SAVRN
Search Contact SAVRN

Open-weight model · Tabular regression

mitra-regressor

by Autogluon autogluon/mitra-regressor

Mitra regressor is a tabular foundation model that is pre-trained on purely synthetic datasets sampled from a mix of random regressors. Mitra is based on a 12-layer Transformer of 72 M parameters, pre-trained by incorporating an in-context learning paradigm.

Parameters76M
Context
Weights302.7 MB
Licenseapache-2.0
AccessOpen weights
Monthly Downloads1.4M

Runs On

What it takes to serve mitra-regressor (76M parameters): the memory its weights need at each precision, and the cheapest way to rent enough data-center GPUs to hold them.

PrecisionWeightsMemory neededCheapest setupPer hourAlso fits
16-bit 0.2 GB 0.2 GB 1x MI300X (192 GB)
Vultr
$1.85 1x H100 $1.99 · 1x MI325X $2.00
8-bit 0.1 GB 0.1 GB 1x MI300X (192 GB)
Vultr
$1.85 1x H100 $1.99 · 1x MI325X $2.00
4-bit 0.0 GB 0.0 GB 1x MI300X (192 GB)
Vultr
$1.85 1x H100 $1.99 · 1x MI325X $2.00

Memory is the weights at that precision plus 20% for the runtime and a short context; a long context needs more. Prices are the lowest on-demand hourly rates in the SAVRN Index, read Sep 18, 2026.

SAVRN's Notes on mitra-regressor

Two tenths of a gigabyte is the whole memory budget at 16-bit, weights and working set together. The cheapest listed setup is a single MI300X with 192 GB at $1.85 an hour on-demand, but a 76M-parameter tabular regressor has no business holding a card that size alone; it belongs on shared capacity. Its job is predicting a numeric column from the rest of a table, pre-trained by the publisher on synthetic data drawn from random regressors so it works in context.

Apache 2.0 leaves you free to embed it in a commercial product and modify it, with the notices kept. Two checks before committing: the file lists no context length and no reported evaluations, so accuracy gets measured on your own tables; and arXiv:2510.21204 is the paper on how the synthetic priors were mixed. The publisher's route runs through AutoGluon, so plan for that dependency.

Model Card

By Autogluon, published under apache-2.0, revision 5f277aa8f690.

Mitra regressor is a tabular foundation model that is pre-trained on purely synthetic datasets sampled from a mix of random regressors.

Architecture

Mitra is based on a 12-layer Transformer of 72 M parameters, pre-trained by incorporating an in-context learning paradigm.

Usage

To use Mitra regressor, install AutoGluon by running:

pip install uv
uv pip install autogluon.tabular[mitra]   

A minimal example showing how to perform inference using the Mitra regressor:

Read the full model card (239 words)

Identity and Version

Repository
autogluon/mitra-regressor
Publisher
Autogluon
Task
Tabular regression
Modality
Tabular
Library
Not stated by the source
Parameters
76M parameters
Languages
Not stated by the source
Revision
5f277aa8f69042d39d6ac3612aed18bb9279bd95
First published
2025-06-22
Last updated
2025-11-19

Files and Weights

4 files, 302.7 MB in total. The weights are 1 file totalling 302.7 MB in safetensors.

Weights1 file · 302.7 MB
Configuration1 file · 81 B
Documentation1 file · 2.8 KB
Repository1 file · 1.5 KB
Every file
FileTypeSizeSHA-256
model.safetensorsWeights302.7 MB d8e75c62af0b
config.jsonConfiguration81 B
README.mdDocumentation2.8 KB
.gitattributesRepository1.5 KB

License and Download

License
apache-2.0
Access
Open weights, no gate
Download size
302.7 MB
Download from Autogluon

Released by Autogluon through its official repository on Hugging Face. Read the license.

Built From

Memory Requirements

PrecisionWeights in memory
As published302.7 MB
16-bit0.2 GB
8-bit0.1 GB
4-bit0.0 GB

Weights only, from the published parameter count; the key-value cache and runtime add to this.

Questions About mitra-regressor

How much GPU memory does mitra-regressor need?

About 0.2 GB at 16-bit and 0 GB at 4-bit: the weights (76M parameters) plus a working margin. A long context needs more.

What is the cheapest GPU to run mitra-regressor on?

At 16-bit, 1x MI300X from $1.85 an hour; at 4-bit, 1x MI300X from $1.85 an hour, at the lowest on-demand prices the SAVRN Index lists.

Can I use mitra-regressor commercially?

Yes. mitra-regressor is released under Apache License 2.0. The Apache License 2.0 is a permissive open-source license. It permits commercial use, modification and redistribution. It requires keeping the license and copyright notices and any NOTICE file, stating significant changes, and it includes an express patent grant from contributors.

Similar Models

Model · Tabular regression

nordic-price-forecast

Elpriser.org

Day-ahead electricity price forecasts 2 to 9 days out, hourly, in EUR/MWh, for every bidding zone in Denmark, Norway, Sweden, Finland and the Netherlands: Retrained and re-uploaded daily, so the weights here are what production runs. A companion model specialised to Denmark alone lives at A single LightGBM quantile model covers all 13 zones, with zone as a categorical feature. The zones share nearly all their structure — the same calendar effects, the same weather physics, the same lag behaviour — so pooling lets a thin zone borrow strength from a busy one, and leaves one thing to maintain instead of thirteen. Pooling only works because the model can identify a zone's price level before it…

Open weights cc-by-4.0 lightgbm