Karim Douieb

📢 Hugging face is directly integrated with GenAI developers favorite tool "langchain"🦜 Get set go!🚀 💡 pip install langchain-huggingface https://lnkd.in/ddHreVrs #huggingface #GenerativeAI #langchain

22

Want to test Meta, Google, and Mistral models all on one platform ? **Groq.com** Why Groq? Because speed matters. GroqCloud supports a diverse array of models, enabling you to select the best for your needs: ➡ llama3-8b-8192 by Meta (Context Window: 8,192 tokens) ➡ llama3-70b-8192 by Meta (Context Window: 8,192 tokens) ➡ mixtral-8x7b-32768 by Mistral (Context Window: 32,768 tokens) ➡ gemma-7b-it by Google (Context Window: 8,192 tokens) Groq's commitment to GenAI inference speed is transforming real-time AI applications with its LPU. What is an LPU? ➡ Language Processing Unit, is a revolutionary processing system designed specifically for LANGUAGE applications. Unlike traditional CPUs (Central Processing Units) and GPUs (Graphics Processing Units), which handle a broad range of tasks. LPUs are optimized for the rapid processing demands of LLMs This specialized architecture allows for more efficient handling of sequential data, significantly reducing the time per word processed. ➡ That means you get faster outputs You can run Llama-2 70B at over 300 tokens per second per user!

24

Kraken Update 🐙 : Kraken Model now Online on Hugging Face https://lnkd.in/e2miq2Jy Have fun testing it 🤗 ⏩ Experts of Kraken ⏺ defog-llama-3-sqlcoder-8b: Leading model for SQL generation – outperforms GPT-4 Turbo. ⏺ gorilla-llm-gorilla-openfunctions-v2-7b: Leading model for function callings – outperforms GPT-4. ⏺ ise-uiuc-Magicoder-S-DS-6.7B: Top Python model in its size. ⏺ VAGOsolutions-Llama-3-SauerkrautLM-8b-Instruct: Strongest Llama-3-8b Fine-Tune for the German language. #MachineLearning #GPT #LLM #Generativeai #OpenSource

14

2 Comments

Detecting hallucinations in large language models using semantic entropy https://lnkd.in/eP5bT8CM

2

LLM News Post : It’s here! Meet Llama 3, latest generation of models that is setting a new standard for state-of-the art performance and efficiency for openly available LLMs. Key highlights • 8B and 70B parameter openly available pre-trained and fine-tuned models. • Trained on more than 15T tokens, 7x+ larger than Llama 2's dataset! • Improved tokenizer with vocabulary of 128K tokens for better performance. • State-of-the-art performance across industry benchmarks. • New capabilities, including enhanced reasoning and coding. • 3x more efficient training than Llama 2. • New trust and safety tools with Llama Guard 2, Code Shield, and CyberSec Eval 2. • Integrated into Meta AI, and available in more countries across our apps. • And, just the beginning with more models and new capabilities coming soon! #meta #llm #ai

9

IMVC 2024 just released Dr. Amir Alush great talk about going from messy visual data to high quality datasets: https://lnkd.in/dGBDUUHQ . Try it yourself on your own data at https://lnkd.in/ds4vwrpP!

24

Interesting conversation between Yann LeCun and Pedro Domingos here. I am mainly aligned with Yann here. Do not forget that large language models are statistical models smartly implemented on computers. Hence, they will encode the information in their representation and generate outputs accordingly with the information encoded. They will generalize well *into the distribution* of our knowledge but, as they lack reasoning, conceptualization and semantic-qualia information, being just multiplications of matrices optimized to minimize some loss function, their out of distribution behaviour will not satisfy the expectations, because they are not trained to do so and their assumptions about our world are poor if any!!! My opinion may generate controversy but I will give it: We urgently need to differentiate between science fiction and science and read more statistics books and less speculative books. What is science and what is a myth. I like to speculate a lot!!! But I take for granted what theorems, proofs and technical papers say if it makes sense. We may achieve artificial general intelligence, with catastrophic hallucinations by the way, using our current approaches but in no way superintelligence. Will soon publish a paper where I will try to give solid arguments about all these things.

