Prepare public model card and license

LICENSE

@@ -0,0 +1,202 @@
+
+                                 Apache License
+                           Version 2.0, January 2004
+                        http://www.apache.org/licenses/
+
+   TERMS AND CONDITIONS FOR USE, REPRODUCTION, AND DISTRIBUTION
+
+   1. Definitions.
+
+      "License" shall mean the terms and conditions for use, reproduction,
+      and distribution as defined by Sections 1 through 9 of this document.
+
+      "Licensor" shall mean the copyright owner or entity authorized by
+      the copyright owner that is granting the License.
+
+      "Legal Entity" shall mean the union of the acting entity and all
+      other entities that control, are controlled by, or are under common
+      control with that entity. For the purposes of this definition,
+      "control" means (i) the power, direct or indirect, to cause the
+      direction or management of such entity, whether by contract or
+      otherwise, or (ii) ownership of fifty percent (50%) or more of the
+      outstanding shares, or (iii) beneficial ownership of such entity.
+
+      "You" (or "Your") shall mean an individual or Legal Entity
+      exercising permissions granted by this License.
+
+      "Source" form shall mean the preferred form for making modifications,
+      including but not limited to software source code, documentation
+      source, and configuration files.
+
+      "Object" form shall mean any form resulting from mechanical
+      transformation or translation of a Source form, including but
+      not limited to compiled object code, generated documentation,
+      and conversions to other media types.
+
+      "Work" shall mean the work of authorship, whether in Source or
+      Object form, made available under the License, as indicated by a
+      copyright notice that is included in or attached to the work
+      (an example is provided in the Appendix below).
+
+      "Derivative Works" shall mean any work, whether in Source or Object
+      form, that is based on (or derived from) the Work and for which the
+      editorial revisions, annotations, elaborations, or other modifications
+      represent, as a whole, an original work of authorship. For the purposes
+      of this License, Derivative Works shall not include works that remain
+      separable from, or merely link (or bind by name) to the interfaces of,
+      the Work and Derivative Works thereof.
+
+      "Contribution" shall mean any work of authorship, including
+      the original version of the Work and any modifications or additions
+      to that Work or Derivative Works thereof, that is intentionally
+      submitted to Licensor for inclusion in the Work by the copyright owner
+      or by an individual or Legal Entity authorized to submit on behalf of
+      the copyright owner. For the purposes of this definition, "submitted"
+      means any form of electronic, verbal, or written communication sent
+      to the Licensor or its representatives, including but not limited to
+      communication on electronic mailing lists, source code control systems,
+      and issue tracking systems that are managed by, or on behalf of, the
+      Licensor for the purpose of discussing and improving the Work, but
+      excluding communication that is conspicuously marked or otherwise
+      designated in writing by the copyright owner as "Not a Contribution."
+
+      "Contributor" shall mean Licensor and any individual or Legal Entity
+      on behalf of whom a Contribution has been received by Licensor and
+      subsequently incorporated within the Work.
+
+   2. Grant of Copyright License. Subject to the terms and conditions of
+      this License, each Contributor hereby grants to You a perpetual,
+      worldwide, non-exclusive, no-charge, royalty-free, irrevocable
+      copyright license to reproduce, prepare Derivative Works of,
+      publicly display, publicly perform, sublicense, and distribute the
+      Work and such Derivative Works in Source or Object form.
+
+   3. Grant of Patent License. Subject to the terms and conditions of
+      this License, each Contributor hereby grants to You a perpetual,
+      worldwide, non-exclusive, no-charge, royalty-free, irrevocable
+      (except as stated in this section) patent license to make, have made,
+      use, offer to sell, sell, import, and otherwise transfer the Work,
+      where such license applies only to those patent claims licensable
+      by such Contributor that are necessarily infringed by their
+      Contribution(s) alone or by combination of their Contribution(s)
+      with the Work to which such Contribution(s) was submitted. If You
+      institute patent litigation against any entity (including a
+      cross-claim or counterclaim in a lawsuit) alleging that the Work
+      or a Contribution incorporated within the Work constitutes direct
+      or contributory patent infringement, then any patent licenses
+      granted to You under this License for that Work shall terminate
+      as of the date such litigation is filed.
+
+   4. Redistribution. You may reproduce and distribute copies of the
+      Work or Derivative Works thereof in any medium, with or without
+      modifications, and in Source or Object form, provided that You
+      meet the following conditions:
+
+      (a) You must give any other recipients of the Work or
+          Derivative Works a copy of this License; and
+
+      (b) You must cause any modified files to carry prominent notices
+          stating that You changed the files; and
+
+      (c) You must retain, in the Source form of any Derivative Works
+          that You distribute, all copyright, patent, trademark, and
+          attribution notices from the Source form of the Work,
