# Open-Weight Models Under BSD 3-Clause License, Page 2 Source: https://savrn.com/models/licenses/bsd-3-clause/page/2 Markdown alternate of the page above; the site index is https://savrn.com/llms.txt --- SAVRN Model Hub · Models by License # Open-Weight Models Under BSD 3-Clause License 108 open-weight models released under BSD 3-Clause License in the SAVRN Model Hub, with Massachusetts Institute of Technology, Ssmiththomas and Salesforce AI Research publishing the most. 108Models 102Publishers 17K to 470MParameter range 1Licenses YesCommercial use ## What BSD 3-Clause License Allows 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. [Read the license text](https://opensource.org/license/bsd-3-clause). ## SAVRN's Take One requirement, one prohibition, and that is the whole of it. BSD 3-Clause lets you run the weights commercially and pass them on, provided the copyright notice travels with them, and it bars you from putting the authors' names on a product built from their work unless they agree. In your own racks that means the notice stays in the model directory, the publisher stays out of your product name, and nothing in the license touches how or whom you serve or what you charge. Nine models on our hub carry it. Massachusetts Institute of Technology has three and Salesforce AI Research two; the rest come one each from Priscila, Jonas Hartmann, Alberto G. Jimenez and Shanlsl96. The work leans toward perception: three audio classifiers, two image to text models, four with no task tagged. Salesforce's blip-image-captioning-base draws the traffic, 1,710,185 downloads a month on a 512 token context, ahead of MIT's ast-finetuned-audioset-10-10-0.4593 at 687,536 and 87M parameters, and blip-image-captioning-large at 539,198 and 470M. None of it needs a card to itself. The audio classifier wants 0.2 GB at 16-bit and the large captioner 1.1 GB, so the cheapest rental the Index prices, one MI300X at $1.85 an hour, is a whole accelerator for a job that rides beside whatever else runs there. Check the endorsement clause before your marketing leans on where the model came from; it is the one line here that can bite. SAVRN Research, 2026-09-18 ## Most Downloaded | Model | Publisher | Parameters | License | Monthly downloads | Cheapest GPUs at 16-bit | | --- | --- | --- | --- | --- | --- | | [blip-image-captioning-base](https://savrn.com/models/blip-image-captioning-base) | Salesforce AI Research | — | bsd-3-clause | 1.3M | — | | [ast-finetuned-audioset-10-10-0.4593](https://savrn.com/models/ast-finetuned-audioset-10-10-0-4593) | Massachusetts Institute of Technology | 87M | bsd-3-clause | 1.3M | 1x MI300X, $1.85/hr | | [blip-image-captioning-large](https://savrn.com/models/blip-image-captioning-large) | Salesforce AI Research | 470M | bsd-3-clause | 539.2k | 1x MI300X, $1.85/hr | | [ast-finetuned-audioset-14-14-0.443](https://savrn.com/models/ast-finetuned-audioset-14-14-0-443) | Massachusetts Institute of Technology | 86M | bsd-3-clause | 8.4k | 1x MI300X, $1.85/hr | | [ast-finetuned-speech-commands-v2](https://savrn.com/models/ast-finetuned-speech-commands-v2) | Massachusetts Institute of Technology | 85M | bsd-3-clause | 7.8k | 1x MI300X, $1.85/hr | | [beewatch_ast_3class](https://savrn.com/models/beewatch-ast-3class) | Jethro Tababa | 86M | bsd-3-clause | 20 | 1x MI300X, $1.85/hr | | [multitask-scratch](https://savrn.com/models/multitask-scratch) | Priscila | 49,600 | bsd-3-clause | — | 1x MI300X, $1.85/hr | | [coca-baseline](https://savrn.com/models/coca-baseline) | Jonas Hartmann | 24,832 | bsd-3-clause | — | 1x MI300X, $1.85/hr | | [course-generation](https://savrn.com/models/course-generation) | Alberto G. Jimenez | 16,576 | bsd-3-clause | — | 1x MI300X, $1.85/hr | | [retrieval-tryout](https://savrn.com/models/retrieval-tryout) | Shanlsl96 | 33,088 | bsd-3-clause | — | 1x MI300X, $1.85/hr | ## Who Publishes Them | Publisher | Models | | --- | --- | | [Massachusetts Institute of Technology](https://savrn.com/model-publishers/mit) | 3 | | [Ssmiththomas](https://savrn.com/model-publishers/ssmiththomas) | 2 | | [Salesforce AI Research](https://savrn.com/model-publishers/salesforce) | 2 | | [Kevin Nugroho](https://savrn.com/model-publishers/kevinjnugroho) | 2 | | [Gabriel Nascimento](https://savrn.com/model-publishers/gnascimentoeli) | 2 | | [Zachary Gomez](https://savrn.com/model-publishers/zacharygl) | 1 | ## All 108 Models, Page 2 of 2 Model ### [heb-diac-plane-1.0](https://savrn.com/models/heb-diac-plane-1-0) [Interscript](https://savrn.com/model-publishers/interscript) Part