121

12 Comments

LONGEMBED : Benchmarking Extension of Embedding Models for Long-Context Retrieval In this paper, authors introduced a novel LONGEMBED benchmark comprising of two synthetic tasks and four carefully chosen real-world tasks, featuring documents of varying length and dispersed target information. 𝐍𝐨𝐭𝐚𝐛𝐥𝐞 𝐨𝐮𝐭𝐜𝐨𝐦𝐞𝐬 𝐨𝐟 𝐭𝐡𝐢𝐬 𝐩𝐚𝐩𝐞𝐫 𝐚𝐫𝐞: - observing effective extension of the context window of existing embedding models by several folds, regardless of their original context being 512 or beyond 4k. - For models employing absolute position encoding (APE), showing the possibility of harvesting further improvements via fine-tuning while strictly preserving original behavior within the short context. - For models utilizing Rotary Position embeddings (RoPE) substantial enhancements on LONGEMBED are observed when employing methods that fully leverage RoPE’s advantages, such as NTK and SelfExtend. 𝐋𝐢𝐦𝐢𝐭𝐚𝐭𝐢𝐨𝐧𝐬 𝐨𝐟 𝐄𝐱𝐢𝐬𝐭𝐢𝐧𝐠 𝐑𝐞𝐭𝐫𝐢𝐞𝐯𝐚𝐥 𝐁𝐞𝐧𝐜𝐡𝐦𝐚𝐫𝐤𝐬 - candidate documents not long enough -  target information to answer user query not uniformly distributed across the document as possible. 𝐓𝐚𝐬𝐤𝐬 𝐢𝐧 𝐋𝐨𝐧𝐠𝐄𝐦𝐛𝐞𝐝 - Synthetic tasks like Personalized Passkey Retrieval and Needle-in-a-haystack Retrieval - Real tasks for evaluations on long-form QA and summarization tasks across datasets like NarrativeQA , 2WikiMultihopQA , QMSum and SummScreenFD 𝐂𝐨𝐧𝐭𝐞𝐱𝐭 𝐖𝐢𝐧𝐝𝐨𝐰 𝐄𝐱𝐭𝐞𝐧𝐬𝐢𝐨𝐧 𝐟𝐨𝐫 𝐀𝐏𝐄-𝐛𝐚𝐬𝐞𝐝 𝐌𝐨𝐝𝐞𝐥𝐬 Strategies used: - 𝐏𝐚𝐫𝐚𝐥𝐥𝐞𝐥 𝐂𝐨𝐧𝐭𝐞𝐱𝐭 𝐖𝐢𝐧𝐝𝐨𝐰𝐬 (𝐏𝐂𝐖) :  PCW divides the long document into multiple short chunks, processes each chunk in parallel, and aggregates their results - 𝐆𝐫𝐨𝐮𝐩𝐞𝐝 𝐏𝐨𝐬𝐢𝐭𝐢𝐨𝐧𝐬 (𝐆𝐏) & 𝐑𝐞𝐜𝐮𝐫𝐫𝐞𝐧𝐭 𝐏𝐨𝐬𝐢𝐭𝐢𝐨𝐧𝐬 (𝐑𝐏) : position reorganization accommodates longer context by reusing the original position ids. - 𝐋𝐢𝐧𝐞𝐚𝐫 𝐏𝐨𝐬𝐢𝐭𝐢𝐨𝐧 𝐈𝐧𝐭𝐞𝐫𝐩𝐨𝐥𝐚𝐭𝐢𝐨𝐧 (𝐏𝐈) : introduces position embeddings via linear interpolation of existing ones. 𝐂𝐨𝐧𝐭𝐞𝐱𝐭 𝐖𝐢𝐧𝐝𝐨𝐰 𝐄𝐱𝐭𝐞𝐧𝐬𝐢𝐨𝐧 𝐟𝐨𝐫 𝐑𝐨𝐏𝐄-𝐛𝐚𝐬𝐞𝐝 𝐌𝐨𝐝𝐞𝐥𝐬 Strategies used: - 𝐒𝐞𝐥𝐟 𝐄𝐱𝐭𝐞𝐧𝐝 (𝐒𝐄): re-introduces normal relative positions within the nearest neighbor window. - 𝐍𝐓𝐊-𝐀𝐰𝐚𝐫𝐞 𝐈𝐧𝐭𝐞𝐫𝐩𝐨𝐥𝐚𝐭𝐢𝐨𝐧 (𝐍𝐓𝐊): scales high frequencies less and low frequencies more to spread out the interpolation pressure across multiple dimensions 𝐑𝐞𝐬𝐮𝐥𝐭𝐬 - APE based models: Plug and play methods like GP, RP, LPI and PCW strategies yield comparable results with no significant disparities. - RoPE-Based Models: RoPE-specific methods including NTK and SE yield significant improvements for both models across all datasets, surpassing PCW, PI and GP by a large margin. 🏆 For my detailed analysis of this paper, please refer to my blogpost on it at: https://lnkd.in/dRMbdH39 #generativeai #largelanguagemodels

