Together AI

🎤 Together AI Rap – “Gold Chain Compute” 🪙💻💥 🎵 Listen here https://lnkd.in/grbhg6FE Yo, We in the cloud game, heavy with the flame, SemiAnalysis just certified the name. Gold chain hangin’, no need to explain, Together AI — we built to train brains. 🧠🔥 GPU clusters, we scale like a beast, Inference so fast, got your tokens on lease. Slurm on lock, kernel code tight, We do it in milliseconds — day or night. ⚡🌙 This the AI gold rush, stakes run deep, We mine pure performance — not just compute cheap. No smoke, no mirrors, no need to pretend, Just that gold-tier power from start to end. 💪🏅 So wear that chain, let the benchmarks shout, Together AI — now that’s what acceleration’s about. 🚀🎤

New deep-dive blog on preference tuning using DPO! We cover direct preference optimization (DPO), how and when to use it to shape models to human preference data. Also includes full code examples of preference tuning small models. Read below! Link: https://lnkd.in/gq6pDnRP

Introducing Chipmunk: Joint work w/ SandyResearch @ UCSD for training-free acceleration of Diffusion Transformers w/ attention/MLP step deltas! ⚡️ Up to 3.7x faster video and 1.6x faster image generation w/ dynamic column sparsity (while preserving Vbench quality)! 🚀 Open-source framework & CUDA kernels! 🔹 What is Chipmunk? Chipmunk accelerates Diffusion Transformers (DiT) without additional training through a combination of two techniques: (1) caching and (2) sparsity. Attention and MLP layers cache their outputs and subsequent steps of the diffusion process are reformulated to compute activation deltas against this cache.  🔹 Why does caching + sparsity work? DiT activations are naturally sparse and change slowly across diffusion steps. Chipmunk exploits this to cache previous step activations and compute sparse deltas against this cache directly within attention and MLP layers – up to 93% dynamic attention sparsity in HunyuanVideo! At Together AI, we’re always exploring the acceleration frontier to serve the highest quality models at the lowest cost! 📚 Read more: https://lnkd.in/dNN8T6NP 📝 In-depth blog: https://lnkd.in/dicJYXdX 🖥️ GitHub: https://lnkd.in/dAAB44_y

🎙️ New to streaming or looking to level up? Check out StreamYard and get $10 discount! 😍 https://lnkd.in/gFadPJSK

🔔 Epic sight on the floor of the @NYSE! The NYSE lit up its iconic boards to celebrate Together AI’s selection to the 2025 Forbes AI 50 list. Seeing Together AI spotlighted on Wall Street is pretty neat 😊 Thank you to the NYSE team and Brian J. Baumann for the shout‑out and to Forbes for recognizing our work. As The AI Acceleration Cloud, Together AI is turbocharging open‑source AI with world‑class research, purpose‑built GPU clusters, and lightning‑fast inference & fine-tuning. We're honored by the NYSE nod—and even more excited to work with customers and partners to accelerate their AI ambitions next. #ForbesAI50 #NYSE

We’re thrilled to be heading to ICLR 2025 in Singapore next week! ⚡️ Stop by our booth C04 to meet our Research & Recruiting teams! 📅  We will also be hosting a special Recruiting Networking Event on Saturday April 26, 6 – 8 PM. Whether you're looking to explore career opportunities or simply expand your network, we’d love to see you there! 🍸 Connect over food, drinks & career conversations 🔗 RSVP here as space is limited: https://lu.ma/j07gg3yk Our team will also be presenting papers throughout the whole week: • Dan Fu on ThunderKittens: Simple, Fast & Adorable Kernels – Thursday 4/24 at 3 PM • Pragaash P., James Liu, and Ben Athiwaratkun on Training-Free Activation Sparsity in Large Language Models – Saturday 4/26 at 10 AM • Avner May on Speculative Decoding - Saturday 4/26 at 3 PM • Ce Zhang, Ben Athiwaratkunu, Junlin W., Jue Wang, & James Liu on Mixture‑of‑Agents Enhances LLM Capabilities - Saturday 4/26 at 3 PM • Linda He, Jue Wang, Maurice Weber, Ben Athiwaratkun, and Ce Zhang on Scaling Instruction-tuned LLMs to Million-token Contexts via Hierarchical Synthetic Data Generation – Saturday 4/26 at 3 PM • Workshop on Modularity for Collaborative, Decentralized, and Continual Deep Learning by Max Ryabinin – Sunday 4/27 at 2:15 PM See you in Singapore! 🚀

🛠️ Together Fine-Tuning: Now with Preference Optimization and Continued Training! Your AI models shouldn’t be static—they should evolve with your users. With our new Fine-Tuning Platform, you can now build custom AI models that evolve over time—optimizing behavior based on user preferences and continuously improving with fresh data. The updated Together Fine-Tuning Platform includes: ✨ Direct Preference Optimization (DPO) for aligning models with user feedback 🔁 Continued Training from previous runs 🖥️ A brand-new browser UI for launching jobs—no code required Fine-tune leading open models like Llama, Gemma, and DeepSeek distilled versions at industry-leading prices—with no minimum spend and full control over model weights. 📖 Read the blog: https://lnkd.in/gvvtgW7u 🧪 Quickstart guide: https://lnkd.in/g4h3t3yc

Introducing Open Deep Research! A fully open-source Deep Research tool that: • writes comprehensive reports • does multi-hop search and reasoning • generates cover images & pod-casts! We’re releasing everything: evaluation dataset, code and blog.🔥 📝 Blog: https://lnkd.in/gMKSFGX2 🤗 Evaluation Dataset: https://lnkd.in/gECSKwwP 🖥️ GithHub: https://lnkd.in/ghj2QfbD Architecture We are releasing a deep-dive technical blog sharing practical tips and also the full codebase including all new features: - improved prompting and workflow implementation - cover images and pod-cast generation We also evaluated the effect of the number of search steps in our Deep Research, and the configurability of our framework allows users freely to adjust the research budget depending on their use cases. Our Deep Research tool can also output Mermaid JS charts to help with data visualization!

It's been a very exciting month at Together AI! Catch up on all the updates to our product, research and tools in our latest "Together We Build" issue 📰

🏆 Together AI has been named to the 2025 Forbes AI 50 — Forbes' annual list of the most promising privately-held companies using AI to transform industries and shape the future. This recognition is a testament to our momentum and mission: to build The AI Acceleration Cloud — turbocharged by cutting-edge AI research and engineered for massive scale 🌐 From high-performance GPU clusters to simple developer APIs for inference and fine-tuning, Together AI powers advanced models like DeepSeek 🐋 and Llama 🦙 We’re proud to be recognized in #ForbesAI50 And we’re just getting started 🚀