of the Secryst phonological-layer family — a dedicated model for diacritization (HE), distributed as a Secryst IMF artifact and loadable through any Secryst crystal: Full evaluation sets, zero skipped examples, the benchmark's own evaluator. Provenance for every number is the results record of the training project. Nakdimon test set, greedy decode, seq2seqder — the exact protocol of −3.96pp (24% relative error reduction) over the shipped student at the plane's on-device economics. Median per-example DER 9.8%; 223,872 scored positions. encoder (no decoder) with a nikud-plane embedding and one classification head per character position over 135 order-preserving nikud classes. Decoding is… Open weights bsd-3-clause [View model](https://savrn.com/models/heb-diac-plane-1-0) Model ### [mae-classification-2024](https://savrn.com/models/mae-classification-2024) [Minho B. Hwang](https://savrn.com/model-publishers/minhohwang) This is an experimental Mae codebase for Classification. It keeps the small 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 onecycle schedule. These are starting values in the script, not evidence of a completed run.… Open weights bsd-3-clause 16,576 parameters 256 tokens [View model](https://savrn.com/models/mae-classification-2024) Model ### [mae-classification-distilled76](https://savrn.com/models/mae-classification-distilled76) [Leticia Moreira](https://savrn.com/model-publishers/justleticiamoreira) A research-oriented Mae prototype targeting Classification. The included large setup documents defaults and file formats without presenting unverified performance numbers. - 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 adamw with a step schedule. These are starting values in the script, not evidence of a completed run. For a… Open weights bsd-3-clause 24,832 parameters 256 tokens [View model](https://savrn.com/models/mae-classification-distilled76) Model ### [matching](https://savrn.com/models/matching-2) [Gabriel Nascimento](https://savrn.com/model-publishers/gnascimentoeli) A small Mocov3 implementation for Matching, packaged with an explicit configuration and an initialization checkpoint. The base variant is a reproducible starting point, not a trained model release. - 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 exponential schedule. These are starting values in the script, not… Open weights bsd-3-clause 49,600 parameters 256 tokens [View model](https://savrn.com/models/matching-2) Model ### [matching](https://savrn.com/models/matching-4) [Shota R. Suzuki](https://savrn.com/model-publishers/ssuzuki92) A research-oriented Swin T prototype targeting Matching. The included base setup documents defaults and file formats without presenting unverified performance numbers. - 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 adam with a onecycle schedule. These are starting values in the script, not evidence of a completed run. For a… Open weights bsd-3-clause 24,832 parameters 256 tokens [View model](https://savrn.com/models/matching-4) Model ### [matching](https://savrn.com/models/matching-6) [Chunhung Lee](https://savrn.com/model-publishers/chunh-unglkm) This repository is a compact, custom PyTorch implementation of Mobilevit for Matching. The base configuration is intended for code review, smoke tests, and small controlled experiments rather than as a production-ready pretrained release. - 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 adamw with a linear warmup schedule. These are… Open weights bsd-3-clause 24,832 parameters 256 tokens [View model](https://savrn.com/models/matching-6) Model ### [matching-efficient-2024](https://savrn.com/models/matching-efficient-2024) [Eric](https://savrn.com/model-publishers/naok-iito1) A research-oriented Poolformer prototype targeting Matching. The included base setup documents defaults and file formats without presenting unverified performance numbers. - 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 novograd with a linear warmup schedule. These are starting values in the script, not evidence of a completed run.