Antonio Bray’s Post

🚀 Breakthrough in Audio Transcription Speed! 🎙️ Imagine transcribing a feature-length movie in less time than it takes to make a cup of coffee. That's now possible with 'insanely-fast-whisper', a game-changing GitHub project. Key highlights: • Transcribe 2.5 hours of audio in just 98 seconds • Works locally on Mac or Nvidia GPUs • Combines Whisper + Pyannote for rapid transcription and speaker segmentation For the tech-savvy, here's a quick setup: 1. pip install insanely-fast-whisper 2. Run with your file and settings This tool isn't just fast—it's revolutionizing how we process audio data. Think about the implications for: • Journalists transcribing interviews • Researchers analyzing focus groups • Content creators captioning videos What would you do with local and near-instant transcription? Share your ideas below! 👇

pip install insanely-fast-whisper insanely-fast-whisper --file-name <FILE NAME or URL> --batch-size 2 --device-id mps --hf_token <HF TOKEN>

You can now transcribe 2.5 hours of audio in 98 seconds, locally. A new implementation called insanely-fast-whisper is blowing up on Github. It works on works on Mac or Nvidia GPUs and uses the Whisper + Pyannote library speed up transcriptions and speaker segmentations. Here's how you can use it: pip install insanely-fast-whisper insanely-fast-whisper --file-name <FILE NAME or URL> --batch-size 2 --device-id mps --hf_token <HF TOKEN> ♻️ Repost this if you found it useful. ↓ Are you technical? Check out https://AlphaSignal.ai to get a daily summary of breakthrough models, repos and papers in AI. Read by 200,000+ devs.

🚀 NVIDIA launches Llama-3.1-Nemotron-70B! https://lnkd.in/efahSbKW 📊 As of Oct 1, 2024, it's topping the charts: • Arena Hard: 85.0 • AlpacaEval 2 LC: 57.6 • MT-Bench: 8.98 🔑 Key points: • Outperforms other models across multiple benchmarks • Longer responses (avg 2199.8 characters) • Consistent performance (narrow confidence interval) 𝗛𝗲𝗿𝗲 𝗮𝗿𝗲 𝘀𝗼𝗺𝗲 𝗶𝗻𝗳𝗼𝗿𝗺𝗮𝘁𝗶𝗼𝗻 𝗮𝗯𝗼𝘂𝘁 𝘁𝗵𝗲 𝗘𝘃𝗮𝗹 𝗺𝗲𝘁𝗿𝗶𝗰𝘀 𝘂𝘀𝗲𝗱 𝑨𝒓𝒆𝒏𝒂-𝑯𝒂𝒓𝒅 is a challenging AI benchmark created from real user queries on Chatbot Arena. It uses the BenchBuilder pipeline to select 500 high-quality, diverse prompts that test language models on complex, real-world tasks across various domains. The process involves: * Question Source: Real user queries from Chatbot Arena (initially 200,000) * Question Selection Process:    - BenchBuilder pipeline filters and evaluates queries    - AI (GPT-4-Turbo) scores questions on 7 key qualities:    - Specificity    - Domain knowledge    - Complexity    - Problem-solving    - Creativity    - Technical accuracy    - Real-world relevance    - Topic modeling clusters similar queries    - High-quality clusters are selected    - 500 challenging prompts sampled for final benchmark Evaluation uses an LLM-as-judge approach, comparing model outputs to a baseline. This method provides a more comprehensive, updatable, and cost-effective evaluation than previous benchmarks, better separating top models and aligning well with human judgments. Arena-Hard Paper: https://lnkd.in/eh_hK7NK Github: https://lnkd.in/e9qBbd3E #NVIDIA #AI #LLM #TechInnovation #LLM

NVIDIA has silently released its new open-source model, Llama-3.1-Nemotron-70B, which is reported to surpass OpenAI's GPT-4o and Anthropic's Claude 3.5 in many crucial benchmarks with relatively lesser parameters Key Features of Llama-3.1-Nemotron-70B: - Can answer How many r's are in strawberry? accurately 😅 - Performance: The model has performed well on several alignment benchmarks, for example: - Arena Hard: 85.0 - AlpacaEval 2 LC: 57.6 - GPT-4-Turbo MT-Bench: 8.98 These scores show that the Nemotron model does not only outperform but also beats the performance of its larger counterparts, which have more parameters. - Architecture: The Nemotron model is built on the Llama 3.1 framework and uses transformers to generate coherent responses. It has relatively modest 70 billion parameters but is quite efficient and capable in the processing of user inquiries. - Open Source Availability: The model and its corresponding reward model with its training dataset are open sourced and are available on the Hugging Face so there's a scope for us to experiment with the model - RLHF (Reinforcement Learning from Human Feedback) : This model was trained on RLHF techniques that relied on the REINFORCE algorithm to optimize the best performance based on human preference. If you are interested in exploring or trying out this model further, preview the official website of NVIDIA. #AI #NVIDIA #Llama #MachineLearning #OpenSource #Innovation https://lnkd.in/gd-_p_rP