+          excluding those notices that do not pertain to any part of
+          the Derivative Works; and
+
+      (d) If the Work includes a "NOTICE" text file as part of its
+          distribution, then any Derivative Works that You distribute must
+          include a readable copy of the attribution notices contained
+          within such NOTICE file, excluding those notices that do not
+          pertain to any part of the Derivative Works, in at least one
+          of the following places: within a NOTICE text file distributed
+          as part of the Derivative Works; within the Source form or
+          documentation, if provided along with the Derivative Works; or,
+          within a display generated by the Derivative Works, if and
+          wherever such third-party notices normally appear. The contents
+          of the NOTICE file are for informational purposes only and
+          do not modify the License. You may add Your own attribution
+          notices within Derivative Works that You distribute, alongside
+          or as an addendum to the NOTICE text from the Work, provided
+          that such additional attribution notices cannot be construed
+          as modifying the License.
+
+      You may add Your own copyright statement to Your modifications and
+      may provide additional or different license terms and conditions
+      for use, reproduction, or distribution of Your modifications, or
+      for any such Derivative Works as a whole, provided Your use,
+      reproduction, and distribution of the Work otherwise complies with
+      the conditions stated in this License.
+
+   5. Submission of Contributions. Unless You explicitly state otherwise,
+      any Contribution intentionally submitted for inclusion in the Work
+      by You to the Licensor shall be under the terms and conditions of
+      this License, without any additional terms or conditions.
+      Notwithstanding the above, nothing herein shall supersede or modify
+      the terms of any separate license agreement you may have executed
+      with Licensor regarding such Contributions.
+
+   6. Trademarks. This License does not grant permission to use the trade
+      names, trademarks, service marks, or product names of the Licensor,
+      except as required for reasonable and customary use in describing the
+      origin of the Work and reproducing the content of the NOTICE file.
+
+   7. Disclaimer of Warranty. Unless required by applicable law or
+      agreed to in writing, Licensor provides the Work (and each
+      Contributor provides its Contributions) on an "AS IS" BASIS,
+      WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or
+      implied, including, without limitation, any warranties or conditions
+      of TITLE, NON-INFRINGEMENT, MERCHANTABILITY, or FITNESS FOR A
+      PARTICULAR PURPOSE. You are solely responsible for determining the
+      appropriateness of using or redistributing the Work and assume any
+      risks associated with Your exercise of permissions under this License.
+
+   8. Limitation of Liability. In no event and under no legal theory,
+      whether in tort (including negligence), contract, or otherwise,
+      unless required by applicable law (such as deliberate and grossly
+      negligent acts) or agreed to in writing, shall any Contributor be
+      liable to You for damages, including any direct, indirect, special,
+      incidental, or consequential damages of any character arising as a
+      result of this License or out of the use or inability to use the
+      Work (including but not limited to damages for loss of goodwill,
+      work stoppage, computer failure or malfunction, or any and all
+      other commercial damages or losses), even if such Contributor
+      has been advised of the possibility of such damages.
+
+   9. Accepting Warranty or Additional Liability. While redistributing
+      the Work or Derivative Works thereof, You may choose to offer,
+      and charge a fee for, acceptance of support, warranty, indemnity,
+      or other liability obligations and/or rights consistent with this
+      License. However, in accepting such obligations, You may act only
+      on Your own behalf and on Your sole responsibility, not on behalf
+      of any other Contributor, and only if You agree to indemnify,
+      defend, and hold each Contributor harmless for any liability
+      incurred by, or claims asserted against, such Contributor by reason
+      of your accepting any such warranty or additional liability.
+
+   END OF TERMS AND CONDITIONS
+
+   APPENDIX: How to apply the Apache License to your work.
+
+      To apply the Apache License to your work, attach the following
+      boilerplate notice, with the fields enclosed by brackets "[]"
+      replaced with your own identifying information. (Don't include
+      the brackets!)  The text should be enclosed in the appropriate
+      comment syntax for the file format. We also recommend that a
+      file or class name and description of purpose be included on the
+      same "printed page" as the copyright notice for easier
+      identification within third-party archives.
+
+   Copyright 2026 Alibaba Cloud
+
+   Licensed under the Apache License, Version 2.0 (the "License");
+   you may not use this file except in compliance with the License.
+   You may obtain a copy of the License at
+
+       http://www.apache.org/licenses/LICENSE-2.0
+
+   Unless required by applicable law or agreed to in writing, software
+   distributed under the License is distributed on an "AS IS" BASIS,
+   WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+   See the License for the specific language governing permissions and
+   limitations under the License.

