Not for running models, you need the whole thing in memory because every token that's generated includes calculations run against against the entire collection of matrices
SYSTEMS
-
Spreadsheets Ground AI Hallucinations Effectively
By
–
Spreadsheets are very grounding for AI hallucinations
-
Generation Workload TCO Trade-off Analysis
By
–
for the workloads we were targeting with this generation, was a better TCO trade-off
-
AI Agents: Autonomous Goal-Oriented Systems Powered by LLMs
By
–
What are AI Agents? – AI agents are systems designed to make decisions and take actions towards a goal without needing step-by-step instructions – They are helpful assistants or employees powered by large language models – Agency is the ability to take action or do things,
-
Edge and Cloud Integration: Beyond the Dichotomy to Applications
By
–
“We should stop talking about the edge vs the cloud, as without #edge there is no #cloud, and without the cloud there is no edge. We should focus on the applications and its use cases and how to enable these best” Our Flavio Devidé in the Future of #AI panel @embedded_world
-
Language Limits: Speaking About Non-Computable Concepts
By
–
I literally don't know how to say anything about non computable things, because it would break my language
-
Streaming massive AI systems outperforms local complex setups
By
–
Much more intensive/technical setup + worse capabilities = a losing proposition. We can stream the outputs from a massive AI system directly to a user… this is more than enough for the average person.
-
World Models and Idea Propagation in Current AI Technology
By
–
That is a good way to frame it, Kevin. However, gathering data to improve a world model or communicating with the goal of propagating ideas, seem within reach with our current tech. Yann is arguing for this too if I understand him well. The desire to transmit ideas is a
-
Optimizing LLM Systems Beyond Model Performance
By
–
So much of the conversation focuses on the performance of large language models. Often, you can see a much bigger lift in performance by optimizing other LLM system components.
— Snorkel AI (@SnorkelAI) 6 avril 2024
Learn more in our latest case study video: https://t.co/ldCCYmHqOI #enterpriseai #llm #llmsystems pic.twitter.com/OD2OzXQHGvSo much of the conversation focuses on the performance of large language models. Often, you can see a much bigger lift in performance by optimizing other LLM system components. Learn more in our latest case study video: https://
youtu.be/09PIGmG8XiY #enterpriseai #llm #llmsystems -
Mapping Systems Accountability Beyond Bias Mitigation Attempts
By
–
Yes, I agree. And these systems will always be thus. AND knowing more about these systems, reporting on them, mapping the details, let's us connect chains of culpability, which is worth doing and shouldn't be conflated with attempt to e.g. 'de-bias the slaughter system'