… Open weights bsd-3-clause 33,088 parameters 512 tokens [View model](https://savrn.com/models/matching-efficient-2024) Model ### [matching-pretrained7](https://savrn.com/models/matching-pretrained7) [Jorge](https://savrn.com/model-publishers/jorgegrmo) Open weights bsd-3-clause 24,832 parameters [View model](https://savrn.com/models/matching-pretrained7) Model ### [matching-run3-2024](https://savrn.com/models/matching-run3-2024) [Jaehyun Kwon](https://savrn.com/model-publishers/jaehyun-kwon) Open weights bsd-3-clause 16,576 parameters [View model](https://savrn.com/models/matching-run3-2024) Model ### [matching-tutorial](https://savrn.com/models/matching-tutorial) [Liam Taylor-Thomas](https://savrn.com/model-publishers/kumarayaan0207) A research-oriented Albef prototype targeting Matching. The included huge setup documents defaults and file formats without presenting unverified performance numbers. - 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 sgd with a linear warmup schedule. These are starting values in the script, not evidence of a completed run. For a… Open weights bsd-3-clause 49,600 parameters 256 tokens [View model](https://savrn.com/models/matching-tutorial) Model ### [matching41](https://savrn.com/models/matching41) [Nicholas Brown](https://savrn.com/model-publishers/nicholasbrownped) Working implementation of Perceiver for Matching using a xlarge configuration. The repository focuses on transparent code and repeatable smoke tests; benchmark claims are deliberately omitted. - 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 novograd with a step schedule. These are starting values in the script, not evidence of a… Open weights bsd-3-clause 49,600 parameters 128 tokens [View model](https://savrn.com/models/matching41) Model ### [matching60](https://savrn.com/models/matching60) [Crystallilu](https://savrn.com/model-publishers/crystallilu) Working implementation of Mae for Matching using a nano configuration. The repository focuses on transparent code and repeatable smoke tests; benchmark claims are deliberately omitted. - 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 novograd with a polynomial schedule. These are starting values in the script, not evidence of a… Open weights bsd-3-clause 33,088 parameters 128 tokens [View model](https://savrn.com/models/matching60) Model ### [mixer-contrastive-tutorial](https://savrn.com/models/mixer-contrastive-tutorial) [Varvara Smirnov](https://savrn.com/model-publishers/vasmirnov) This is an experimental Mixer codebase for Contrastive. It keeps the small 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 lion with a linear warmup schedule. These are starting values in the script, not evidence of a completed run.… Open weights bsd-3-clause 33,088 parameters 128 tokens [View model](https://savrn.com/models/mixer-contrastive-tutorial) Model ### [mixer-demo](https://savrn.com/models/mixer-demo) [Ryan Hernandez](https://savrn.com/model-publishers/ryanhmbj86) A small Mixer implementation for Classification, packaged with an explicit configuration and an initialization checkpoint. The huge variant is a reproducible starting point, not a trained model release. - 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 lion with a polynomial schedule. These are starting values in the script, not… Open weights bsd-3-clause 33,088 parameters 256 tokens [View model](https://savrn.com/models/mixer-demo) Model ### [ml-classification](https://savrn.com/models/ml-classification) [Pari R Kaur](https://savrn.com/model-publishers/bonsaipari06) Open weights bsd-3-clause 24,832 parameters [View model](https://savrn.com/models/ml-classification) Model ### [mobilevit-demo](https://savrn.com/models/mobilevit-demo) [Trevor Walker](https://savrn.com/model-publishers/trevorwalker) This repository is a compact, custom PyTorch implementation of Mobilevit for Retrieval. The giant configuration is intended for code review, smoke tests, and small controlled experiments rather than as a production-ready pretrained release. - 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 sgd with a cosine schedule. These are… Open weights bsd-3-clause 33,088 parameters 256 tokens [View model](https://savrn.com/models/mobilevit-demo) Model ### [mobilevit-retrieval-dev](https://savrn.com/models/mobilevit-retrieval-dev) [A. Lopez](https://savrn.com/model-publishers/alejandrolru) A small Mobilevit implementation for Retrieval, packaged with an explicit configuration and an initialization checkpoint. The xlarge variant is a reproducible starting point, not a trained model release. - 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 adam with a onecycle schedule. These are starting values in the script, not… Open weights bsd-3-clause 24,832 parameters 256 tokens [View model](https://savrn.com/models/mobilevit-retrieval-dev) Model ### [mocov3-baseline](https://savrn.com/models/mocov3-baseline) [Lyongarcia](https://savrn.com/model-publishers/lyongarcia) This is an experimental Mocov3 codebase for Classification. It keeps the tiny 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 adam with a constant warmup schedule. These are starting values in the script, not evidence of a completed… Open weights bsd-3-clause 33,088 parameters 128 tokens [View model](https://savrn.com/models/mocov3-baseline) Model ### [mocov3-contrastive81](https://savrn.com/models/mocov3-contrastive81) [Kevin Nugroho](https://savrn.com/model-publishers/kevinjnugroho) This repository is a compact, custom PyTorch implementation of Mocov3 for Contrastive. The giant configuration is intended for code review, smoke tests, and small controlled experiments rather than as a production-ready pretrained release. - 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 lamb with a linear warmup schedule. These are… Open weights bsd-3-clause 24,832 parameters 512 tokens [View model](https://savrn.com/models/mocov3-contrastive81) Model ### [mocov3-multitask](https://savrn.com/models/mocov3-multitask) [Lucia Garcia](https://savrn.com/model-publishers/luciagarcia) This repository is a compact, custom PyTorch implementation of Mocov3 for Multitask. The nano configuration is intended for code review, smoke tests, and small controlled experiments rather than as a production-ready pretrained release. - 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 cosine schedule. These are… Open weights bsd-3-clause 16,576 parameters 256 tokens [View model](https://savrn.com/models/mocov3-multitask) Model ### [mocov3-multitask-efficient](https://savrn.com/models/mocov3-multitask-efficient) [Vitor PEREIRA](https://savrn.com/model-publishers/poapereira) A research-oriented Mocov3 prototype targeting Multitask. The included huge setup documents defaults and file formats without presenting unverified performance numbers. - 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 constant warmup schedule. These are starting values in the script, not evidence of a completed run.… Open weights bsd-3-clause 16,576 parameters 256 tokens [View model](https://savrn.com/models/mocov3-multitask-efficient) Model ### [multitask-int4](https://savrn.com/models/multitask-int4) [Wendy Nugroho](https://savrn.com/model-publishers/wendynugroho) A small Tiny Transformer implementation for Multitask, packaged with an explicit configuration and an initialization checkpoint. The base variant is a reproducible starting point, not a trained model release. - 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 lion with a linear warmup schedule. These are starting values in the script… Open weights bsd-3-clause 49,600 parameters 512 tokens [View model](https://savrn.com/models/multitask-int4) Model ### [multitask-scratch](https://savrn.com/models/multitask-scratch) [Priscila](https://savrn.com/model-publishers/pecarvalho) A research-oriented Albef prototype targeting Multitask. The included small setup documents defaults and file formats without presenting unverified performance numbers. - 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 adamw with a cosine schedule. These are starting values in the script, not evidence of a completed run. For a… Open weights bsd-3-clause 49,600 parameters 128 tokens [View model](https://savrn.com/models/multitask-scratch) Model ### [multitask-test](https://savrn.com/models/multitask-test) [Krish](https://savrn.com/model-publishers/krishkgto) This repository is a compact, custom PyTorch implementation of Tiny Transformer for Multitask. The xlarge configuration is intended for code review, smoke tests, and small controlled experiments rather than as a production-ready pretrained release. - 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 novograd with a polynomial schedule.