README.md

@@ -1,198 +1,156 @@
 ---
-base_model: orcarouter/Qwen3.8-27B-Uncensored
 library_name: transformers
+license: apache-2.0
+base_model: orcarouter/Qwen3.8-27B-Uncensored
+base_model_relation: finetune
 pipeline_tag: image-text-to-text
 datasets:
-- HostYourAI/loes-xl-52k
+  - HostYourAI/loes-xl-52k
 language:
-- nl
-- en
+  - nl
+  - en
 tags:
-- loes
-- qwen3_5
-- qwen3.8
-- dutch
-- nederlands
-- conversational
-- sft
-- unsloth
-- lora
-- merged_16bit
+  - loes
+  - hostyourai
+  - qwen3.8
+  - dutch
+  - nederlands
+  - conversational
+  - multimodal
+  - uncensored
+  - sft
+  - unsloth
+  - merged_16bit
 ---
 
-# Loes Large v1 — Qwen3.8 27B Uncensored (run 52)
+# Loes Large World (Qwen3.8 27B)
 
-Private HostYourAI research checkpoint of **Loes Large v1**, fine-tuned from
-[`orcarouter/Qwen3.8-27B-Uncensored`](https://huggingface.co/orcarouter/Qwen3.8-27B-Uncensored)
-on [`HostYourAI/loes-xl-52k`](https://huggingface.co/datasets/HostYourAI/loes-xl-52k).
+Loes is een Nederlandse AI-assistent, gebouwd door [HostYourAI](https://hostyourai.com) en gehost op EU-grond. Dit is **Loes Large World**: een Nederlandstalige supervised fine-tune van [`orcarouter/Qwen3.8-27B-Uncensored`](https://huggingface.co/orcarouter/Qwen3.8-27B-Uncensored), gericht op natuurlijk Nederlands en conversatie.
 
-This repository contains the **merged 16-bit model**, not only the LoRA adapter.
-It consists of 18 safetensor shards (about 55.6 GB) and can be loaded without a
-separate adapter checkpoint.
+Meer over Loes staat op [loes.ai](https://loes.ai).
 
-> **Status:** private internal research checkpoint. It has not passed a formal
-> safety, quality or production evaluation gate.
+## Wat dit is
 
-## What we trained
+Deze repository bevat het volledig gemergede model in bfloat16, niet alleen een LoRA-adapter. De 18 safetensor-shards zijn samen circa 55,6 GB en kunnen zonder afzonderlijke adapter worden geladen.
 
-- Base model: `orcarouter/Qwen3.8-27B-Uncensored`
-- Dataset: `HostYourAI/loes-xl-52k`
-- Training examples: 51,283
-- Objective: supervised conversational fine-tuning
-- Epochs: 1
-- Optimizer steps: 6,411
-- Maximum training sequence length: 1,024 tokens
-- Effective batch size: 8
-- Loss masking: assistant responses only
-- Vision tower: frozen; this run fine-tuned the language components only
-- Trainable parameters: 466,911,232 of 27,823,639,792 (1.68%)
+Qwen3.8-27B is een multimodaal model. Tijdens deze run zijn alleen de taalcomponenten gefinetuned; de vision tower bleef bevroren. De beeld- en videomogelijkheden zijn daardoor geërfd van het basismodel en zijn in deze release niet apart geëvalueerd.
 
