Activité
3 k abonnés
-
Xavier Fischer a republié ceciHad a great time representing Anyshift at DevOps Day Dallas ! 💫Xavier Fischer a republié ceciAnyshift was proud to be at DevOps Day Dallas! 🤠 Here’s a quick recap of our session: 🛠 Topic: "5 Reasons You're Struggling to Debug Your Infrastructure in Under an Hour" Key takeaways: - 👁️ Lack of Centralized Visibility: Siloed systems make it tough to get the full picture of your infra. - 🕰️ Scattered Historical Data: Tracking changes across environments is a mess. - 📍 Weak Impact Analysis: You still struggle to see how changes affect your entire infrastructure. There’s always a shiver before applying your change to production. Bottom line: Between your cloud, K8s, data providers, Cloudflare, DNS, and more—your single source of truth is fractured. 🌐 Learn how Anyshift helps you confidently deploy changes by answering crucial questions like: - Where is this resource defined? - What depends on it? - Who owns it? 📽 Watch the full replay here! https://lnkd.in/d-MdYbgq
-
Xavier Fischer a partagé ceciIA: Apple rachète la start-up parisienne DatakalabIA: Apple rachète la start-up parisienne Datakalab
-
Xavier Fischer a partagé ceciXavier Fischer a partagé ceciGet a glimpse of images generated by our fully quantized, Stable Diffusion model running on NVIDIA GPUs using TensorRT. Not only is this model 4x smaller, but it's also faster than its full precision equivalent! ( 👏 Thomas Lespargot ) #AI #MachineLearning #DiffusionModels #Quantization #TensorRT #NVIDIA
-
Xavier Fischer a partagé ceciXavier Fischer a partagé ceciAnother week, other newS on LLM compression ! This week we are happy to showcase our results on quantizing Dolly v2 (Databricks) in 4 bits. You will find links to the papers and method used in comment. If you're eager to learn more or put our cutting-edge quantization methods to the test, don't hesitate to get in touch with us.
-
Xavier Fischer a partagé ceciXavier Fischer a partagé ceciExciting news for the large language model (LLM) community! 🎉 Datakalab is proud to unveil our groundbreaking methods for LLM quantization. We've successfully achieved quantizing a 13 billion parameter OPT down to 4 bits using our proprietary techniques, all without sacrificing accuracy! (tested on Hellaswag, OpenbookQA, ARC-e, ARC-c, Winogrande, PiQA, BoolQ) 🚀 In the table below, we showcase two of our innovative methods: Datakalab (1) and Datakalab (2). The former leverages sparse binary matrices, while the latter employs INT 4 weights exclusively. 💥 Both techniques yield impressive results on their own, but the magic truly happens when they're combined! 💡 If you're eager to learn more or put our cutting-edge quantization methods to the test, don't hesitate to get in touch with us. We're always happy to help you unlock the full potential of your AI models! 📩 Contact us to discover the future of AI optimization with Datakalab's compression breakthroughs.
-
Xavier Fischer a partagé ceciProud to see our R&D recognized ICLR 2023. If you need to optimize LLM deployment - contact us !Xavier Fischer a partagé ceciIf you are in Kigali, Rwanda - Datakalab's team is happy to present its latest achievement in compression at ICLR 2023. The method that we are showcasing allows low bit quantization of all kinds of standard networks from tinyML to LLM and diffusion models ! Edouard Yvinec arnaud dapogny Link to the paper in the first comment.
-
Xavier Fischer a partagé ceciHappy to see that our techniques are used and featured by Intel Corporation in their Neural compressor tool !Xavier Fischer a partagé ceciDatakalab team is happy to see its work featured and leveraged by Intel Corporation inside the Intel Neural compressor. SPIQ (Data Free Per Channel Input Quantization), method created by Datakalab, is used by Intel to compress the biggest LLM. Our goal there was to fix the issues of outliers. It is great to see that Intel engineers have recognized the value of Datakalab technology and are incorporating it into its compression solutions for the new generation of models. SPIQ was published last year at WACV. If you want to learn more about our latest developments, which are even more promising - come and speak to us!GitHub - intel/neural-compressor: SOTA low-bit LLM quantization (INT8/FP8/MXFP8/INT4/MXFP4/NVFP4) & sparsity; leading model compression techniques on PyTorch, TensorFlow, and ONNX RuntimeGitHub - intel/neural-compressor: SOTA low-bit LLM quantization (INT8/FP8/MXFP8/INT4/MXFP4/NVFP4) & sparsity; leading model compression techniques on PyTorch, TensorFlow, and ONNX Runtime
-
Xavier Fischer a partagé ceciExciting! Datakalab was recently mentioned in The Wall Street Journal article discussing the importance of prioritizing privacy in the deployment of AI-based technologies for the Olympics - Paris 2024 - Comité d'organisation des Jeux Olympiques et Paralympiques de 2024. At Datakalab, we're dedicated to developing solutions that provide valuable insights while ensuring that individual privacy is never sacrificed. In 2021, our hard work to enact French law allowing for the deployment of private by design AI based video analysis solutions was successful but also underscored a wider market need for neural network compression at scale. We firmly believe that for everyone to benefit from AI, and not just 5 global companies, there is a multi billion dollar need for plug-and-play neural network compression solutions that allow organizations large and small to effectively deploy AI on the edge while ensuring privacy, optimizing energy consumption or bandwidth and increasing algorithm accuracy. This approach allows for data to be processed at the sensor level - and never sent to 3rd party centralized servers or data lakes - thus ensuring that everyone benefits from AI without sacrificing privacy. Thank you to Catherine Supp for featuring us in her thought provoking article. We're committed to continuing our work in promoting responsible AI and safeguarding privacy. Stay tuned for more updates soon and reach out if you have any questions on what we do! #ai #privacy #quantization #pruning #edgePrivacy Concerns Surround Plans for AI Camera Surveillance at 2024 Paris OlympicsPrivacy Concerns Surround Plans for AI Camera Surveillance at 2024 Paris Olympics
-
Xavier Fischer a partagé ceciXavier Fischer a partagé ceciHappy to be presenting our products and research at the tinyML Foundation summit! If you are there or in the Bay Area pass by our booth . #research #tinyml23 #tinyml #onnx #edge Xavier Fischer Lucas Fischer Edouard Yvinec
Xavier Fischer a commenté un post
1 ans
Congrats Sam!!!