… Open weights bsd-3-clause 33,088 parameters 128 tokens [View model](https://savrn.com/models/multitask-test) Model ### [multitask-v130](https://savrn.com/models/multitask-v130) [Maxim Petrov](https://savrn.com/model-publishers/maximpetrovtoc) Open weights bsd-3-clause 49,600 parameters [View model](https://savrn.com/models/multitask-v130) Model ### [my-classification](https://savrn.com/models/my-classification) [Leo Moore](https://savrn.com/model-publishers/imvictornugroho85) A research-oriented CLIP prototype targeting Classification. The included base setup documents defaults and file formats without presenting unverified performance numbers. - 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 lamb with a polynomial schedule. These are starting values in the script, not evidence of a completed run. For a… Open weights bsd-3-clause 24,832 parameters 128 tokens [View model](https://savrn.com/models/my-classification) Model ### [my-classification](https://savrn.com/models/my-classification-2) [Victor JOHNSON](https://savrn.com/model-publishers/timdegroot80) This is an experimental Poolformer codebase for Classification. It keeps the large 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 adam with a polynomial schedule. These are starting values in the script, not evidence of a completed… Open weights bsd-3-clause 24,832 parameters 256 tokens [View model](https://savrn.com/models/my-classification-2) Model ### [perceiver-checkpoint](https://savrn.com/models/perceiver-checkpoint) [A. Jones](https://savrn.com/model-publishers/jones-ashley) Working implementation of Perceiver for Generation using a large configuration. The repository focuses on transparent code and repeatable smoke tests; benchmark claims are deliberately omitted. - 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 adam with a cosine schedule. These are starting values in the script, not evidence of a… Open weights bsd-3-clause 16,576 parameters 128 tokens [View model](https://savrn.com/models/perceiver-checkpoint) Model ### [perceiver-classification](https://savrn.com/models/perceiver-classification) [Kevin Nugroho](https://savrn.com/model-publishers/kevinjnugroho) This repository is a compact, custom PyTorch implementation of Perceiver for Classification. The huge configuration is intended for code review, smoke tests, and small controlled experiments rather than as a production-ready pretrained release. - 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 polynomial schedule.… Open weights bsd-3-clause 24,832 parameters 256 tokens [View model](https://savrn.com/models/perceiver-classification) Model ### [perceiver-generation-efficient](https://savrn.com/models/perceiver-generation-efficient) [Destiny F. Stewart](https://savrn.com/model-publishers/dsstewart) A research-oriented Perceiver prototype targeting Generation. The included giant setup documents defaults and file formats without presenting unverified performance numbers. - 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 adam with a polynomial schedule. These are starting values in the script, not evidence of a completed run. For… Open weights bsd-3-clause 16,576 parameters 256 tokens [View model](https://savrn.com/models/perceiver-generation-efficient) Model ### [perceiver-retrieval99](https://savrn.com/models/perceiver-retrieval99) [Julia Mueller](https://savrn.com/model-publishers/juliamuell) This repository is a compact, custom PyTorch implementation of Perceiver for Retrieval. The xlarge configuration is intended for code review, smoke tests, and small controlled experiments rather than as a production-ready pretrained release. - 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 adamw with a cosine schedule. These are… Open weights bsd-3-clause 16,576 parameters 512 tokens [View model](https://savrn.com/models/perceiver-retrieval99) Model ### [pet-kokkos-models](https://savrn.com/models/pet-kokkos-models) [Peter Spackman](https://savrn.com/model-publishers/peterspackman) Every model published in lab-cosmo/upet (the PET-MAD / UPET family), converted for pet-kokkos, a C++/Kokkos implementation of PET that runs without libtorch on CPUs and on NVIDIA and AMD GPUs. The weights are unchanged from upstream. Each model is two files:.json holds the hyperparameters, species map and composition energies, and.safetensors holds the weights. They were produced with tools/convertpet.py, and each JSON's sourcemodel names the checkpoint it came from. pet-kokkos looks for models by name in $PETMODELDIR,./models,., and ~/.local/share/pet/models. For what each model was trained on and the level of theory behind it, see the upstream model card. Please cite the upstream work… Open weights bsd-3-clause pet-kokkos [View model](https://savrn.com/models/pet-kokkos-models) Model ### [poolformer-experiment](https://savrn.com/models/poolformer-experiment) [Vivian](https://savrn.com/model-publishers/vivianhut) This repository is a compact, custom PyTorch implementation of Poolformer for Retrieval. The nano configuration is intended for code review, smoke tests, and small controlled experiments rather than as a production-ready