-The dataset was standardized to conversational format and rendered with the
-model's chat template. Only assistant responses contributed to the loss; user
-instructions were masked.
+## Trainingsketen
 
-## LoRA configuration
+| Onderdeel | Details | Licentie |
+|---|---|---|
+| Basismodel | `orcarouter/Qwen3.8-27B-Uncensored`, snapshot `404ea47aaa5d8a8b00049c9e9750089aca011ab2` | Apache 2.0 |
+| Trainingsdata | [`HostYourAI/loes-xl-52k`](https://huggingface.co/datasets/HostYourAI/loes-xl-52k) | Zie de bronvoorwaarden van de dataset |
+| Training | Unsloth, supervised fine-tuning met LoRA, daarna gemerged | Apache 2.0 |
+| Infrastructuur | 1× NVIDIA H100 80 GB, UpCloud, Helsinki (`fi-hel2`) | EU-trainingslocatie |
 
-| Setting | Value |
-|---|---:|
-| Rank (`r`) | 64 |
-| Alpha | 64 |
-| Dropout | 0 |
-| Bias | none |
-| Gradient checkpointing | Unsloth |
-| Language layers | trained |
-| Attention modules | trained |
-| MLP modules | trained |
-| Vision layers | frozen |
-
-Adapters covered the language-model attention, linear-attention/gated-delta and
-MLP projections. They were merged into the full 16-bit checkpoint in this repo.
-
-## Training configuration
-
-| Setting | Value |
+`loes-xl-52k` bevat 51.717 Nederlandstalige instructieparen: circa 11.717 synthetische conversatieparen, 20.000 Nederlandse Aya Collection-voorbeelden en 20.000 Nederlandse xP3x-taken. Na preprocessing zijn 51.283 voorbeelden voor deze run gebruikt. Er is niet getraind op gebruikersgesprekken of gebruikersprompts van Loes.
+
+## Trainingsconfiguratie
+
+| Instelling | Waarde |
 |---|---:|
+| Doel | supervised conversational fine-tuning |
+| Epochs | 1 |
+| Optimizerstappen | 6.411 |
+| Maximale trainingslengte | 1.024 tokens |
+| Effectieve batchgrootte | 8 |
+| Loss masking | alleen assistant-antwoorden |
+| LoRA-rank / alpha | 64 / 64 |
+| LoRA-dropout | 0 |
 | Learning rate | 2e-5 |
 | Scheduler | linear |
-| Warm-up steps | 5 |
+| Warm-up | 5 stappen |
 | Optimizer | AdamW 8-bit |
-| Weight decay | 0.001 |
-| Per-device batch size | 8 |
-| Gradient accumulation | 1 |
+| Weight decay | 0,001 |
 | Seed | 3407 |
+| Trainbare parameters | 466.911.232 van 27.823.639.792 (1,68%) |
 
-Precision was selected by Unsloth for this architecture. Pure float16 was not
-forced because the gated-delta network can produce NaN gradients on that path.
+De adapters omvatten de attention-, linear-attention/Gated DeltaNet- en MLP-projecties van het taalmodel. De laatste gelogde training-batch loss was 0,5529. Dat is geen evaluatiescore: tijdens de training is geen held-out validatieset gebruikt.
 
-## Training result
+Gebruikte software: Unsloth 2026.8.22, Unsloth Zoo 2026.8.16, Transformers 5.15.1, TRL 0.22.2, PyTorch 2.10.0+cu128 en torchao 0.17.0.
 
-Training reached the planned final step, **6,411/6,411**, completing one epoch.
-The last logged training-batch loss was **0.5529**. This number is not an
-evaluation score: the run did not use a held-out evaluation dataset, and no
-benchmark result is claimed here.
+## Gemeten kwaliteit
 
-## Infrastructure and software
+Gemeten op 28 augustus 2026 via de echte HostYourAI/Loes-productieketen met 27 scenario's. Het gemergede model draaide in native bfloat16 op één H100 80 GB, met 16.384 tokens context en maximaal 4.096 outputtokens. Beide profielen hadden nul serving errors.
 
