Read more here: https://lnkd.in/gHbbR_J4 #USFMDS #DataScience #MachineLearning #ModelCompression #Quantization

15

Good resource in fastRAG

2

New repository for applying Constrained Deep Learning! Check out this blog post to learn how to design constrained models, and how constrained deep learning makes neural networks easier to verify 👉 https://spr.ly/6045gmKQv

206

1 Comment

Learn the basics of nonlinear dimensionality reduction, kernel PCA, and multidimensional scaling by following along Biman Chakraborty's clear, hands-on Python tutorial.

18

Text embeddings have become crucial for transforming textual data into dense vector representations (with context), enabling efficient processing across various modalities like text, images, and audio. This article provides a good summary on some of these (not all) embeddings available for text https://lnkd.in/gKp4ta9a 1. Gensim Word Embeddings (Free) Why Use It: Specializes in NLP with open-source models like Word2Vec and FastText, promoting transparency and ease of custom training. 2. Facebook Embeddings (Free) Why Use It: Trained on extensive corpora and supports over 100 languages, making it versatile for global applications. 3. AllenNLP Embeddings (Free) Why Use It: Provides NLP-focused embeddings like BERT and ELMo, with tools for fine-tuning and visualization. 4. OpenAI Embeddings (Paid) Why Use It: Comprehensive training on massive datasets allows for effective semantic understanding, making it suitable for advanced NLP tasks. Ideal for tasks requiring zero-shot learning. 5. MultiLingual BERT (Free) Why Use It: Handles multilingual datasets effectively, offering robust support for multiple languages. Perfect for applications requiring consistent performance across different languages, such as global customer support systems. 6. DistilBERT Base Uncased (Free) Why Use It: Smaller and faster than BERT, maintaining near-comparable performance with reduced resource usage. 7. MPNet V2 (Free) Why Use It: Excels in text similarity tasks with a Siamese architecture, providing accurate similarity measurements. 8. HuggingFace Embeddings (Free/Paid) Why Use It: Offers versatility with a wide range of embeddings (text, image, audio) and the ability to fine-tune on custom data. HF also provides wrappers to almost all open sourced models. 9. Longformer Base 4096 (Free) Why Use It: Supports long texts up to 4096 tokens, making it unique in handling lengthy documents without truncation. 10. RoBERTa (2022) (Free) Why Use It: Offers updated text representations with comprehensive training on recent data. 11. GTE-Base (Free) Why Use It: Offers a good balance of performance and resource requirements with a 768-dimensional embedding. 12. GTE-Large (Free) Why Use It: Provides high-quality embeddings with more detailed representations, useful for nuanced text analysis. Best for applications requiring in-depth text understanding, such as detailed content recommendations. 13. GTE-Small (Free) Why Use It: Optimized for speed with a smaller model size, making it efficient for applications needing quick processing. Useful for real-time systems where fast text similarity searches are critical. 14. E5-Small (Free) Why Use It: Combines compact size with good performance, making it suitable for fast, general-purpose text embedding tasks.

11

📢 New Breakthrough in Time Series Forecasting: TSMamba Model 📢 Researchers from Tsinghua University and several esteemed institutions have introduced a novel time series foundation model named TSMamba. This model leverages the Mamba architecture to tackle the complexity of time series forecasting across diverse domains. Unlike traditional models reliant on quadratic complexity architectures like Transformers, TSMamba offers a linear complexity alternative, enhancing both efficiency and scalability. Key Highlights: - Bidirectional Encoding: TSMamba incorporates both forward and backward Mamba encoders to capture time dependencies effectively. - Two-Stage Transfer Learning: To reduce the need for extensive data and computational resources, TSMamba utilizes pretrained Mamba models, allowing for effective adaptation to specific time series tasks. - Cross-Channel Attention: A compressed attention module captures dependencies across channels, enhancing predictive accuracy on multivariate datasets. Experiments show that TSMamba performs on par or even better than current state-of-the-art models, especially in scenarios with limited training data, such as zero-shot and full-shot forecasting. This approach could revolutionize forecasting across industries, from finance to healthcare, by offering more data-efficient and adaptable solutions. 📚 Read more about the research here: https://lnkd.in/d_k3sfgv #TimeSeries #Forecasting #MachineLearning

52

Model-Based Transfer Learning for Contextual Reinforcement Learning https://lnkd.in/ez-UMPmU

2

Cerebras Introduces CePO (Cerebras Planning and Optimization): An AI Framework that Adds Sophisticated Reasoning Capabilities to the Llama Family of Models Cerebras has introduced CePO (Cerebras Planning and Optimization), an AI framework designed to enhance the reasoning and planning capabilities of the Llama family of models. CePO integrates optimization algorithms with Llama’s language modeling capabilities, enabling it to address complex reasoning tasks that previously required multiple tools CePO’s core innovation lies in embedding planning capabilities directly into the Llama models. This eliminates the need for external optimization engines, allowing the models to reason through multi-step problems, manage trade-offs, and make decisions autonomously. These features make CePO suitable for applications in logistics, healthcare planning, and autonomous systems where precision and adaptability are essential CePO enhances Llama models with a specialized planning and reasoning layer. This layer employs reinforcement learning and advanced constraint-solving techniques to facilitate long-term decision-making. Unlike traditional AI systems, which often require predefined rules or domain-specific training data, CePO generalizes its optimization strategies across various tasks. A key technical feature of CePO is its integration of neural-symbolic methods. By combining neural network learning with symbolic reasoning, CePO achieves both adaptability and interpretability. It also includes a dynamic memory module that enables it to respond effectively to evolving scenarios, improving performance in real-time planning tasks..... Read the full article here: https://lnkd.in/gjE8uikP Check out more details: https://lnkd.in/gv66CTvM Cerebras Systems Emma Call Connor Anderson Jessica Liu Ryan Humble Ganesh Venkatesh Daria Soboleva Natalia Vassilieva Angela Yeung

