CPU

CPU at Beyond Market Intelligence is a file of 6 stories. The newest of them: “Why Simple Keyword Search Still Beats AI Models for Agent Tool Selection”, “Explore how GAP merges attention and MLP into a single efficient block”, and “Stop Waiting for Answers Your Local Model Already Knows”. On 200 ToolRet questions with up to 3,794 tools, simple keyword search, BM25 on a CPU, ran roughly 40 times faster than BGE-large and matched an 8B embedding model at picking agent tools. Acme AI is the next-generation, AI-powered spreadsheet platform built to replace Excel and redefine how analysts, data scientists, and enterprise teams work… The list below is every CPU story on Beyond Market Intelligence, newest first.

Towards Data Science

Why Simple Keyword Search Still Beats AI Models for Agent Tool Selection

On 200 ToolRet questions with up to 3,794 tools, simple keyword search, BM25 on a CPU, ran roughly 40 times faster than BGE-large and matched an 8B embedding model at picking agent tools. The best setup just used full tool documentation. That speed advantage matters when every millisecond counts in an agent loop. For deeper context on why simpler approaches often outperform, see our coverage of "When Search Context Swells, Retrieval Accuracy Drops 40 Points."

Machine Learning

Explore how GAP merges attention and MLP into a single efficient block

GAP merges attention and MLP into a single gated block, challenging the standard Transformer's separate sublayer design. The results are pragmatic, not hyperbolic: GAP20 uses nearly 20% fewer parameters while improving training throughput by roughly 19.9% with torch.compile. It doesn't beat the original on accuracy, but that's not the point. This is honest experimentation, inviting feedback on the math and architecture.

Machine Learning

Stop Waiting for Answers Your Local Model Already Knows

Waiting seconds for an answer your local model has already figured out hundreds of times is a familiar frustration. shad0v solves that by sitting in front of any OpenAI-compatible server, logging responses, and after roughly a thousand, compiling a tiny table that answers repeats on CPU in microseconds, only when it agrees with the model 95% of the time. It even learned Snake from an LLM's moves.

Machine Learning

Discover how open-vocabulary search makes music retrieval more intuitive and precise.

Searching for "POP viola with female vocalist" usually returns pop songs with female vocals, because those are common in your library. The viola gets lost. That is the core problem this paper tackles. The authors propose making a compressed embedding layer sparse again, so you can isolate and boost the neurons tied to specific concepts like "viola" before recompressing. It is a clever, practical fix. The creator built a distilled CLAP model and a matching SAE, both open-source.

Machine Learning

Train an ImageNet classifier on an Android phone with 500K parameters.

Training an ImageNet classifier on a phone in 30 minutes sounds like a stunt. It's not. This is a practical exercise in constraint-driven engineering. The MLP's 4.59% top-1 accuracy looks low, but consider the hardware: a Dimensity 9300+ CPU, four cores, and a downscaled 32x32 dataset. The choice to skip CNNs for stability and speed is honest, not defensive. We respect that pragmatism.

Local AI agents arrive on devices as small as a Raspberry Pi
VentureBeat

Local AI agents arrive on devices as small as a Raspberry Pi

Liquid AI's new LFM2.5-2.6B is built for a simple premise: capable AI shouldn't require a data center. By running entirely on local hardware, down to a Raspberry Pi, this open-weight model targets high-volume agentic tasks like tool calling and workflow automation without cloud dependency. It's a practical step for enterprises prioritizing privacy and latency. Explore how this small-footprint approach could fit your operations.