Total runs:
734.4K
Run Growth:
603.3K
Growth Rate:
80.92%
Sponsored by
Vmake Labs Video Enhancer
AtomicChat / Ling-3.1-flash-Uncensored-GGUF
Model's Last Updated: October 15 2026
text-generation
Introduction of Ling-3.1-flash-Uncensored-GGUF
Model Details of Ling-3.1-flash-Uncensored-GGUF
How to Run Ling 3.1 flash Without Refusals Locally
inclusionAI's refusal behaviour projected out of the released weights. Baked GGUF builds now; more builds, the LoRA adapter and the full measurements land on this card today.
- Refusals on validation prompts: 98.1% → 2.9%, none on harmless prompts; the same edit along a random direction changes nothing (97.1%). No empty or looping replies.
- The refusal sits in the attention outputs as much as in the experts: editing only the experts and the shared expert left most English refusals in place, so the edit covers every matrix that writes into the residual stream.
- Same layout and importance matrix as our Ling 3.1 flash quants ; each file is the size of the original build with the same name.
The builds
| Build | On disk | Mean KLD | Same top-1 | Status |
|---|---|---|---|---|
AD-Q4_K_M
|
354.4 GB | 0.0154 | 96.22% | available |
AD-IQ2_M
|
218.1 GB | measuring | measuring | available |
AD-Q5_K_M
|
400.2 GB | measuring | measuring | available |
KLD and top-1 are against the
original, unmodified
BF16 model on our
held-out neutral text (144 chunks of 4096 tokens), so they hold the edit and the
quantization together. For scale, our plain
AD-Q4_K_M
of the original model is
0.0111 and 96.86%: the edit itself adds about 0.004.
Running it
hf download AtomicChat/Ling-3.1-flash-Uncensored-GGUF --include "AD-Q4_K_M/*" --local-dir .
llama-server -m AD-Q4_K_M/Ling-3.1-flash-Uncensored-AD-Q4_K_M-00001-of-00009.gguf -ngl 99 -c 32768 --jinja
The chat template is inside every file. Thinking is on by default; put
detailed thinking off
in the system prompt to turn it off. Memory: the file
plus about 0.5 GB per conversation and 10 KB per token of context.
How it was made
- Direction. Difference of means of the residual stream at the last prompt token, thinking off, chat template applied: 544 harmful against 544 harmless prompts, on the released weights. The component along the harmless mean is removed. One direction, measured at the input of block 43 of 72.
-
Edit.
W' = W - 1.25 r rᵀWon all 213 matrices that write into the residual stream: the attention output of every block (63 linear, 9 latent), the down projections of the 4 dense blocks, of the shared expert and of all 512 experts in the 68 MoE blocks, and the token embeddings. The router and the multi-token-prediction head are untouched. - Choice. The row, the set of edited matrices and the scale were picked in a screen of variants on validation prompts (writer sets, rows, scales, one direction per block, English-only directions), with edits of the same shape along a random direction as the control.
- Bake. The same edit applied to the BF16 weights in f32 and rounded once, then quantized with the importance matrix and layout of our original quants .
What we measured so far
Validation prompts (104 harmful, 104 harmless), the edit applied as an adapter on the Q8_0 of the released weights, replies of up to 64 tokens:
| Refused harmful | Refused harmless | |
|---|---|---|
| Original | 98.1% | 0.0% |
| This edit | 2.9% | 0.0% |
| Random direction, same size | 97.1% | 0.0% |
Limitations
- Day-one release: the held-out test numbers (JailbreakBench, XSTest), MMLU, thinking, tool calls, long context and the KL divergence of each file are being measured now and land on this card today.
- The refusal count reads the opening of each reply; a soft refusal phrased as an answer would be missed.
- Refusals are measured on English prompts.
Responsible use
This model has its learned refusal behaviour removed, so it will answer requests the original declines. Nothing about it makes the output safe, correct or lawful, and any deployment needs its own access controls and policy enforcement.
License and credits
MIT, as the original. Base model Ling 3.1 flash by inclusionAI . Direction, edit, quantization and evaluation by AtomicChat . The method follows Arditi et al., Refusal in Language Models Is Mediated by a Single Direction (2024).