-
Xavier Fischer a aimé ceciXavier Fischer a aimé ceciAn internal version of Astra, OpenAI’s next major model class, solved 10 major open problems in mathematics, quantum complexity, and theoretical computer science. The cost of generating the proofs for all 10 of these breakthroughs combined was under $2,000 at Sol API prices. We’re excited to see what scientists and researchers are able to create with our upcoming Astra models! https://lnkd.in/gAca354A
-
Xavier Fischer a aimé ceciXavier Fischer a aimé ceciPascal Weinberger , Co-CEO of Owkin, joins the #GlobalInvestmentSummit2026 to explore how advanced artificial intelligence is reshaping biomedical research and accelerating drug discovery, drawing on experience spanning innovation, research, and entrepreneurship. Paris | 1–2 September 2026 Register now and join leading figures in investment and innovation #GIS2026
-
Xavier Fischer a aimé ceciXavier Fischer a aimé ceci🚀 Not a point solution. A complete, clinically validated X-ray AI suite, from image analysis and triage to the draft report. The Rayvolve AI Suite is one of the broadest CE-marked, peer-reviewed X-ray AI offerings. Backed by a recent multicenter study published in Diagnostics, drawing on 21,581 X-rays from 52 sites in 20 countries across 5 continents, the new scope extension strengthens both AZtrauma and AZchest. Rather than stitching together separate AI tools, the suite unifies a broader clinical range within one workflow, all the way to the draft report. AZtrauma extends beyond fractures, dislocations, and joint effusions to include osteoarthritis, tumoral, and degenerative lesions. AZchest builds on its core thoracic findings to cover 13 abnormalities in total. With Rayvolve, additional findings can now be addressed through a single, consistent approach, helping imaging teams analyze more X-rays in more settings. For more information, visit azmed.co #AZtrauma #AZchest #RadiologyAI #MedicalImaging
-
Xavier Fischer a aimé ceciXavier Fischer a aimé ceciAnd another one ☝️ In a world where agents do the work, all we need is big ambition and great food. Amazed by what people are building in SF. We’re continuing our “Taste & Decision in the Age of AI” dinner series. Reach out for the next one.
-
Xavier Fischer a aimé ceciXavier Fischer a aimé ceciWhat should move to the top of the remediation queue? The credential with the strongest risk signals or the resource connected to the greatest downstream impact? Anyshift.io's Louis Fradin explores how those two views strengthen prioritization. GitGuardian surfaces exposed credentials and risky machine identities. Anyshift traces the services and applications that depend on the resources they reach. Together, credential context and infrastructure topology provide a clearer picture of what demands attention first. Read the full analysis: https://lnkd.in/eXpCpEmV #SecretsSecurity #NonHumanIdentity #DevSecOps
-
Xavier Fischer a aimé ceciXavier Fischer a aimé ceciNEW INVESTMENT | Quadrille Capital is pleased to participate in the $100M Series B of TwelveLabs, co-led by New Enterprise Associates (NEA) and NAVER Ventures. Amazon, Radical Ventures, Korea Investment Partners, Index Ventures and Red Bull Ventures also participated in the round. 🎥TwelveLabs builds video-native, multimodal AI (its Marengo and Pegasus foundation models, available on Amazon Bedrock) that lets enterprises search, analyze and reason about any video at scale. With this round the team moves from foundation models to a full-stack, agentic video intelligence system. Congratulations to Jae Lee and the entire team. Proud to support you on the road to video superintelligence! Find out more 👉https://lnkd.in/e744HfKr CC: Vianney Barre, Edward Colby, Adrien Galand, Tiffany Luck, Mason Murray #AI #VideoAI #VentureCapital
-
Xavier Fischer a aimé ceciXavier Fischer a aimé ceciHappy birthday to our French icon Yann LeCun 😁 Good to take the opportunity of RAISE summit to catch up from robotics to AI companies Steeve Morin Nicolas Essayan Drysdale Ventures Hadrien de Cournon
-
Xavier Fischer a aimé ceciXavier Fischer a aimé ceciUse case #3 Anyshift Graph API: find who keeps changing production config before it breaks something Production config changes come from everywhere: kubectl edits, CI pipelines, operators, Terraform and hotfixes. Most stay invisible until an incident starts, so we used the Anyshift Graph API to: -🎯 See every configuration change in one place -🎯 Identify the resource creating the churn -🎯 Understand when it started and what changed So engineering teams catch risky config churn before it becomes an incident, without digging through audit logs. ⏱️
Expérience et formation
-
Drysdale Ventures
Voir toute l’expérience de Xavier
Découvrez son poste, son ancienneté et plus encore.
ou