Downloads · 30 days
36
51% of all-time downloads
Walkowicz/kairos-sotos-1.0-14B
kairos-sotos-1.0-14B is a text generation model from Walkowicz. Use it when you need the model to write or continue text. The card lists the license as apache-2.0.
<p align="center" <img src="kairos-logo.png" alt="Kairos" width="720" </p
Downloads · 30 days
36
51% of all-time downloads
All-time downloads
70
Public
Repo size
12.1 GB
Likes
0
Public
Click a slice to open those files.
.gguf12.1 GB · 100%
From the Hugging Face model README
Local security and privacy auditor. It reports the finding, the smallest safe fix, and the CWE or OWASP id. It does not provide an exploit payload or attack steps. Fine-tune of Qwen/Qwen3-14B. This repository is the shipped Q6_K GGUF. Q4_K_M of the same merge drops below the gate, so Q6_K is the floor.
The file is 11.3 GiB (6.56 BPW). On a 12 GB GPU with num_ctx 2048, about 9.9 GiB stays on the GPU and the rest is offloaded, at about 10.8 tok/s. It does not fit an 8 GB GPU.
Both judges scored these Q6 generations, not only the adapter.
| In-domain | Held-out | |
|---|---|---|
| Grok 4.7 | 18/20 | 16/20 |
| Opus 5.0 | 17/20 | 17/20 |
Family bars are 80% in-domain and 70% held-out. Replies in the user's language (EN or PT). Code identifiers stay in English.
Download kairos-sotos-1.0-14B-Q6_K.gguf and Modelfile into the same folder, then:
ollama create kairos-sotos:1.0 -f Modelfile
ollama run kairos-sotos:1.0
The Modelfile sets ChatML stops, num_ctx 2048, temperature 0, and an empty <think> block before the answer. Qwen3 needs that prefix. A bare ollama pull of the GGUF uses the chat template baked into the file and will not match this deploy.
Apache 2.0, same as the Qwen3-14B base. This repository does not include the training set.