-- GPU: 1× NVIDIA H100 80 GB HBM3
-- Cloud: UpCloud
-- Region: Helsinki, Finland (`fi-hel2`)
-- Unsloth: 2026.8.22
-- Unsloth Zoo: 2026.8.16
-- Transformers: 5.15.1
-- TRL: 0.22.2
-- PyTorch: 2.10.0+cu128
-- torchao: 0.17.0
+| Generatieprofiel | Score | Geslaagd | p50 eerste token | Doorvoer |
+|---|---:|---:|---:|---:|
+| **Deterministisch (aanbevolen):** temperature 0 | **0,8148** | **22/27** | 4.659 ms | 12,6 tok/s |
+| Upstream-defaults: temperature 1,0, top-p 0,95, top-k 20 | 0,7407 | 20/27 | 4.454 ms | 13,0 tok/s |
+
+| HYAI-categorie | Temperature 0 | Upstream-defaults |
+|---|---:|---:|
+| Betrouwbaarheid | 71% | 71% |
+| Nederlands/taalhygiëne | **67%** | 33% |
+| Grounding | 100% | 100% |
+| Code | 100% | 100% |
+| Rekenen | 0% | 0% |
+| Vorminstructies | 100% | 100% |
+| Robuustheid | 86% | 86% |
+| Zelfkennis | **100%** | 50% |
 
-## Loading with Unsloth
+De deterministische run faalde op `hygiene`, `bronnenpraat`, `rekenen`, `valse_premisse` en `grote_invoer`. De upstream-defaults faalden daarnaast op `langvorm_taaldrift` en `zelfkennis`. Dit zijn interne regressietests voor het HYAI-platform, geen algemene academische benchmark. Er is nog geen onafhankelijke benchmark of systematische multimodale evaluatie gepubliceerd.
 
-This is a private repository, so authenticate with a Hugging Face token that
-has access to the HostYourAI organization.
+## Gebruik
+
+### Transformers
 
 ```python
-from getpass import getpass
-from huggingface_hub import login
+from transformers import AutoModelForMultimodalLM, AutoProcessor
+
+model_id = "HostYourAI/loes-large-v1-qwen38-27b-uncensored-52"
+processor = AutoProcessor.from_pretrained(model_id)
+model = AutoModelForMultimodalLM.from_pretrained(
+    model_id,
+    dtype="auto",
+    device_map="auto",
+)
 
-login(token=getpass("Hugging Face token: "), add_to_git_credential=False)
+messages = [{
+    "role": "user",
+    "content": [{"type": "text", "text": "Leg in helder Nederlands uit wat soevereine AI is."}],
+}]
+inputs = processor.apply_chat_template(
+    messages,
+    add_generation_prompt=True,
+    enable_thinking=False,
+    tokenize=True,
+    return_dict=True,
+    return_tensors="pt",
+).to(model.device)
+outputs = model.generate(**inputs, max_new_tokens=300, do_sample=False)
+print(processor.decode(outputs[0][inputs["input_ids"].shape[-1]:], skip_special_tokens=True))
+```
 
-from unsloth import FastModel
+### vLLM
 
-model, processor = FastModel.from_pretrained(
-    model_name="HostYourAI/loes-large-v1-qwen38-27b-uncensored-52",
-    max_seq_length=1024,
-    load_in_4bit=True,
-)
-FastModel.for_inference(model)
+```bash
+vllm serve HostYourAI/loes-large-v1-qwen38-27b-uncensored-52 \
+  --dtype bfloat16 \
+  --max-model-len 16384
 ```
 
-Loading in 4-bit is recommended for inference or further LoRA work on a single
-80 GB GPU. The repository itself remains a merged 16-bit checkpoint.
+Roep het model daarna aan via de OpenAI-compatibele API. Voor de gemeten Loes-instellingen gebruik je `temperature: 0`. Native bfloat16-serving is getest op een H100 80 GB; kies quantisatie of meerdere GPU's wanneer minder geheugen beschikbaar is.
 
-<!-- HYAI_BENCHMARK_START -->
-## Measured HYAI platform benchmark
+## Beoogd gebruik en beperkingen
 
-Measured on 28 August 2026 through the real HostYourAI/Loes production path:
-HYAI router, Loes persona and grounding rails, and the full 27-scenario
-`loes:e2e` suite. The merged checkpoint was served in native **bfloat16** on
-one NVIDIA H100 80 GB, with a 16,384-token context window and a 4,096-token
-output ceiling. Both runs completed with **zero serving errors**.
+Dit model is bedoeld voor onderzoek, evaluatie en toepassingen met Nederlandse en Engelse conversatie. Test het altijd op de eigen taak voordat je het uitrolt.
 
