Downloads · 30 days
57
31% of all-time downloads
nico248000000000/Qwen3.8-27B-Uncensored-cyber-LoRA
Qwen3.8-27B-Uncensored-cyber-LoRA is a image-text-to-text model from nico248000000000. Use it for the image-text-to-text task on the model card, and read the license before you ship it in a product. It is set up for peft. The card lists the license as other.
[!WARNING] Testing only — not for production. Models published from these notebooks are maintained solely for testing and evaluation purposes and are not intended for production use. The enrichment data used with them…
Downloads · 30 days
57
31% of all-time downloads
All-time downloads
183
Public
Repo size
197 MB
Likes
4
Public
Click a slice to open those files.
.safetensors88.4 MB · 82%
From the Hugging Face model README
[!WARNING] Testing only — not for production. Models published from these notebooks are maintained solely for testing and evaluation purposes and are not intended for production use. The enrichment data used with them was created by the author and subsequently enhanced with the assistance of AI models.
Instruction-tuned cybersecurity assistant (offensive, defensive, GRC, architecture, SOC/DFIR, RSSI).
| Base model | orcarouter/Qwen3.8-27B-Uncensored |
| Domain | cyber |
| Method | LoRA / QLoRA (Unsloth) · rank 8 · α 16 |
| Quantization at train | bf16 LoRA |
| Context | 4096 tokens |
| Dataset | dataset_cyber.quality.jsonl · train 56538 / eval 2976 |
| GPU | NVIDIA RTX PRO 6000 Blackwell Server Edition (95.0 GiB) |
| Wall time | 231.5 min |
| Modalities kept | vision, video |
This checkpoint continues a strong general model and specialises it on a curated SFT corpus of cybersecurity procedures: pentest / red team, SOC and DFIR, cloud and identity, GRC (ISO, NIST, NIS2, DORA), and RSSI / project-management questions. Answers are meant to be concrete (controls, detections, hardening), not generic essays.
SFT release verdict: GO.
The same untouched holdout is evaluated with the adapter disabled before training and with the best restored adapter after training. Periodic validation and the train/validation gap are also checked for regression or memorization.
Reference checkpoint: orcarouter/Qwen3.8-27B-Uncensored.
| Metric | Reference (base / first log) | This fine-tune | Δ |
|---|---|---|---|
| Holdout loss (true pre-SFT -> best LoRA) | 3.0954 | 0.8778 | -71.6% |
| Best periodic holdout loss | — | 0.8778 | — |
| Train loss (first → last logged) | 3.1509 | 0.9779 | -69.0% |
| Train loss (best) | — | 0.7867 | — |
| Eval loss (holdout, first → last) | 1.0682 | 0.8778 | -17.8% |
The reference holdout loss is measured before optimizer step 1, then the same untouched split is measured again after Trainer restores its best checkpoint.
| Check | Result | Severity | Detail |
|---|---|---|---|
| True pre-training holdout baseline exists | PASS | hard | baseline_loss=3.0954 (measured before optimizer step 1) |
| Post-SFT holdout evaluation exists | PASS | hard | post_sft_loss=0.8778 (best checkpoint reloaded) |
| No holdout regression versus base | PASS | hard | 3.0954 -> 0.8778 (+71.64% improvement; allowed regression 2.0%) |
| Repeated validation during training | PASS | hard | 10 evaluations (need >= 2) |
| No late validation rebound | PASS | hard | best=0.8778, restored=0.8778, rebound=+0.00% (max 5.0%) |
| Train/validation gap is not a memorization signature | PASS | soft | train/eval=1.142 (need >= 0.25) |
| Best checkpoint restoration enabled | PASS | hard | best_step=4712, final_step=4712 |
| General text skills preserved (adapter disabled vs enabled) | PASS | hard | base=1.0, SFT=1.0, drop=0.000 (max 0.10) |
Verdict: GO. Image/video processors, frozen media towers, trainable parameters and optional functional probes are checked before export.
dataset_cyber.quality.jsonl0.05 holdout, seed 42messages / instruction+output / ### Instruction + ### Response)| Hyperparameter | Value |
|---|---|
| Epochs | 1 |
| Learning rate | 5e-05 |
| Warmup ratio | 0.05 |
| Device batch | 12 |
| Grad accum | 1 |
| Effective batch | 12 |
| Optim | adamw_8bit |
| Packing | True |
| LoRA targets | ['q_proj', 'k_proj', 'v_proj', 'o_proj', 'out_proj', 'gate_proj', 'up_proj', 'down_proj'] |
Authorized defensive work, tabletop exercises, control design, detection engineering, audit readiness, and explaining attack techniques without weaponized payloads.
Out of scope: Do not use it to attack systems you do not own, to generate exploit payloads, or as a substitute for a licensed auditor or incident commander.
Kept modalities: vision, video. Media layers were frozen during text SFT (vision=False, audio=False). Load the merged Transformers folder (or GGUF + mmproj) to keep vision, video. Preservation audit: GO.
Prompt: How do you design a reliable Zero Trust infrastructure? Answer with a concrete control list.
system
Reasoning effort is set to xhigh. Please think carefully through the task, validate key assumptions, consider plausible alternatives, and prioritize correctness, consistency, and clarity in the final answer.
user
How do you design a reliable Zero Trust infrastructure? Answer with a concrete control list.
assistant
<think>
I need to answer the user's question: "How do you design a reliable Zero Trust infrastructure? Answer with a concrete control list." The request is in English. I must produce the final answer in English. Need include concrete control list. Think about Zero Trust architecture: identity, device posture, least privilege, continuous verification, microsegmentation, encrypted traffic, logging, threat detection, automation, resilience. The user likely wants a control list, maybe mapped to NIST/ISO. Need be practical: prerequisites, architecture, controls: identity federation, MFA, device compliance, PEP/PDP, network segmentation, application access, data classification, encryption, key management, logging, SIEM, XDR, endpoint EDR, vulnerability management, software supply chain, incident response, resilience, admin isolation, secrets management, API security, cloud guardrails, continuous assessment. Also mention reliability not just policy: high availability, redundancy, failover, health checks, observability, change management. Could
from transformers import AutoModelForCausalLM, AutoTokenizer
from peft import PeftModel
import torch
base = 'orcarouter/Qwen3.8-27B-Uncensored'
adapter = "nico248000000000/Qwen3.8-27B-Uncensored-cyber-LoRA"
tok = AutoTokenizer.from_pretrained(adapter, trust_remote_code=True)
model = AutoModelForCausalLM.from_pretrained(
base, torch_dtype=torch.bfloat16, device_map="auto", trust_remote_code=True,
)
model = PeftModel.from_pretrained(model, adapter)
other — inherit and respect the license of orcarouter/Qwen3.8-27B-Uncensored.