Nacho Martínez Rincón

AI agents are only as powerful as their memory. Redis has been named the #1 data storage choice for AI agents in 2025. The reason? Redis delivers what AI truly needs: - Real-time vector storage - Semantic caching - Long-term memory for LLMs At Bassirah, we see this as validation of why Redis is central to the next generation of AI-native data platforms powering everything from agentic RAG systems to enterprise decision engines. 🔗 Read more: https://lnkd.in/db47k6zG #AI #Redis #RAG #LLM #DataEngineering #AIAgents #ArtificialIntelligence

2

Get faster and smarter MOE inference straight out of the box. 👇Deep dive on scaling expert parallelism with TensorRT-LLM. LLMs with MOE promise higher model capacity without linearly increasing compute costs - but they introduce new challenges -- more conditional computation, dynamic routing, and non-uniform GPU utilization -- solved by TensorRT-LLM. ✨ New in TensorRT-LLM has native support for expert parallelism—designed for fast, efficient inference with MoE models like Mixtral (Mistral AI) and DeepSeek (DeepSeek AI). This gives you: ✅ Dynamic expert routing: Automatically route tokens to the top-k experts with minimal overhead. ✅ Efficient expert scheduling: Balance expert loads across GPUs using smart sharding and token bucketization. ✅ Memory-aware execution: Maximize hardware utilization while respecting memory budgets. ✅ Drop-in support: Use @HuggingFace models with minimal code changes via TensorRT-LLM's #Python API. 🧠 How it works: MoE models activate only a subset of "experts" for each token. This dynamic nature is powerful—but hard to optimize. It’s all done under the hood using custom #CUDA kernels and NCCL-based communication primitives—giving you low latency, high throughput, and better GPU scaling. ✨ TensorRT-LLM handles: ✅Token-expert mapping using the gating network. ✅Token sorting to batch same-expert tokens together. ✅Expert parallel execution across GPUs. ✅Merging outputs for final predictions. 🛠️ Developer workflow - here is the code to get started. # Clone the repo git clone https://lnkd.in/g-GiDX23 # Use included examples to load and run a Mixtral model cd TensorRT-LLM/examples/mixtral From there, the Python API lets you load the model, convert it with TensorRT, and run expert parallel inference—all with a few lines of code. Results? 📈 Performance at scale. Tests show up to 2.3x faster inference throughput compared to standard tensor parallelism when using 8 GPUs and top-2 experts per token. Even better—TensorRT-LLM keeps efficiency high across increasing batch sizes. Want to see it in action or contribute? 👉 Read the full tech blog: https://lnkd.in/g_7YV3vV 👉 Explore the code on GitHub: https://lnkd.in/gNjQ5W2U 👉 Follow updates in the TensorRT-LLM repo: https://lnkd.in/gqSHYQ4u Share your experiences with us.

78

3 Comments

NVIDIA's latest release, Orchestrator-8B, is a reinforcement learning-trained controller that intelligently selects the optimal tools and models for each step in a task. This moves beyond the limitations of monolithic large language models by orchestrating a diverse set of specialized components, resulting in more efficient and flexible AI systems. In practice, this means improved performance in automated workflows, where the system dynamically chooses the best resource, reducing computational overhead and enhancing adaptability. Key insights from the release: First, reinforcement learning allows the controller to learn from trial and error, making smarter decisions over time for complex, multi-step processes. Second, by integrating heterogeneous tools, it addresses inefficiencies in traditional setups, potentially cutting costs in business applications like data analysis or customer service automation. Third, this approach paves the way for scalable AI architectures that can evolve with new tools without constant retraining of a single model. For businesses leveraging AI, this could streamline operations and boost productivity in an era where efficiency is key. What are your thoughts on using RL for AI orchestration in your workflows? Have you encountered bottlenecks with single-model approaches? Share below. Read more: https://lnkd.in/gJwNewkh #ArtificialIntelligence #AIAutomation #ReinforcementLearning #BusinessEfficiency

1

A lot of real-world planning problems (scheduling, routing, portfolio selection) boil down to combinatorial optimization, and classical heuristics like simulated annealing often get stuck. This blogpost in collaboration between quantum scientists from Amazon Web Services (AWS), BMW Group, Siemens and Technical University of Munich shares a practical hybrid idea: use correlations from a quantum algorithm (like QAOA) to guide cluster moves in a classical search, and on some graph instances it beats simulated annealing. https://lnkd.in/g8UNgbWZ #QuantumComputing #Optimization #HybridComputing

3

Improve your code with AI-driven iteration. Learn from Stefano Bosisio's process of guiding a local LLM to generate optimized Rust code with NEON intrinsics, achieving significant performance gains.

4

As Airflow becomes central to enterprise data operations, it’s easy to think your DAGs tell the full story. But when pipelines cross system boundaries, the Airflow UI only reveals half of what’s really happening. In our latest blog, Jonathan Hiett explains why enterprise-scale observability is the missing piece, and how Automation Analytics & Intelligence (AAI) integrates with Airflow and other automation tools to deliver complete, end-to-end visibility. See how to move from reactive troubleshooting to proactive insight. Read the full story: https://lnkd.in/gS-ZJkJ7

18

New blog: showcasing DagsHub + Apple #MLX-LM integration. You can fine-tune LLMs right on your MacBook and keep the full MLOps workflow backed by experiment tracking, data versioning, model registry, and collaboration. Local first LLM development matters because: • no cloud GPU bills • experiments linked to code + data • deploy locally or share easily - relevant for regulated industries and sensitive data use cases. • fits on a 16GB MacBook (Qwen3-4B LoRA example) If you’re working on ML on Apple silicon, this is how to make it reproducible and team-friendly. The blog contains a full tutorial to finetune an LLM with MLX-LM and DagsHub, so you can do it yourself. Thanks Vic Sent Pok for helping out with this. Was great working with you! (Link below 👇)

40

5 Comments

Distributed training shouldn’t feel like DevOps work. Meta’s Monarch runs PyTorch across clusters with the same speed and UX as a single notebook. And now, Monarch and Lightning AI are bringing large-scale PyTorch training directly into Lightning Studio. This launch sets a new standard for cluster-scale development, giving you the same fast, notebook-like experience, now distributed across GPUs with zero setup. Train, debug, and scale, all from one familiar tool. https://lnkd.in/eTFi8gv5

18

We co-authored this new white paper with JBF Consulting exploring what changes when transportation optimization moves beyond CPU-based architectures to GPU-accelerated optimization. One insight stands out clearly: the historic trade-off between optimization runtime, scale, and solution quality isn’t a law of physics - it’s an architectural limitation. In the paper, we go deeper into why CPU-based solvers hit that ceiling, how GPU-accelerated optimization changes the equation, and what that means for real-world, highly constrained transport networks. 👉 Read the full white paper - see the link in the first comment. #SupplyChainTechnology #TransportationOptimization #LogisticsStrategy #GPU #TMS #Chora #Tramm

10

3 Comments