15

Python Tutorial: Use TensorFlow to Generate Predictive Text https://buff.ly/444Uh1K

2

Exciting news ❤️! I am proud to finally announce the publication of the « 𝐅𝐢𝐫𝐬𝐭-𝐞𝐯𝐞𝐫 𝐓𝐮𝐧𝐢𝐬𝐢𝐚𝐧 𝐒𝐢𝐠𝐧 𝐋𝐚𝐧𝐠𝐮𝐚𝐠𝐞 𝐃𝐚𝐭𝐚𝐬𝐞𝐭 ». This dataset includes hand gestures that represent common words of Tunisian sign language. It is the first ever dataset of TunSL, it was collected in collaboration with Tunisian Association of Sign Language Interpreters (ATILS) as one of the tasks for the developement of a mobile app for the recognition of TunSL with my colleague Rihab Jerbi under the supervision of Mr. Riadh Abdelfattah at SUP'COM. We believe that this dataset will be an important resource for the study and understanding of TunSL, and we hope that it will pave the way for further research and development in this field. We would like to thank the participants (especially our colleague Hazem Bouaziz) who helped make this dataset possible. The dataset is available for use by researchers and developers. For more information please contact us (& don't forget to star the repo ⭐) 🆕Link to Kaggle dataset: https://lnkd.in/eWRhcT2G 🆙Link to 𝐌𝐞𝐧𝐝𝐞𝐥𝐞𝐲 𝐃𝐚𝐭𝐚 Mendeley: https://lnkd.in/gqkHpF5u 🆙Link to the GitHub repo: https://lnkd.in/eE-yUgW8 🆙Link to the report on ResearchGate: https://lnkd.in/dT2fUqae 🆙Link to the project's post : https://lnkd.in/efnufzbN #signlanguage #TunisianSignLanguage #SignLanguageResearch #dataset #linguistics #data #computervision #artificialintelligence #collaboration #research #keepworking #keepwinning

45

11 Comments

Your Contribution for the start of the week 📆: botpress 👇 Co-founded by Sylvain Perron and Justin Watson, Botpress allows developers to build chatbots without a team of data scientists and machine learning experts. Here’s why you should contribute to this repo: 1️⃣ By contributing to botprocess, you can dive into complex bot development using the Botpress SDK/CLI and your contributions have the chance of positively impacting the experience of hundreds of thousands of developers. Since maintainers take extra care in getting back to you (with a ~100% reply rate and an average of 1 day to answer your queries), your contributions will get the attention they deserve. 2️⃣ Nearly 50% of contributors submit a second PR. This means the repo has a supportive environment for building upon past contributions. If you enjoy working on this repo, chances are you can continue to work on it! ⚡️ 3️⃣ Botpress is an accessible repo. At Quira, we rank Developers by their DevRank (more on this here: https://lnkd.in/eNtrs97F). We found that the median developer working on the repo belongs to the ~85th percentile of GitHub developers. In comparison, many of the open source companies we reported on have had a median contributor level of ~95th percentile. This means that you have a real opportunity to contribute. ⚡️ You can check more of replicate/cog's open-source stats on Quira (https://lnkd.in/efemmXwH) or access the GitHub page https://lnkd.in/eD8SdZBH 📖