74

Join Hasham Ul Haq, our Senior Machine Learning Engineer presenting ‘Running State-of-the-art #MedicalLLMs on Databricks’ at the Data + AI Databricks Summit on June 12, 2024, at 2:50 pm PDT. Healthcare-specific #largelanguagemodels provide superior accuracy to general-purpose LLMs. This session shares current benchmarks of John Snow Labs’ Medical Large Language Models in three areas – question answering, information extraction, and chat quality – as evaluated on a combination of widely used medical LLM benchmarks and a blind evaluation by human doctors. Then, the recommended architecture for running these models on Databricks will be shared, in a way that complies with the strict security and privacy requirements of healthcare, while still enabling a fully managed, scalable, and highly performant environment.   Book a meeting with our Team: https://lnkd.in/dD_RBEtr #Databricks #LLMs #HealthcareLLMs #nocode #GenerativeAI #HealthcareAI #MedicalLLMs #nlp #naturallanguageprocessing #DataScience #DataEngineering #DataAnalysis #DataInsights #TextProcessing #DeepLearning #DataScience

3

Welcome to our new blog Series on Friendli Model Optimizer! Friendli Model Optimizer is a quantization tool that you can use to improve inference speed, lower resource consumption and maintain model accuracy. In this article, we’ll explore how to use the FMO library and understand its quantization feature. Stay tuned for part 2 of this blog, where we'll dive deeper into the performance analysis. Read Full blog 👉 https://lnkd.in/d-WU5MgJ Try friendli Suite 👉 https://suite.friendli.ai/ #FriendliAI #Friendli #FriendliModelOptimizer #LLM #Inference

14

Models like OpenAI's CLIP combine text and images for search, but they have limitations, particularly with longer texts and complex relationships. This article talks about JINA's new model that improves on CLIP by integrating both text and image embeddings in one model, making it faster and more efficient for multimodal search tasks. Experiments and Findings: → Search with Both Text and Images: Combining both embeddings leads to better and more diverse search results, improving relevance and variety in results. → Image-Driven Diversification: By using images for diversification, Jina-CLIP v1 can present a broader range of results, offering users more varied options. Results Comparison: → Using both text and image embeddings (via averaging) consistently outperformed searches based solely on text or images, providing more precise and diverse retrieval across both broad and fine-grained categories. https://lnkd.in/guNRQpBU

15

"Our topline metrics for llama 3 are - 97% speedup for Llama 3 8B inference using autoquant with int4 weight only quantization and hqq - 73% peak VRAM reduction for Llama 3.1 8B inference at 128K context length with a quantized KV cache - 50% speedup for Llama 3 70B pretraining using float8 training on H100 - 30% peak VRAM reduction for Llama 3 8B using 4 bit quantized optimizers. " https://lnkd.in/d8W4PUG5

1

Researchers from Salesforce AI and UIUC have introduced a workflow for Online Iterative Reinforcement Learning from Human Feedback (RLHF), which enhances the performance of large language models compared to traditional offline methods. This approach addresses the challenge of incorporating human feedback in open-source projects by using proxy models constructed from diverse open-source datasets to approximate human feedback, making it more accessible to resource-constrained open source projects.   The workflow involves several key components, including reward modeling, iterative policy optimization, practical implementation, and evaluation. Reward modeling constructs preference models using datasets like HH-RLHF and SHP, employing the Bradley-Terry model to approximate human preferences and connect RLHF with reward maximization. Iterative policy optimization continuously updates the policy based on new data, mitigating over-optimization issues associated with finite offline datasets and ensuring robust performance with out-of-distribution data. Practical implementation offers a comprehensive guide for the framework, featuring supervised fine-tuning (SFT) and iterative direct preference learning with a hybrid batch learning approach to balance exploitation and exploration.   The trained LLM, SFR-Iterative-DPO-LLaMA-3-8B-R, achieved state-of-the-art results on benchmarks such as AlpacaEval-2, Arena-Hard, MT-Bench, HumanEval, and TruthfulQA, outperforming larger models like GPT-3.5-turbo-1106. Additionally, an ablation study addressing length bias by incorporating a length penalty into the reward function effectively improved performance on length-control benchmarks. The approach integrates supervised fine-tuning with iterative RLHF, achieving state-of-the-art performance using fully open-source datasets.    Arxiv: https://lnkd.in/eQP8ivYt

4

This week in #TheBatch: 🎨 All about Claude’s Artifacts feature 🌱 AI growth hinders carbon emissions goals 🧠 GaLore, a memory-efficient method for fine-tuning and pretraining Plus: Andrew Ng shares his concerns about the proposed California regulation SB 1047. Read The Batch now: https://hubs.la/Q02G5YpM0

49

🤖 📄 Looking to build a highly reliable Summarization system using LLMs? Check out the guide we put out to see how you leverage DSPy to build a reliable Summarization system that fits your use case. https://lnkd.in/gnShmHaJ

6