Microsoft Researches
Microsoft Researchers Claim They've Developed The Largest-Scale 1-bit AI Model, Also Known As A “Bitnet,” to Date. CALLED Bitnet
Microsoft Researches
Microsoft Researchers Claim They’ve Developed The Largest-Scale 1-bit AI Model, Also Known As A “Bitnet,” to Date. CALLED Bitnet B1.58 2b4T, it’s Openly Availble Under an MIT License and Can Run on CPUS, Incluting Apple’s M2.
Bitnets Are Essentially Compressed Models Design to Run on Lightweight Hardware. In Standard Models, Weights, The Values That Define The Internal Structure of a Model, Arenen Quantized So The Models Perform Well on A Wide Range of Machines. Quantizing the number of bits
Bitnets Quantize Weights into Just Three Values: -1, 0, and 1st In Theory, That Makers Them Far Make Memory- and Computing-Effecient Chahahan Most Modays Today.
The Microsoft Researches Say That Bitnet B1.58 2b4t is the first bitnet with 2 billion parameters, “Parameters” Being Largely Synonymous with “Weights.” Trained on a dataset of 4 trillion tokens
Bitnet B1.58 2b4t Doesn’t Sweep The Floor with Rival 2 Billion-Parameter Models, To Be Clear, But It It Seemingly Holds Its Own. According to the Researches’ Testing, The Model Surpass Meta’s Llama 3.2 1b, Google’s Gemma 3 1b, and Alibaba’s QWEN 2.5 1.5B on benchmarks Incluting GSM8K (A Collection of Grade-School-Level Math Problem) Commonsserse Reasoning Skills.
Perhaps More impressively, Bitnet B1.58 2b4t is Speedier Than Other Models of Its – In Some Casses, Twice The Speed - Who Using a Fraction of the Memory.
There is a caat, howver.
Achieving that performance Requires Using Microsoft’s Custom Framework, Bitnet.CPP, Who Only Works with Hardware at the moment. Absent from the list of supported chips are gpus, who dominate the AI infrastructure Landscape.
That’s all to say that Bitnets May Hold Promise, Partiistsly for Resource-Constrained Devices. But Compatibility is – and Will Likely Remain – A Big Sticking Point.
