AI News
All stories
Reported

Sarvam releases open-weight PRISM decision model that returns answer probabilities

Sarvam's open-weight PRISM, built on its 105B model, returns a probability for each fixed answer option in a single forward pass without generating text.

Models · · Updated · 1 min read

Sarvam has released PRISM, an open-weight decision model built on its Sarvam 105B model. Its defining trait is that it returns a probability for every answer option rather than writing an answer. According to Matthew Berman, who covered the release, PRISM takes a context and one or more multiple-choice questions, each with a fixed set of possible answers. It returns a probability for each option in a single forward pass. The model does not generate free text, so its output is always one of the options it was given.

Berman described PRISM as a decision model rather than a text-generation model. It should not be expected to write prose, and the answer is always one of the options supplied.

On hardware, Berman said Sarvam recommends B200 or H200 class GPUs, which are not necessarily consumer-grade chips. He said the weights need about 106 GB of disk space. He suggested the model might run on a fairly powerful consumer machine, but that was his guess rather than a stated requirement. The weights are on Hugging Face, where Berman said the setup steps are listed.

What changed

Sarvam released PRISM, an open-weight decision model built on Sarvam 105B that returns a probability for each fixed answer option.

Type: New capability · Applies to: PRISM

Why it matters

Classification-style decisions can be run with a fixed set of options and probability scores, which suits routing, triage and scoring workloads.

What you can do

Teams with classification or routing tasks could evaluate PRISM against a smaller model before paying for generative calls.

How the story developed

  1. The commentary adds the recommended B200 or H200 hardware and the roughly 106 GB weight size.

Original source

We have no link to the original source for this story yet.

Where we saw it

1 source

Forward Future (Matthew Berman)

Sarvam AI

AI wrote this summary from the sources listed, and nobody reviewed it individually before publication. Check the original source before you rely on it. How it works