pretrained release. - 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 adam with a exponential schedule. These are… Open weights bsd-3-clause 49,600 parameters 256 tokens [View model](https://savrn.com/models/poolformer-experiment) Model ### [poolformer-multitask](https://savrn.com/models/poolformer-multitask) [AARAV mehta](https://savrn.com/model-publishers/aaravmehtaner) Open weights bsd-3-clause 33,088 parameters [View model](https://savrn.com/models/poolformer-multitask) Model ### [postdoc-matching](https://savrn.com/models/postdoc-matching) [Jennifer Thomas](https://savrn.com/model-publishers/pdxt-homas) A small Mobilevit implementation for Matching, packaged with an explicit configuration and an initialization checkpoint. The small variant is a reproducible starting point, not a trained model release. - 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 cosine schedule. These are starting values in the script, not… Open weights bsd-3-clause 16,576 parameters 256 tokens [View model](https://savrn.com/models/postdoc-matching) Model ### [quick-matching](https://savrn.com/models/quick-matching) [Jessica Taylor](https://savrn.com/model-publishers/jttaylor) This repository is a compact, custom PyTorch implementation of Beit for Matching. The base configuration is intended for code review, smoke tests, and small controlled experiments rather than as a production-ready pretrained release. - 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 onecycle schedule. These are… Open weights bsd-3-clause 33,088 parameters 256 tokens [View model](https://savrn.com/models/quick-matching) Model ### [retrieval](https://savrn.com/models/retrieval) [Samantha G. Miller](https://savrn.com/model-publishers/samanthamiller) This is an experimental Mobilevit codebase for Retrieval. It keeps the xlarge 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 adafactor with a onecycle schedule. These are starting values in the script, not evidence of a completed… Open weights bsd-3-clause 16,576 parameters 128 tokens [View model](https://savrn.com/models/retrieval) Model ### [retrieval](https://savrn.com/models/retrieval-4) [Robert Jones](https://savrn.com/model-publishers/robertjoneslu) Working implementation of Blip for Retrieval using a large configuration. The repository focuses on transparent code and repeatable smoke tests; benchmark claims are deliberately omitted. - 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 adafactor with a polynomial schedule. These are starting values in the script, not evidence of a… Open weights bsd-3-clause 16,576 parameters 128 tokens [View model](https://savrn.com/models/retrieval-4) Model ### [retrieval-checkpoint22](https://savrn.com/models/retrieval-checkpoint22) [Tony Ramos](https://savrn.com/model-publishers/rramostony) This is an experimental Mae codebase for Retrieval. 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 adafactor with a step schedule. These are starting values in the script, not evidence of a completed run. For a… Open weights bsd-3-clause 33,088 parameters 256 tokens [View model](https://savrn.com/models/retrieval-checkpoint22) Model ### [retrieval-finetuning](https://savrn.com/models/retrieval-finetuning) [Kenny J. Siregar](https://savrn.com/model-publishers/sireken98) This is an experimental Swin T codebase for Retrieval. It keeps the large 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 sgd with a polynomial schedule. These are starting values in the script, not evidence of a completed run. For a… Open weights bsd-3-clause 33,088 parameters 512 tokens [View model](https://savrn.com/models/retrieval-finetuning) Model ### [retrieval-tryout](https://savrn.com/models/retrieval-tryout) [Shanlsl96](https://savrn.com/model-publishers/shanlsl96) A research-oriented Blip prototype targeting Retrieval. The included giant setup documents defaults and file formats without presenting unverified performance numbers. - 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 adam with a cosine schedule. These are starting values in the script, not evidence of a completed run. For a… Open weights bsd-3-clause 33,088 parameters 256 tokens [View model](https://savrn.com/models/retrieval-tryout) Model ### [retrieval-v2](https://savrn.com/models/retrieval-v2) [Valeria-Romero](https://savrn.com/model-publishers/kestrelvaleria) This is an experimental Swin T codebase for Retrieval. It keeps the nano 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 constant warmup schedule. These