8

Contextual Position Encoding : Learning to Count What’s Important One of the most insightful papers I read this week ! Having worked on the LLM space previously, I have seen how transformers struggle with position. With standard independent positional embedding, if two words have the same context, there attention will depend solely on their position embedding which is either sinusoidal or a sequence. The paper brings forward the idea : what if positions were dependent on context rather than being completely independent ! Which means that the position depends on a non linear function of the keys and the query vectors. Still grasping the entire idea, but potentially this should be a turning point for transformers both in the NLP and Vision Space.

24

2 Comments

LlamaIndex + MLflow RAG contains a lot of parameters to tune, from chunking to indexing - everything affects downstream answer accuracy. It’s important to have a systematic approach to track and tune these parameters, and define the right eval metrics/dataset to evaluate changes. Check out Jino Rohit's video on experimenting with a LlamaIndex pipeline with MLflow for experiment tracking. https://lnkd.in/gyYXfyyb

553

14 Comments

A method to generate predictor circuits for the classification of tabular data https://buff.ly/3WSXjEw

5

Llama 3 by Meta was just released last week .. and it is already on the Chatbot Arena leaderboard 🏆. The battle for the best (open ?) LLM continues stronger than ever ! The 𝐂𝐡𝐚𝐭𝐛𝐨𝐭 𝐀𝐫𝐞𝐧𝐚 is a crowdsourced open platform for blind comparisons of LLMs, developed by LMSYS. It is one of the most popular ways to evaluate LLMs, and companies releasing new models thrive to reach the top of the leaderboard https://lnkd.in/g_sWj5pF. 𝐇𝐨𝐰 𝐝𝐨𝐞𝐬 𝐢𝐭 𝐰𝐨𝐫𝐤 ? ▶ The approach is very simple: users enter prompts, and the platform generates responses from two different LLMs, then users chose the best answer, without knowing which model generated which answer ▶ Scores are calculated with the Elo ranking systems ▶ They have collected hundreds of thousands of votes so far 𝐖𝐡𝐲 𝐢𝐬 𝐭𝐡𝐢𝐬 𝐩𝐨𝐰𝐞𝐫𝐟𝐮𝐥 ? ▶ Realistic Evaluations: traditional benchmarks are not as helpful as they might seem because they can be manipulated and tailored for optimization.. with the Chatbot Arena, LLMs are evaluated through real-world user interactions for more realistic evaluations ▶ The data is always fresh = no risk of data contamination, as compared to static benchmarks ▶ Continuous Feedback: Rankings are updated based on ongoing user votes, to ensure up-to-date performance metrics ▶ Transparent Process: the Elo rating system is a clear and explainable method 𝐒𝐨𝐦𝐞 𝐥𝐢𝐦𝐢𝐭𝐚𝐭𝐢𝐨𝐧𝐬 𝐨𝐟 𝐭𝐡𝐞 𝐂𝐡𝐚𝐭𝐛𝐨𝐭 𝐀𝐫𝐞𝐧𝐚 ▶ Short user interactions might not reveal more advanced model capabilities ▶ Lack of diversity and complexity in the prompts ▶ Crowdsourcing is not representative: it is likely that most of the platform users are in a similar group/race/gender, so their preferences might not fairly reflect the entire population ▶ Users tend to be biased towards more verbose responses 👌 Overall, I love this platform, and it is very user-friendly so if you haven't yet tried it, give it a go! -- #AI #GenAI #Llama3 #LLMs #ChatbotArena

10

2 Comments

Impressive results. It would be interesting to see how it performs compared to GPT4/Claude 3 Opus-level LLMs.

2

This tutorial is perfect for anyone looking to deepen their understanding of combining vision and language models in practical applications (VQA: Visual Question Answering Models)

6

2 Comments

Hey everyone ! I've fine-tuned a model using my own dataset to differentiate between AI-generated and real images. Check out the demo and model here: Hugging Face model : https://lnkd.in/d9qg3Gas

25

2 Comments