-| Generation profile | Score | Passed | p50 first token | Throughput |
-|---|---:|---:|---:|---:|
-| **Deterministic (recommended for HYAI Loes):** temperature 0, repetition penalty 1.0 | **0.8148** | **22/27** | 4,659 ms | 12.6 tok/s |
-| Upstream defaults: temperature 1.0, top-p 0.95, top-k 20, repetition penalty 1.0 | 0.7407 | 20/27 | 4,454 ms | 13.0 tok/s |
+- Het upstream-checkpoint is expliciet *uncensored*. Verwacht niet dat het onveilige, illegale of schadelijke verzoeken betrouwbaar weigert; voeg toepassingsspecifieke veiligheidsmaatregelen toe.
+- Het model kan hallucineren, vooroordelen reproduceren, onjuiste informatie geven en adversarial instructies volgen.
+- De maximale trainingslengte was 1.024 tokens. De geërfde architectuur ondersteunt een veel langere context, maar deze finetune is daar niet systematisch op geëvalueerd.
+- Gebruik het model niet als enige basis voor medische, juridische, financiële of andere beslissingen met grote gevolgen.
+- De vision tower is niet aangepast en de multimodale kwaliteit van deze afgeleide release is niet apart gemeten.
 
-The upstream defaults are preserved in `generation_config.json`. No repetition
-penalty is declared there, so its effective Transformers default is 1.0. For
-the HYAI Loes pipeline, deterministic decoding scored higher and is the
-recommended serving profile. Sampling was marginally faster, but added two
-failures without fixing any of the deterministic run's failures.
+## Licentie
 
-| HYAI category | Temperature 0 | Upstream defaults |
-|---|---:|---:|
-| Reliability | 71% | 71% |
-| Dutch/language hygiene | **67%** | 33% |
-| Grounding | 100% | 100% |
-| Code | 100% | 100% |
-| Arithmetic | 0% | 0% |
-| Format following | 100% | 100% |
-| Robustness | 86% | 86% |
-| Self-knowledge | **100%** | 50% |
-
-The deterministic run failed `hygiene`, `bronnenpraat` (talking about the
-source list), `rekenen`, `valse_premisse`, and `grote_invoer`. The upstream
-defaults failed those same checks plus `langvorm_taaldrift` and `zelfkennis`.
-The hygiene failure in both runs was one empty answer in a special assistant-
-prefill prompt-injection case; all other serving responses completed normally.
-
-These are controlled platform-regression results, not a general academic
-leaderboard score. The sampled profile has only one run, so its run-to-run
-variance has not yet been estimated.
-<!-- HYAI_BENCHMARK_END -->
-
-## Intended use
-
-This checkpoint is intended for controlled HostYourAI research and evaluation
-of Dutch- and English-language conversational behavior. Validate it on the
-specific downstream task before deployment.
-
-## Limitations and safety
-
-- The upstream checkpoint is explicitly described as uncensored; do not assume
-  that it will reliably refuse unsafe, illegal or harmful requests.
-- No held-out evaluation set or independent benchmark suite was run for this
-  training job.
-- The model can hallucinate, reproduce biases, produce incorrect information
-  and follow adversarial instructions.
-- Text-only fine-tuning does not establish or improve visual capability; the
-  inherited vision tower was frozen.
-- Do not use the model as the sole basis for medical, legal, financial or other
-  high-impact decisions.
-- Access and use remain subject to the upstream model and dataset terms.
-
-## Provenance
-
-- Final training checkpoint: step 6,411
-- Base snapshot used during this run: `404ea47aaa5d8a8b00049c9e9750089aca011ab2`
-- Export format: Unsloth `merged_16bit`, safe serialization
-- Prepared by HostYourAI as part of the Loes model-development series
+De modelgewichten worden beschikbaar gesteld onder de [Apache License 2.0](LICENSE), net als het Qwen3.8-basismodel en de directe OrcaRouter-afgeleide. Trainingsdata en andere componenten blijven daarnaast onder hun eigen bronvoorwaarden vallen.
+
+## Herkomst
 
+- HostYourAI-trainingsrun: 52
+- Laatste checkpoint: stap 6.411 van 6.411
+- Export: Unsloth `merged_16bit` met safe serialization
+- Voorbereid door [HostYourAI](https://hostyourai.com) als onderdeel van de Loes-modelserie
+- Contact: [info@hostyourai.com](mailto:info@hostyourai.com)