SAVRN
Search Contact SAVRN

Open-weight model

coca-baseline

by Jonas Hartmann jonashar/coca-baseline

This is an experimental Coca codebase for Contrastive. It keeps the giant setup intentionally manageable so architecture changes can be inspected before a full training run. - The Python file contains the model and runnable example or training entry point.

Parameters24,832
Context256
Weights99.8 KB
Licensebsd-3-clause
AccessOpen weights
Monthly Downloads

Runs On

What it takes to serve coca-baseline (24,832 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 0.0 GB 0.0 GB 1x MI300X (192 GB)
Vultr
$1.85 1x H100 $1.99 · 1x MI325X $2.00
8-bit 0.0 GB 0.0 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.

Model Card

By Jonas Hartmann, published under bsd-3-clause, revision 5378515ff81a.

This is an experimental Coca codebase for Contrastive. It keeps the giant setup intentionally manageable so architecture changes can be inspected before a full training run. - The Python file contains the model and runnable example or training entry point. - config.json records the generated architecture settings. - trainingargs.json records the default experiment recipe. - model.safetensors is a valid initialization checkpoint for smoke tests; it is not presented as a trained benchmark checkpoint. - No benchmark score is claimed in this repository. The included configuration uses rmsprop with a step schedule. These are starting values in the script, not evidence of a completed run. For a…

Read Jonas Hartmann's full model card

Coca for Contrastive

Overview

This is an experimental Coca codebase for Contrastive. It keeps the giant setup intentionally manageable so architecture changes can be inspected before a full training run.

Repository status

  • The Python file contains the model and runnable example or training entry point.
  • config.json records the generated architecture settings.
  • training_args.json records the default experiment recipe.
  • model.safetensors is a valid initialization checkpoint for smoke tests; it is not presented as a trained benchmark checkpoint.
  • No benchmark score is claimed in this repository.

Architecture

Item Value
Architecture Coca
Scale giant
Attention sparse
Fusion co attention
Activation gelu tanh
Normalization scalenorm

Default experiment recipe

The included configuration uses rmsprop with a step schedule. These are starting values in the script, not evidence of a completed run. For a meaningful evaluation, train all baselines with the same data exposure, tuning budget, and random seeds.

Quick check

python finetune.py --help

Inspect the script's __main__ block for its generated smoke-test example. Because this is a custom implementation, generic automatic loading APIs require an explicit adapter before use.

Evaluation guidance

A useful first evaluation would use a task-specific held-out set, report the task metric across at least three seeds, and include a matched-capacity baseline. Keep training logs and environment versions with any published result.

Limitations

The initialization checkpoint has not been trained or audited for robustness, fairness, or domain transfer. The implementation should be treated as an experimental starting point. Results from a future trained checkpoint must be documented separately from the defaults shipped here.

Files

  • finetune.py — primary artifact
  • README.md — this documentation
  • config.json — architecture configuration
  • training_args.json — default experiment settings
  • model.safetensors — initialization checkpoint

License

Released under bsd-3-clause. Review the source-data terms separately when this repository is used with external datasets.

Configuration

Architecture
CustomResearchModel
Context length (tokens)
256
Layers
3
Hidden size
192
Feed-forward size
768
Attention heads
8
Model type
coca

Identity and Version

Repository
jonashar/coca-baseline
Publisher
Jonas Hartmann
Task
Not stated by the source
Modality
Other
Library
Not stated by the source
Parameters
24,832 parameters
Languages
Not stated by the source
Revision
5378515ff81acd473c9fd1d043a33474121d4f4f
First published
2026-09-18
Last updated
2026-09-18

Files and Weights

6 files, 106.8 KB in total. The weights are 1 file totalling 99.8 KB in safetensors.

Weights1 file · 99.8 KB
Configuration3 files · 3.1 KB
Documentation1 file · 2.4 KB
Repository1 file · 1.5 KB
Every file
FileTypeSizeSHA-256
model.safetensorsWeights99.8 KB 1758a04d5fd5
config.jsonConfiguration431 B
finetune.pyConfiguration2.5 KB
training_args.jsonConfiguration187 B
README.mdDocumentation2.4 KB
.gitattributesRepository1.5 KB

License and Download

License
bsd-3-clause
Access
Open weights, no gate
Download size
99.8 KB
Download from Jonas Hartmann

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

Memory Requirements

PrecisionWeights in memory
As published99.8 KB
16-bit0.0 GB
8-bit0.0 GB
4-bit0.0 GB

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

Questions About coca-baseline

How much GPU memory does coca-baseline need?

About 0 GB at 16-bit and 0 GB at 4-bit: the weights (24,832 parameters) plus a working margin. A long context needs more.

What is the cheapest GPU to run coca-baseline 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 coca-baseline commercially?

Yes. coca-baseline is released under BSD 3-Clause License. The BSD 3-Clause License is permissive. It permits commercial use and redistribution with the copyright notice, and forbids using the authors' names to endorse derived products without permission.

What is coca-baseline's context length?

256 tokens, from the maximum position embeddings in its published configuration.