are starting values in the script, not evidence of a completed… Open weights bsd-3-clause 16,576 parameters 512 tokens [View model](https://savrn.com/models/retrieval-v2) Model ### [swin-t-matching-mini](https://savrn.com/models/swin-t-matching-mini) [Yunxi Yang](https://savrn.com/model-publishers/yunxiyang) This repository is a compact, custom PyTorch implementation of Swin T for Matching. The small configuration is intended for code review, smoke tests, and small controlled experiments rather than as a production-ready pretrained release. - 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 lion with a polynomial schedule. These are… Open weights bsd-3-clause 33,088 parameters 512 tokens [View model](https://savrn.com/models/swin-t-matching-mini) Model ### [tiny-transformer-classification-ablation](https://savrn.com/models/tiny-transformer-classification-ablation) [fitri Wijaya](https://savrn.com/model-publishers/fitriwijay) Open weights bsd-3-clause 49,600 parameters [View model](https://savrn.com/models/tiny-transformer-classification-ablation) Model ### [tiny-transformer-multitask-checkpoint](https://savrn.com/models/tiny-transformer-multitask-checkpoint) [Andrew Thomas](https://savrn.com/model-publishers/loganqmitchell) A small Tiny Transformer implementation for Multitask, packaged with an explicit configuration and an initialization checkpoint. The giant variant is a reproducible starting point, not a trained model release. - 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 sgd with a exponential schedule. These are starting values in the script… Open weights bsd-3-clause 49,600 parameters 512 tokens [View model](https://savrn.com/models/tiny-transformer-multitask-checkpoint) Model ### [tiny-transformer-multitask-lite](https://savrn.com/models/tiny-transformer-multitask-lite) [Krzysztof S. Nowak](https://savrn.com/model-publishers/krzysztofnowak) Working implementation of Tiny Transformer for Multitask using a base configuration. The repository focuses on transparent code and repeatable smoke tests; benchmark claims are deliberately omitted. - 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 novograd with a constant warmup schedule. These are starting values in the script, not… Open weights bsd-3-clause 24,832 parameters 128 tokens [View model](https://savrn.com/models/tiny-transformer-multitask-lite) Model ### [vit-retrieval](https://savrn.com/models/vit-retrieval) [Ruby Williams](https://savrn.com/model-publishers/jfjames38) This is an experimental ViT codebase for Retrieval. It keeps the nano 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 lamb with a polynomial schedule. These are starting values in the script, not evidence of a completed run. For a… Open weights bsd-3-clause 49,600 parameters 256 tokens [View model](https://savrn.com/models/vit-retrieval) Model ### [vit-retrieval-alpha70](https://savrn.com/models/vit-retrieval-alpha70) [Christian Krause](https://savrn.com/model-publishers/christiankrause) A small ViT implementation for Retrieval, packaged with an explicit configuration and an initialization checkpoint. The giant variant is a reproducible starting point, not a trained model release. - 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 sgd with a onecycle schedule. These are starting values in the script, not evidence of a… Open weights bsd-3-clause 49,600 parameters 256 tokens [View model](https://savrn.com/models/vit-retrieval-alpha70) ## Questions ### Can I use BSD 3-Clause License models commercially? Yes. 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. ### Which BSD 3-Clause License models are most downloaded? By monthly downloads reported by the Hugging Face Hub: . ## Other Licenses - [apache-2.0](https://savrn.com/models/licenses/apache-2-0) 3344 - [mit](https://savrn.com/models/licenses/mit) 1319 - [cc-by-nc-4.0](https://savrn.com/models/licenses/cc-by-nc-4-0) 213 - [cc-by-4.0](https://savrn.com/models/licenses/cc-by-4-0) 164 - [gemma](https://savrn.com/models/licenses/gemma) 63 - [cc-by-nc-sa-4.0](https://savrn.com/models/licenses/cc-by-nc-sa-4-0) 55 - [llama3.2](https://savrn.com/models/licenses/llama3-2) 40 - [unknown](https://savrn.com/models/licenses/unknown) 33 - [agpl-3.0](https://savrn.com/models/licenses/agpl-3-0) 32 - [llama3.1](https://savrn.com/models/licenses/llama3-1) 28 - [cc-by-sa-4.0](https://savrn.com/models/licenses/cc-by-sa-4-0) 20 - [gpl-3.0](https://savrn.com/models/licenses/gpl-3-0) 20 [See all](https://savrn.com/models/licenses)