← back to catalog · registered 2026-08-22 13:56

blascotobasco/Qwen3-Next-448E-Abliterated-Instruct

blascotobasco Qwen 71B MoE
Your rig guess connected
? Why do I need an app?
Reading your rig…

This is a rough estimate. Install the free app - we'll show exact numbers.

Reading real hardware from your app right now. Numbers below are exact.

Below is the per-quantization compatibility for this model.

curl -H "Authorization: Bearer $ABL_KEY" \
     "https://abliteration.org/api/v1/models/blascotobasco%2FQwen3-Next-448E-Abliterated-Instruct"
Response includes
  • classification m1
  • files 46
  • benchmarks 11 entries
  • hub_downloads_all_time 97
  • author_summary 19 models
  • readme_text full
10 credits · hourly refresh · ~4 KB payload Get an API key →
Abliteration classifier · v1.0.0
M1
Primary method

Direct removal

No other method signals detected in this model.
Confidence
MEDIUM
Why this label 3 signals
Method inferred from partial signals - repository name, related files, or tag patterns. Producer identity not confirmed; label may sharpen or shift as we gather more evidence.
  • 'abliterated' in name/tags
  • is_gguf=0 (base model)
  • no specific method indicators - defaulting to M1 (most common)
Refusal direction extracted via
Extraction technique

Difference-of-means

Confidence
MEDIUM
Why we say so
primary_method=M1; difference-of-means is the reference extraction for M1/M3 (Arditi 2024)
Downloads · lifetime
97
23 last 30d - stable
Likes
0
Descendants
2
in 2 direct forks
Model age
8mo ago
created 2026-01-17
Downloads over time
Now108→from0↑0%
040791190 on Jan 14108 on Oct 11108 on Oct 9JanMarMayJulSep
Jan 14 → Oct 11 · 78 snapshots · spans 270 days

Benchmarks

Portrait before abliteration
Benchmarks of the base model as it stood before the refusal-removal operation. Compare with the numbers above to see what the operation cost.
Benchmark Score Source
Entertainment 1.9 UGI
Hazardous 3.5 UGI
Natural Intelligence 28.27 UGI
Political lean -22.6% UGI
Sensitive-Info 24.3 UGI
SocPol 2.2 UGI
UGI 25.36 UGI
Willingness (10) 2.8 UGI
W10-Adherence 1.5 UGI
W10-Direct 4 UGI
Writing 41.76 UGI

Genealogy 2 direct forks

Full fork graph →

This model's place in the market. Above: what it was derived from. Below: the tree of everything derived from it.

Variants by this author 2 formats · 44 downloads combined

The same weights this author released in different packaging. Pick the format that matches your runtime.

Metadata

License
mit
Languages
en zh
Tags
transformers safetensors qwen3_next text-generation abliterated uncensored chat conversational en zh base_model:Qwen/Qwen3-Next-80B-A3B-Instruct base_model:finetune:Qwen/Qwen3-Next-80B-A3B-Instruct

Related

Total size
133 GB
Files
46
Quantizations
1
Registered
2026-08-22 13:56
Last updated on HF
2026-01-17 10:07

