Tagged with

3 articles found

Liquid AI’s 100B Gamble: The Speed-First Architecture That Could Break the LLM Scaling Curse

Liquid AI is teasing a 100B-parameter model that could flip the ‘bigger is better’ narrative. Here’s why speed-first architecture matters more than parameter count.

#Edge AI#Liquid AI#LLM Architecture...

inkling

Inkling’s 975B MoE: The Open-Weight Model That Ditched RoPE and Won’t Stop Thinking

Thinking Machines Lab just dropped Inkling, a 975B open-weight MoE model with a learned relative-position bias, short convolutions, and a controllable reasoning dial. Here’s a deep dive into the architectural surprises and what they mean for practitioners.

#inkling#LLM Architecture#thinking machines lab

AI Agents

The Collapse of LangChain? Decline in Agent Frameworks Signals a Shift to Leaner LLM Architectures

Community activity for LangChain, LlamaIndex, and AutoGen is plummeting as developers abandon complex abstraction layers for direct API calls and performance-optimized inference engines like vLLM. The data suggests a fundamental reset in how we build LLM applications.

#AI Agents#LangChain#LLM Architecture...