Is having a background in AI imperative to get into Generative AI? Maybe Not.. Maybe not if you want to build an RAG pipeline. But most definitely yes if you want to Fine Tune LLMs. I consider myself lucky to have that background, so I was able to transition seamlessly into the task of Fine Tuning models but honestly, most work you might do with LLMs would involve building an RAG pipeline with Prompt Engineering. With Autogen, it has become even more easier to setup a RAG workflow. So do you need a background in AI to work with LLMs? - YES - if you need to Fine Tune models - NOT NECESSARILY - If you want to build an RAG pipeline. PS: If you have an NVIDIA GPU>= 4070 powering your laptop, you can just about Fine Tune your own LLM models using PEFT. Or use Google Collab to try it out. #llm #rag

Ever imagine an AI that could not only chat with you about Shakespeare’s sonnets but also critique the artistry in your vacation photos? Enter NVIDIA’s latest marvel, the NVLM-D-72B model, a veritable Swiss Army knife in the AI world, blending text and visuals with the ease of a seasoned bartender mixing your favorite cocktail 🍹. NVIDIA’s brainchild, the NVLM-D-72B, isn’t just smart—it’s a smart aleck that can understand pictures and words together 📸📖. Picture this: an AI that can look at a doodle and not only tell you what it sees but also spin up a story about it! Trained on NVIDIA’s ultra-sophisticated Megatron-LM and cozying up with Hugging Face, this model is like having a supercomputer in your pocket, only it doesn’t weigh you down. Whether you’re looking to spice up your meme game or develop an app that finally understands fashion advice from a picture, NVLM-D-72B is your go-to 🚀. So, if you’re ready to ride the wave of AI innovation, NVIDIA’s got your ticket to the future, and it’s first class! Dive into the details on their Hugging Face page, and let’s get this techno-party started! 🎉 For a deeper dive and more giggles, check out the full tech extravaganza at NVIDIA's Hugging Face space: #AI #MachineLearning #DataScience #NVIDIA #Technology #Innovations #DeepLearning #AITransformation

The Biggest problem in AI is today getting GPU for LLM, below formula is handy for this:- 🚀 **Understanding GPU Memory Requirements for AI Models** 🚀 The formula is simple: **M** = (P*4B)*1.2/(32/Q) Where: - **M**: GPU memory required (in GB) - **P**: Number of parameters in the model (e.g., a 7B model has 7 billion parameters) - **4B**: Bytes used per parameter (4 bytes) - **32**: Total bits in 4 bytes - **Q**: Bits used to load the model (e.g., 16 bits, 8 bits, or 4 bits) - **1.2**: Accounts for a 20% overhead for additional memory usage 🔍 **Example**: For a model with 7B parameters loaded with 16-bit precision, you can estimate the GPU memory required using this formula. It's an essential consideration when planning your AI infrastructure! 💡 #AI #MachineLearning #DeepLearning #GPUMemory #ModelOptimization

𝐍𝐕𝐈𝐃𝐈𝐀 𝐨𝐮𝐭𝐩𝐞𝐫𝐟𝐨𝐫𝐦𝐬 𝐆𝐏𝐓4𝐨 𝐚𝐧𝐝 𝐒𝐨𝐧𝐧𝐞𝐭 3.5 NVIDIA has released a fine-tuned version of Llama 3.1 70B that outperforms both OpenAI's GPT4o and Anthropic's Sonnet 3.5 on multiple benchmarks. This model was trained using Reinforcement Learning from Human Feedback (RLHF), Llama-3.1-Nemotron-70B-Reward and HelpSteer2-Preference prompts on a Llama-3.1-70B-Instruct model as initial policy. NVIDIA models are now available worldwide under a Llama 3.1 Open Source license on Hugging Face, and according to it, the Llama-3.1-Nemotron-70B-Reward-HF model is "#1 on all three automatic alignment benchmarks, edging out strong frontier models such as GPT-4o and Claude 3.5 Sonnet." as o October 1st. 🔗 Hugging face repository: https://lnkd.in/dqexeux6 🔗 Paper: https://lnkd.in/db7YfEDP

AI Reading for Thursday March 14, 2024! Cerebras shows giant AI training chip, betting on bifurcation of GPU market into separate training and inference chips. - SiliconANGLE https://lnkd.in/e7kDyANc Nvidia to talk about new H200 chips with more memory, upcoming Blackwell architecture at GTC conference next week - Spiceworks OpenAI briefly posted hints at GPT 4.5 coming late summer 2024. - Android Authority A profile of Scott Wu, creator of Devin - Analytics India Magazine Black Nazis and Vikings are back, this time via Adobe Firefly, and the outrage machine snowflakes need a safe space from AI - Mail Online Demo of Figure Humanoid Android now with OpenAI - YouTube Watch Anymal robot dog run an obstacle course, climb rubble. Kurzweil and Musk talk about the singularity, and when we'll have more intelligence and computational capacity in data centers than all human brains - The Independent Testing an AI's moral compass by giving it 50,000 trolley problems - Ars Technica Making AI-generated YouTube video content spam targeted at kids for fun and profit - WIRED Satya Nadella says Google should have been 'default winner' in AI - Business Insider Softbank eyes stake in Mistral NYT responds to OpenAI: no 'hacking' here, just OpenAI ignoring our paywalls and IP rights - Ars Technica Akaso night-vision binoculars turn night into day with AI. For all your peeping-Tom adventures j/k! - Yanko Design - Modern Industrial Design News A copilot for your bike to record and warn you about your surroundings and possible close calls - Ars Technica Fabric adds LLM intelligence to your command line, lets you complete many daily tasks through a prompt library.