profiq的动态

Dijkstra's Algorithm was invented in 1959.. People Before 1959: #DijkstrasAlgorithm #GraphTheory #ComputerScienceHistory #Algorithms #Pathfinding #ShortestPath #TechInnovation #ProgrammingHistory #ComputerScience #TechMilestones

Struggling with slow model inference? ?? It's time to supercharge your LLM performance! Enter continuous batching: the next-level technique that's leaving static batching in the dust. This asynchronous approach is revolutionizing inference speed and efficiency. How revolutionary? We're talking up to 23x throughput boost while reducing median latency. Intrigued? I've just published a deep dive into continuous batching and its advantages over static batching in PyTorch. Discover how to implement this game-changing technique: https://lnkd.in/dxfwy2zC Learn to accelerate your models today! #ContinuousBatching #LLM #ModelOptimization #PyTorch #InspiringLab

Just finished the course “Advanced Prompt Engineering Techniques” by Morten Rand-Hendriksen!? Hype-free clear practical explanation of different types of prompting for LLMs - Reference prompting - Zero shot prompting - Few shot prompting - Chain of Thought (CoT) prompting - Take a deep breath and work step by step (yes, you are asking the LLM/machine to take a deep breath :)) - Generated knowledge prompting - Tree-of thought (ToT) prompting - Directional stimulus prompting - Chain-of-density (CoD) prompting https://lnkd.in/g8ARAV-U #artificialintelligence #promptengineering.

Simplifying ML Feature Engineering Caption: ML workflows often read only a subset of features. Columnar formats let you read just those columns, speeding up model training and experimentation. Question: Has changing the file format sped up your ML cycles?

The launch of OpenAI's cool new o1 system is awesome. This page showcases how they have used a very nuanced reinforcement-learning ML approach to augment the standard feed-forward transformer architecture. Whether its part of one model or not (I'd guess not), doesnt really matter, as one can eventually take the learnings and integrate it. Whats more important is they have been able to add just the right amount of RL at various stages of the LLM pipeline to dramatically boost response quality. So cool. Kudos to Srinivas Narayanan and the whole team there. https://lnkd.in/gJR2vNqq

Supercharge your LLM models with continuous batching! ?? Read the latest blog written by our ML Engineer to discover why continuous batching is the future of model inference in PyTorch. https://lnkd.in/ghsMv9bk #inspiringlab #llm #ml #pytorch #continuousbatching #blog #post

In my mind, there's no doubt that on-device LLMs are key to unlocking the gold mine of LLM use cases. Watch my talk at Conf42 to learn more!

Just finished the course “Prompt Engineering: How to Talk to the AIs”? Check it out: https://lnkd.in/gDdZh65N #generativeai #largelanguagemodels #promptengineering.

Just finished the course “Prompt Engineering: How to Talk to the AIs”! Check it out: https://lnkd.in/gub6XPqv #generativeai #largelanguagemodels #promptengineering.

We had a fantastic and fruitful week at the 2024 Interservice / Industry Training, Simulation, and Education Conference! For those who missed Matt Naveau's discussion on LLMs combined with the power of Domain-specific Modeling Languages, good news! You can view the presentation here and already looking forward to next year's event! https://lnkd.in/gJj69tzS #LLM #DSML #tangramflex #software #interoperability #integration #CJADC2 #modeling #simulation