CyberNative-AI/Colibri_8b_v0.1_q5_gguf

Llama 3 Colibri 8B v0.1 Q5 GGUF

Built with Meta Llama 3

This is a Q5 GGUF quantization of Llama 3 Colibri 8B v0.1.

Provenance limitation

The source model card previously claimed a training corpus of around 35k Q/A pairs derived from cybersecurity-related texts. CyberNative AI cannot currently evidence the source corpus, source URLs, rights holders, licenses, or transformation records for those pairs. Provenance is unverified. Do not infer a rights basis for those pairs from this card.

Known-risk and acceptable-use notice

Repository records previously reported Dolphin 2.9 Llama 3 8B as the upstream model for the source model, but CyberNative AI cannot currently verify that lineage from a primary training or weight-lineage record. Its public model card describes it as uncensored; conservatively, this model may produce unsafe, inaccurate, or inappropriate cybersecurity content. Do not use it for illegal, malicious, or harmful activity. Use of Meta Llama 3 is subject to the Meta Llama 3 Acceptable Use Policy.

What you download, and how to prompt it

One file, colibri_ggml-model-Q5_K_M.gguf, 5,732,999,488 bytes. The GGUF header reports architecture llama, 8,030,277,632 parameters, a context length of 8192, and an end-of-turn token of <|im_end|>.

Prompts use ChatML, not the stock Llama 3 instruct format:

<|im_start|>system
You are Colibri, an advanced cybersecurity AI assistant developed by CyberNative AI.<|im_end|>
<|im_start|>user
{prompt}<|im_end|>
<|im_start|>assistant

That template is also embedded in the file, so a runner that reads the GGUF chat template applies it for you; if yours does not, set the format above and stop generation on <|im_end|>.

Every number in this section comes from the file's own header:

curl -sL https://huggingface.co/api/models/CyberNative-AI/Colibri_8b_v0.1_q5_gguf | jq .gguf

License and notice

The Meta Llama 3 Community License Agreement is distributed with this repository. The required attribution is in NOTICE.

Custom GGUF quantization

We quantize a model you own — up to 8B parameters — to Q4_K_M, Q5_K_M or Q8_0. You get a loadable GGUF file plus a compatibility note: file size, load test and prompt format. $99 flat, 3 business days.

To request a quote, email hello@cybernative.ai with the subject "GGUF quote" and a link to the model you own. Scope is loadability only: no evaluation report, no accuracy numbers, no provenance review, no fine-tuning.

Contact

CyberNative AI LLC — hello@cybernative.ai

We also run a fixed-price document-to-CSV extraction service: https://cybernative.ai/services/documents-to-csv/