Seedream 4 passed this prompt test pretty well, only missing the tear in the gold backdrop.

You can now run Ovi by @character_ai Simultaneously generate both audio & video in under 40 seconds replicate.com/character-ai/o… Another @PrunaAI collab to deliver the fastest speeds possible

𝗜’𝗺 𝘀𝘁𝗮𝗿𝘁𝗶𝗻𝗴 𝗮 𝗻𝗲𝘄 𝘀𝗲𝗿𝗶𝗲𝘀: 𝗣𝗮𝗽𝗲𝗿 𝗛𝗶𝗴𝗵𝗹𝗶𝗴𝗵𝘁𝘀! I’ll share research papers about (efficient) AI I’ve read including their code when available. 𝗣𝗮𝗽𝗲𝗿 𝗛𝗶𝗴𝗵𝗹𝗶𝗴𝗵𝘁 #𝟬𝟭: “Olica: Efficient Structured Pruning of Large Language Models without Retraining” | 𝗔𝘂𝘁𝗵𝗼𝗿𝘀: Jiujun He, Huazhen Lin | 𝗩𝗲𝗻𝘂𝗲: @icmlconf 2025 This paper explores how to efficiently prune large language models 𝘄𝗶𝘁𝗵𝗼𝘂𝘁 𝗳𝘂𝗹𝗹 𝗿𝗲𝘁𝗿𝗮𝗶𝗻𝗶𝗻𝗴. Key contributions: • 𝗦𝘁𝗿𝘂𝗰𝘁𝘂𝗿𝗲𝗱 𝗽𝗿𝘂𝗻𝗶𝗻𝗴 𝘄𝗶𝘁𝗵 𝗳𝗮𝘀𝘁 𝗣𝗖𝗔: They analyze matrix products in the multi-head attention (MHA) and remove neurons with the lowest importance scores. • 𝗙𝗮𝘀𝘁 𝗿𝗲𝗰𝗮𝗹𝗶𝗯𝗿𝗮𝘁𝗶𝗼𝗻 𝘄𝗶𝘁𝗵 𝗳𝗲𝘄 𝗱𝗮𝘁𝗮 𝘀𝗮𝗺𝗽𝗹𝗲𝘀: Residual errors are compensated via a low-rank decomposition, requiring only a small calibration dataset. • 𝗣𝗿𝗮𝗰𝘁𝗶𝗰𝗮𝗹 𝗲𝗳𝗳𝗶𝗰𝗶𝗲𝗻𝗰𝘆: An LLM can be pruned and recalibrated in ~7 minutes using just hundreds of samples, resulting in smaller and faster models. 𝗣𝗲𝗿𝘀𝗼𝗻𝗮𝗹 𝗛𝗼𝘁 𝘁𝗮𝗸𝗲: This does not feel like zero retraining because of the recalibration phase but it makes retraining very efficient!

🚀 We offer tech croissants and pizzas in post-holiday Paris | Wednesday 17 September 🔥 Join AI developers, ML engineers, and researchers for an evening of cutting-edge optimization insights and practical implementation strategies! This meetup is built for practitioners scaling LLMs, fine-tuning diffusion models, and deploying real-time inference pipelines. 📅 What's Coming: 🎯 7:00 PM: Welcome & Networking 🧠 7:30-8:15 PM: Technical Sessions 🍕 8:15-9:00 PM: Deep Tech Conversations & Food • James Martin (CEO, Better Tech): "AI's Impacts: Solving Real Problems" • Jules Belveze (Software Engineer, Dust): "Early Exiting: Under-Hyped Compression Methods" 🎤 Speaking Opportunity: Submit your talk proposal - we want YOUR breakthrough insights! 📍 Location: Paris 9th (exact venue shared with approved attendees) ⚡️ Limited spots available: lu.ma/i24gy980

The much anticipated image editing model from Qwen is now on Replicate replicate.com/qwen/qwen-imag… Edit images in just 3 seconds, for $0.03 per image. We've worked with Pruna to deliver you the fastest way to use Qwen Image Edit.

The much anticipated image editing model from Qwen is now on Replicate replicate.com/qwen/qwen-imag… Edit images in just 3 seconds, for $0.03 per image. We've worked with Pruna to deliver you the fastest way to use Qwen Image Edit.

The much anticipated image editing model from Qwen is now on Replicate replicate.com/qwen/qwen-imag… Edit images in just 3 seconds, for $0.03 per image. We've worked with Pruna to deliver you the fastest way to use Qwen Image Edit.