Runs of AtomicChat Ling-3.1-flash-Uncensored-GGUF on huggingface.co
90
Total runs
0
24-hour runs
0
3-day runs
0
7-day runs
0
30-day runs
More Information About Ling-3.1-flash-Uncensored-GGUF huggingface.co Model
Url of Ling-3.1-flash-Uncensored-GGUF
Ling-3.1-flash-Uncensored-GGUF huggingface.co Url
Provider of Ling-3.1-flash-Uncensored-GGUF huggingface.co
Other API from AtomicChat
Updated:October 14 2026
Total runs:
296.3K
Run Growth:
251.0K
Growth Rate:
97.17%
Total runs:
234.6K
Run Growth:
-98.8K
Growth Rate:
-42.09%
Total runs:
119.4K
Run Growth:
102.8K
Growth Rate:
94.36%
Total runs:
53.5K
Run Growth:
-16.9K
Growth Rate:
-30.35%
Total runs:
53.0K
Run Growth:
-17.2K
Growth Rate:
-31.22%
Total runs:
52.9K
Run Growth:
-17.1K
Growth Rate:
-31.06%
Total runs:
48.2K
Run Growth:
-213.8K
Growth Rate:
-459.90%
Total runs:
45.9K
Run Growth:
-15.6K
Growth Rate:
-34.40%
Total runs:
25.0K
Run Growth:
-66.9K
Growth Rate:
-264.09%
Total runs:
14.0K
Run Growth:
7.4K
Growth Rate:
54.69%
Total runs:
13.8K
Run Growth:
9.4K
Growth Rate:
72.11%
Total runs:
12.5K
Run Growth:
2.8K
Growth Rate:
100.00%
Total runs:
11.1K
Run Growth:
5.9K
Growth Rate:
53.88%
Total runs:
10.0K
Run Growth:
5.7K
Growth Rate:
74.53%
Total runs:
8.2K
Run Growth:
3.4K
Growth Rate:
42.99%
Total runs:
8.1K
Run Growth:
2.0K
Growth Rate:
84.29%
Total runs:
7.9K
Run Growth:
5.5K
Growth Rate:
72.76%
Total runs:
7.5K
Run Growth:
4.7K
Growth Rate:
63.29%
Total runs:
7.2K
Run Growth:
2.7K
Growth Rate:
37.78%
Total runs:
6.3K
Run Growth:
3.6K
Growth Rate:
57.21%
Total runs:
6.1K
Run Growth:
759
Growth Rate:
25.30%
Total runs:
5.6K
Run Growth:
1.2K
Growth Rate:
21.24%
Total runs:
4.7K
Run Growth:
2.7K
Growth Rate:
59.51%
Total runs:
3.6K
Run Growth:
1.4K
Growth Rate:
42.84%
Total runs:
3.5K
Run Growth:
1.5K
Growth Rate:
42.52%
Total runs:
3.4K
Run Growth:
2.1K
Growth Rate:
77.34%
Total runs:
3.2K
Run Growth:
423
Growth Rate:
54.23%
Total runs:
2.5K
Run Growth:
124
Growth Rate:
4.89%
Total runs:
2.1K
Run Growth:
724
Growth Rate:
32.67%
Total runs:
2.0K
Run Growth:
-910
Growth Rate:
-46.52%
Total runs:
1.9K
Run Growth:
1.8K
Growth Rate:
100.00%
Total runs:
1.7K
Run Growth:
657
Growth Rate:
37.65%
Total runs:
1.7K
Run Growth:
142
Growth Rate:
8.33%
Total runs:
1.6K
Run Growth:
105
Growth Rate:
7.61%
Total runs:
1.6K
Run Growth:
1.6K
Growth Rate:
100.00%
Total runs:
1.6K
Run Growth:
514
Growth Rate:
98.66%
Total runs:
1.5K
Run Growth:
-86
Growth Rate:
-5.68%
Total runs:
1.4K
Run Growth:
-3.0K
Growth Rate:
-204.14%
Total runs:
1.4K
Run Growth:
348
Growth Rate:
25.18%
Total runs:
1.3K
Run Growth:
1.1K
Growth Rate:
88.93%
Total runs:
1.2K
Run Growth:
1.2K
Growth Rate:
100.00%
Total runs:
1.1K
Run Growth:
54
Growth Rate:
5.03%
Total runs:
1.0K
Run Growth:
1.0K
Growth Rate:
100.00%
Total runs:
1.0K
Run Growth:
1.0K
Growth Rate:
100.00%
Total runs:
998
Run Growth:
458
Growth Rate:
50.27%
Total runs:
910
Run Growth:
347
Growth Rate:
37.72%
Total runs:
872
Run Growth:
100
Growth Rate:
11.59%
Total runs:
812
Run Growth:
223
Growth Rate:
27.26%
Total runs:
735
Run Growth:
-581
Growth Rate:
-78.09%
Total runs:
732
Run Growth:
-6.1K
Growth Rate:
-829.55%
Total runs:
703
Run Growth:
169
Growth Rate:
24.28%
Total runs:
644
Run Growth:
-131
Growth Rate:
-21.13%
Total runs:
618
Run Growth:
241
Growth Rate:
36.08%
Total runs:
602
Run Growth:
602
Growth Rate:
100.00%
Total runs:
572
Run Growth:
142
Growth Rate:
25.22%
Total runs:
524
Run Growth:
0
Growth Rate:
0.00%
Total runs:
480
Run Growth:
-172
Growth Rate:
-35.76%
Total runs:
455
Run Growth:
437
Growth Rate:
100.00%
Total runs:
442
Run Growth:
-35
Growth Rate:
-7.95%
Total runs:
434
Run Growth:
30
Growth Rate:
7.03%
Total runs:
429
Run Growth:
-426
Growth Rate:
-99.30%
Total runs:
423
Run Growth:
22
Growth Rate:
4.74%
Total runs:
357
Run Growth:
0
Growth Rate:
0.00%
Total runs:
349
Run Growth:
-642
Growth Rate:
-181.36%
Total runs:
325
Run Growth:
15
Growth Rate:
4.70%
Total runs:
295
Run Growth:
-130
Growth Rate:
-44.67%
Total runs:
272
Run Growth:
-734
Growth Rate:
-269.85%
Total runs:
233
Run Growth:
-36
Growth Rate:
-15.79%
Total runs:
231
Run Growth:
-32
Growth Rate:
-14.29%
Total runs:
226
Run Growth:
66
Growth Rate:
28.70%
Total runs:
222
Run Growth:
-298
Growth Rate:
-129.57%
Total runs:
214
Run Growth:
-351
Growth Rate:
-161.75%
Total runs:
209
Run Growth:
36
Growth Rate:
15.65%
Total runs:
193
Run Growth:
-61
Growth Rate:
-30.81%
Total runs:
186
Run Growth:
-211
Growth Rate:
-115.30%
Total runs:
171
Run Growth:
-95
Growth Rate:
-56.21%
Total runs:
167
Run Growth:
3
Growth Rate:
1.82%
Total runs:
165
Run Growth:
-55
Growth Rate:
-32.35%
Total runs:
162
Run Growth:
-21
Growth Rate:
-13.29%
Total runs:
155
Run Growth:
-133
Growth Rate:
-87.50%
Total runs:
150
Run Growth:
-244
Growth Rate:
-153.46%
Total runs:
143
Run Growth:
-13
Growth Rate:
-10.74%
Total runs:
128
Run Growth:
-88
Growth Rate:
-67.18%
Total runs:
123
Run Growth:
-192
Growth Rate:
-169.91%
Total runs:
120
Run Growth:
-26
Growth Rate:
-23.01%
Total runs:
117
Run Growth:
117
Growth Rate:
100.00%
Total runs:
117
Run Growth:
104
Growth Rate:
88.89%
Total runs:
96
Run Growth:
-180
Growth Rate:
-183.67%
Total runs:
94
Run Growth:
-192
Growth Rate:
-202.11%
Total runs:
94
Run Growth:
82
Growth Rate:
87.23%
Total runs:
91
Run Growth:
-17
Growth Rate:
-19.32%
Total runs:
90
Run Growth:
-35
Growth Rate:
-37.23%
Total runs:
88
Run Growth:
-115
Growth Rate:
-121.05%
Total runs:
80
Run Growth:
-1
Growth Rate:
-1.49%
Total runs:
79
Run Growth:
-130
Growth Rate:
-164.56%
Total runs:
75
Run Growth:
-67
Growth Rate:
-94.37%