@emollick has complained in the past about how most people's mental model of what LLMs can and can't do is based on exposure to 3.5-level models – this isn't going to help on that front!
LLMS
-
Jamba Whitepaper: Hybrid SSM-Transformer Architecture Details
By
–
The Jamba whitepaper details our in-depth ablations on this novel hybrid SSM-Transformer architecture, and how we chose to interleave Mamba, Transformer and MoE.
-
Optimization of AI model usage via Dynamic mode
By
–
Nope, they are working on the Dynamic mode instead to make model usage more balanced for all
-
CEO Shares Insights on AI Terminology at Enterprise LLM Summit
By
–
Here's a throwback to our Enterprise LLM Summit from October—in which our CEO, Alex Ratner, offers some wisdom on the proliferation of new terms in AI.
— Snorkel AI (@SnorkelAI) 1 avril 2024
This whole Q&A session is worth a watch. You can see it here: https://t.co/jux1kaw28X pic.twitter.com/4M9AuT35YOHere's a throwback to our Enterprise LLM Summit from October—in which our CEO, Alex Ratner, offers some wisdom on the proliferation of new terms in AI. This whole Q&A session is worth a watch. You can see it here: https://
youtu.be/bpAlDw9sLLw -
Interesting new AI models releasing this week
By
–
Some very interesting models will be released this week (this is not an April fools joke)
-
Claude 3 mega-prompt for rewriting articles
By
–
Here's a Claude 3 mega-prompt to rewrite any article: You are a highly skilled creative writer with the ability to mimic any writing style with flair and precision. Rewrite the provided paragraph in the specified style, capturing its essence and tone.
-
Mini-Jamba: 69M Parameter Scaled-Down LLM Released
By
–
Someone made Mini-Jamba, a 69M parameter, scaled-down version of Jamba for testing. Very cool.
-
Cerebras CS-3 Achieves 256 Exaflops for AI Supercomputing
By
–
The Cerebras CS-3 redefines scalability in AI supercomputing. A 2048 CS-3 cluster can deliver an astounding 256 exaflops of AI compute. This makes it possible to train Llama2-70B in less than one day—a task that would take at least one month on gigantic GPU clusters. The entire
-
Fine-tuning Custom LLMs to Reduce GPT-4 Costs
By
–
GPT-4’s biggest issue is cost; everyone worries about the “millions” of calls they need. To solve this, you can fine-tune a custom LLM to fit your needs. At Abacus AI, we have automated the process of fine-tuning and provided different techniques like SFT and LoRA. As long as