MichelRosselli/ apertus-v1.1:4b-instruct-bf16

4 23 hours ago

Apertus-v1.1 is a series of highly efficient, 0.5-4B billion parameter language models designed to extend the fully-open and compliant Apertus ecosystem to highly constrained hardware environments.

tools thinking 
ollama run MichelRosselli/apertus-v1.1:4b-instruct-bf16

Details

23 hours ago

2e99ffc62db2 · 7.7GB ·

model
apertus
·
3.83B
·
BF16
params
{ "num_ctx": 4096, "stop": [ "<SPECIAL_68>", "<SPECIAL_72>", "<SPECI
template
<s> {{- $hasSystem := false -}} {{- $systemContent := "" -}} {{- range $i, $m := .Messages -}} {{- i

Readme

I AM NOT THE ORIGINAL AUTHOR – JUST RE-UPLOADING APERTUS FOR OLLAMA 🤓
CREDITS TO THE APERTUS / SWISS AI INITIATIVE TEAM, SEE BELOW.

Apertus v1.1 - Attribution & Provenance

This is not the original release of Apertus v1.1.

I am not the author of the model and I do not claim ownership of the weights, training, tokenizer, or original chat template.

This Ollama page is simply a community upload intended to make Apertus v1.1 easier to run with Ollama.

Disclaimer

The chat template used here is based on the official Hugging Face chat_template.jinja and was validated against the official tokenizer metadata.

Standard string-based chat prompts match the Hugging Face template byte-for-byte. Some richer Hugging Face-only message shapes are not available through standard Ollama messages, including developer mapping content, user parts, ordered assistant content blocks, and structured parsing of native Apertus tool-call markers.

For the canonical release and the fullest reference behavior, please refer to the official Hugging Face repositories linked below.

Model Variants

The checked Apertus v1.1 instruct variants share the same chat template, tokenizer config, generation config, context size, EOS token, and README sampling recommendation:

  • swiss-ai/Apertus-v1.1-0.5B-Instruct
  • swiss-ai/Apertus-v1.1-1.5B-Instruct
  • swiss-ai/Apertus-v1.1-4B-Instruct

Recommended sampling settings from the official model cards:

  • temperature 0.8
  • top_p 0.9
  • num_ctx 4096

No top_k, min_p, or repeat penalty value is documented in the fetched official metadata.

Official Sources

  1. Original Model / Official Releases
    Apertus by the Swiss AI Initiative, developed as a collaborative effort between EPFL, ETH Zurich, and CSCS.

    Website: https://APERTVS.ai

    Hugging Face: https://huggingface.co/swiss-ai

    Official model repositories:

  2. Ollama Packaging
    This is an unofficial Ollama packaging of Apertus v1.1. The template and parameters were checked against the official Hugging Face metadata. The model weights or GGUF quantization used by a specific Ollama tag should be documented by that tag or uploader.

License

The official Apertus v1.1 instruct releases are published under the Apache 2.0 license.

Please refer to the original model cards and license metadata in the official Hugging Face repositories.

Notes

  • This is an unofficial Ollama upload by MichelRosselli.
  • All credit for the original model belongs to the Apertus / Swiss AI Initiative team.
  • The template is intended for normal Ollama chat use with string-based messages.
  • PARSER passthrough preserves native Apertus markers as raw content; it does not invent structured thinking or tool_calls.
  • For the canonical release and complete reference behavior, please use the official Hugging Face repositories.