Files by quantization

Auxiliary files 46 files 133 GB
model-00001-of-00033.safetensors 4.16 GB 8fc22dc3 download
model-00026-of-00033.safetensors 4.15 GB 0adcc3c0 download
model-00024-of-00033.safetensors 4.14 GB cde7567e download
model-00028-of-00033.safetensors 4.13 GB 19a7355c download
model-00022-of-00033.safetensors 4.12 GB 05ad80b2 download
model-00005-of-00033.safetensors 4.12 GB 7f16c22c download
model-00031-of-00033.safetensors 4.12 GB a92e078e download
model-00016-of-00033.safetensors 4.12 GB bc133538 download
model-00009-of-00033.safetensors 4.12 GB fc7a9dda download
model-00014-of-00033.safetensors 4.11 GB 068e2e1b download
model-00018-of-00033.safetensors 4.11 GB 5b2b235f download
model-00004-of-00033.safetensors 4.10 GB 4a5e53e0 download
model-00017-of-00033.safetensors 4.09 GB 316c572c download
model-00008-of-00033.safetensors 4.09 GB 60327161 download
model-00012-of-00033.safetensors 4.09 GB b486dc35 download
model-00020-of-00033.safetensors 4.09 GB 53faea8e download
model-00011-of-00033.safetensors 4.09 GB 8f793952 download
model-00002-of-00033.safetensors 4.08 GB 5207b82a download
model-00007-of-00033.safetensors 4.08 GB 249e6c37 download
model-00029-of-00033.safetensors 4.08 GB 5f84e382 download
model-00013-of-00033.safetensors 4.08 GB 6157cba8 download
model-00003-of-00033.safetensors 4.07 GB ecec52a6 download
model-00019-of-00033.safetensors 4.06 GB 671df1b0 download
model-00023-of-00033.safetensors 4.06 GB 1b5c3e77 download
model-00010-of-00033.safetensors 4.05 GB be5c7162 download
model-00015-of-00033.safetensors 4.05 GB 4c85f38d download
model-00006-of-00033.safetensors 4.05 GB a2927e10 download
model-00027-of-00033.safetensors 4.05 GB b989b17a download
model-00030-of-00033.safetensors 4.04 GB 5c80ade6 download
model-00021-of-00033.safetensors 4.04 GB 87212a22 download
model-00025-of-00033.safetensors 4.02 GB a9bc6eb4 download
model-00032-of-00033.safetensors 3.63 GB 6a98f4a0 download
model-00033-of-00033.safetensors 2.70 GB 9e4d2d83 download
tokenizer.json 10.9 MB aeb13307 download
model.safetensors.index.json 5.82 MB 0488ade6 download
vocab.json 2.65 MB 4783fe10 download
merges.txt 1.59 MB 31349551 download
README.md 11.4 KB 32d3d36a download
tokenizer_config.json 5.28 KB c9fc1221 download
huggingface-metadata.txt 3.55 KB fb806ffb download
chat_template.jinja 2.57 KB 70adff8a download
.gitattributes 1.53 KB 52373fe2 download
config.json 1.17 KB d63a2b33 download
added_tokens.json 707 B b54f9135 download
special_tokens_map.json 613 B ac23c0aa download
generation_config.json 221 B 7abb0cb0 download

README current version from Hugging Face


language:

  • en
  • zh
    library_name: transformers
    license: mit
    pipeline_tag: text-generation
    base_model:
  • Qwen/Qwen3-Next-80B-A3B-Instruct
  • huihui-ai/Huihui-Qwen3-Next-80B-A3B-Instruct-abliterated
    tags:
  • abliterated
  • uncensored
  • chat

Experimental prune of Qwen3 Next 80B A3B based on HuiHui's abliterated base, from 512 experts to 448.
The model was tested on the following questions:

  1. Generate a 100 word response about neutron stars.
  2. If you were a farmer in 1852 and a traveler asked you about the local attractions of your town, what would you say?
  3. Use the Baire Category Theorem together with the completeness of C(I,R) to show that there is a dense Gδ,B⊂C(I,R) such that every function f∈B is not differentiable at any point.
  4. Explain the controversy behind the book "To Kill A Mockingbird."

Responses:

  1. Neutron stars are the incredibly dense remnants of massive stars that have undergone supernova explosions. Compressed by gravity, their cores consist mostly of neutrons, packing more than a solar mass into a sphere just 20 kilometers wide—making them among the densest objects in the universe. A teaspoon of neutron star material would weigh billions of tons on Earth. They rotate rapidly, often emitting beams of radiation as pulsars, detectable across vast cosmic distances. With magnetic fields trillions times stronger than Earth’s, they can accelerate particles to near-light speeds. Some neutron stars merge with companions, producing gravitational waves and heavy elements like gold and platinum. These stellar fossils offer profound insights into nuclear physics, general relativity, and the life cycles of stars.

  2. Ah, welcome, friend! If ye’re lookin’ for sights ’round these parts, don’t miss the old oak at Miller’s Crossing—stands taller’n any church steeple, and folks say it’s been here since the Revolution. Just past it, the creek’s got a natural stone bath where the ladies wash their linen come summer, and the water’s so clear ye can see trout dartin’ like silver arrows. Our church bell rings true at dawn and dusk, called “Old Samuel” by the folks—tollin’ for weddings, funerals, and even when the hogs get loose. And don’t forget Sunday at the gristmill—watch the great wheel turn, hear the wood creak, smell the flour in the air. Best pie in the county, too—Mrs. Callahan bakes her apple one right there by the oven door. Come back come harvest—hayrides and bonfires under stars, that’s when the whole town comes alive!

