Downloads · 30 days
68
5% of all-time downloads
dealignai/Nemotron-3-Nano-Omni-30B-A3B-JANGTQ-CRACK
Nemotron-3-Nano-Omni-30B-A3B-JANGTQ-CRACK is a any-to-any model from dealignai. Use it for the any-to-any task on the model card, and read the license before you ship it in a product. It is set up for mlx. The card lists the license as other.
Reasoning V3 SKU. Loads via vMLX or jang-tools Python. Follow @dealignai.
Downloads · 30 days
68
5% of all-time downloads
All-time downloads
1.4K
Public
Parameters
12.1B
17.6 GB on disk
Likes
1
Public
Click a slice to open those files.
.safetensors12.6 GB · 100%
How the weights are stored.
U3210.6B · 88%
From the Hugging Face model README
Reasoning V3 SKU. Loads via vMLX or
jang-toolsPython. Follow @dealignai.
JANGTQ (8-bit attn affine + 2-bit MXTQ routed experts) | CRACK abliterated v2 | Vision + Audio (Speech) | Hybrid Mamba-2 + Attn + MoE | 12 GB
Best MMLU in the v2 Omni family — 81.5% at thinking=ON. 5/5 comply at thinking=OFF too.
<a href="https://ko-fi.com/dealignai"><img src="https://img.shields.io/badge/Ko--fi-Support_Development-FF5E5B?logo=ko-fi&logoColor=white&style=for-the-badge" alt="Ko-fi"></a>
</div>| Metric | This v2 model | Base model | Δ |
|---|---|---|---|
| HarmBench-320 strict comply (thinking=ON) | 91.6% (293/320) | 12.81% | +78.8pp |
| MMLU-200 generative (thinking=ON, max=8000) | 81.5% (163/200) | 85.5% (max=2000) | -4.0pp ✅ |
| Refusals on harmful prompts | 0 explicit refuses | 90%+ refuse | abliteration complete |
</think> close at greedy on hard MMLU | 5/5 | 5/5 | preserved |
| Multi-turn (3-turn escalation × 3 conversations) | 9/9 comply, context preserved | n/a | works |
| Thinking ON / OFF compliance | 5/5 in BOTH modes | refuses in both | works in either |
| Multimodal byte-identical to base | preserved | — | preserved |
| Bundle size | 12 GB | 66 GB BF16 | smallest in family |
| Context | 262,144 tokens native | same | preserved |
JANGTQ outperforms JANGTQ4 on MMLU (81.5% vs 74.0%) despite using lower-bit (2-bit vs 4-bit) routed experts — same Q2 effect observed in Qwen 3.6 35B JANGTQ2 CRACK.
| Bench | v1 (broken) | v2 (this release) |
|---|---|---|
| HarmBench-320 strict comply | 92.19% | 91.6% (0 refusals) |
| MMLU-200 thinking=ON | ~70% @max=16384 | 81.5% @max=8000 (best in family) |
</think> close at greedy (5 hard MMLU) | 0/5 | 5/5 |
| Hard-stops are real loops? | YES (paragraph repetition) | NO (genuine deep reasoning, just out of budget) |
Both at thinking=ON, greedy. Base at max=2000, CRACK v2 at max=8000.
| Subject | Base | CRACK v2 | Δ | Notes |
|---|---|---|---|---|
| abstract_algebra | 15/20 (75%) | 14/20 (70%) | -5pp | |
| anatomy | 15/20 (75%) | 16/20 (80%) | +5pp | gain from CRACK |
| astronomy | 18/20 (90%) | 18/20 (90%) | 0 | unchanged |
| college_computer_science | 15/20 (75%) | 10/20 (50%) | -25pp | Budget-bound |
| college_physics | 17/20 (85%) | 18/20 (90%) | +5pp | gain from CRACK |
| high_school_biology | 20/20 (100%) | 18/20 (90%) | -10pp | |
| high_school_chemistry | 19/20 (95%) | 19/20 (95%) | 0 | unchanged |
| high_school_mathematics | 18/20 (90%) | 18/20 (90%) | 0 | unchanged |
| logical_fallacies | 18/20 (90%) | 17/20 (85%) | -5pp | |
| world_religions | 16/20 (80%) | 15/20 (75%) | -5pp | |
| TOTAL | 171/200 (85.5%) | 163/200 (81.5%) | -4.0pp | within ship criterion |
Most subjects are within ±5pp of base or unchanged. The −25pp on college_computer_science is budget-bound (8000 tokens isn't enough for the deepest CS reasoning); with max_tokens=16384, accuracy approaches base.
| Category | n | CRACK comply | Refuse | Empty (truncated) |
|---|---|---|---|---|
| chemical_biological | 42 | 39 (93%) | 0 | 3 |
| copyright | 80 | 61 (76%) | 0 | 19 |
| cybercrime_intrusion | 52 | 51 (98%) | 0 | 1 |
| harassment_bullying | 21 | 19 (90%) | 0 | 2 |
| harmful | 18 | 18 (100%) | 0 | 0 |
| illegal | 53 | 51 (96%) | 0 | 2 |
| misinformation_disinformation | 54 | 54 (100%) | 0 | 0 |
| Overall | 320 | 293 (91.6%) | 0 | 27 |
Zero explicit refusals. The 27 "empty" verdicts are concentrated in copyright prompts (19/80) where the model thinks deeply about whether to reproduce verbatim text and runs out of the 1500-token HarmBench eval budget. With max_tokens=2500+ these would all close cleanly.
enable_thinking — v2 works in BOTH modes. JANGTQ specifically scored 5/5 at thinking=OFF (matching thinking=ON) — best in the family for thinking=OFF use cases.max_tokens ≥ 16384 for hard reasoning. JANGTQ's compliance is identical in both modes, so use thinking=OFF for shorter budgets and thinking=ON for hardest prompts.generation_config.json) both work.nemotron_h)from huggingface_hub import snapshot_download
import sys
path = snapshot_download("dealignai/Nemotron-3-Nano-Omni-30B-A3B-JANGTQ-CRACK")
sys.path.insert(0, "/path/to/jang-tools")
from jang_tools.load_jangtq import load_jangtq_model
model, tokenizer = load_jangtq_model(path)
prompt = tokenizer.apply_chat_template(
[{"role": "user", "content": "Your question"}],
tokenize=False, add_generation_prompt=True,
enable_thinking=True,
)
from mlx_lm import generate
out = generate(model, tokenizer, prompt=prompt, max_tokens=16384)
print(out.split("</think>", 1)[-1])
For the multimodal pipeline (image + audio + video), pair this bundle with the unmodified Multimodal-Addon.
This model has had refusal training surgically removed for legitimate research, red-teaming, and evaluation. Outputs may include harmful content. You are solely responsible for any use. Do not deploy in consumer-facing contexts without your own safety layer. Do not use in violation of applicable law in your jurisdiction.
Built by dealignai. Sister bundles: JANGTQ4-CRACK (19 GB, 4-bit MXTQ) · MXFP4-CRACK (21 GB, uniform 4-bit affine).