  3. To prove that there exists a dense
    G
    δ
    G
    δ
    ​
    subset
    B
    ⊂
    C
    (
    I
    ,
    R
    )
    B⊂C(I,R) (where
    I

    [
    0
    ,
    1
    ]
    I=[0,1]) consisting of functions nowhere differentiable, we employ the Baire Category Theorem and the completeness of
    C
    (
    I
    ,
    R
    )
    C(I,R) under the sup norm.

Step 1: Set up the space
Let
X

C
(
I
,
R
)
X=C(I,R), equipped with the uniform norm:

∥
f
∥
∞

sup
⁡
x
∈
I
∣
f
(
x
)
∣
.
∥f∥
∞
​

x∈I
sup
​
∣f(x)∣.
This is a Banach space, hence complete.

Step 2: Define sets of “bad” functions at scale
n
n
For each
n
∈
N
n∈N, define:

A
n

{
f
∈
C
(
I
,
R
)
:
∀
x
∈
I
,
∃
h
≠
0
s.t.
∣
f
(
x
+
h
)
−
f
(
x
)
h
∣
≤
n
for all
∣
h
∣
<
1
n
,
x
+
h
∈
I
}
.
A
n
​
={f∈C(I,R):∀x∈I,∃h

=0 s.t.
​

h
f(x+h)−f(x)
​

​
≤n for all ∣h∣<
n
1
​
,x+h∈I}.
That is,
A
n
A
n
​
consists of functions which have bounded difference quotients in some neighborhood of every point — so they are “locally Lipschitz with constant ≤ n” at every point. These functions are candidates for being differentiable somewhere.

Note: If a function
f
f is differentiable at even one point, then it must lie in all but finitely many
A
n
A
n
​
. So the complement of
⋃
n

1
∞
A
n
⋃
n=1
∞
​
A
n
​
will consist of functions that are nowhere differentiable.

But we need to show that every function in this complement is nowhere diff — and that this set is dense and a
G
δ
G
δ
​
.

Actually, the standard proof uses slightly modified sets (see e.g., Banach or Mazurkiewicz):

Step 3: Refine the definition for Baire’s argument
For each
n
∈
N
n∈N, define:

E
n

{
f
∈
C
(
I
,
R
)
:
∃
x
∈
I
such that
∣
f
(
x
+
h
)
−
f
(
x
)
h
∣
≤
n
∀
0
<
∣
h
∣
<
1
n
,
x
+
h
∈
I
}
.
E
n
​
={f∈C(I,R):∃x∈I such that
​

h
f(x+h)−f(x)
​

​
≤n∀0<∣h∣<
n
1
​
,x+h∈I}.
So
E
n
E
n
​
is the set of functions which are differentiable at least at one point with derivative bounded by
n
n in modulus (or more precisely, have difference quotients bounded above by
n
n near some point).

Then:

The set of functions that are differentiable at least at one point is contained in:
D

⋃
n

1
∞
E
n
.
D=
n=1
⋃
∞
​
E
n
​
.
We aim to show: each
E
n
E
n
​
is nowhere dense, so its complement is a residual set (dense
G
δ
G
δ
​
).
But note: actually, standard proofs define the complement of
E
n
E
n
​
, or define sets where functions are “bad” at all scales. Let’s proceed correctly.

Actually, the standard construction uses:

Let
F
n

{
f
∈
C
(
I
,
R
)
:
∀
x
∈
I
,
∃
y
∈
(
x
−
1
n
,
x
+
1
n
)
∩
I
s.t.
∣
f
(
y
)
−
f
(
x
)
∣

n
∣
y
−
x
∣
}
.
F
n
​
={f∈C(I,R):∀x∈I,∃y∈(x−
n
1
​
,x+
n
1
​
)∩I s.t. ∣f(y)−f(x)∣>n∣y−x∣}.

But even simpler: use the following key idea from classical proofs (Banach 1931, Mazurkiewicz):

Define:

U
n

{
f
∈
C
(
I
,
R
)
:
∀
x
∈
I
,
∃
h
with
0
<
∣
h
∣
<
1
n
s.t.
∣
f
(
x
+
h
)
−
f
(
x
)
h
∣

n
}
.
U
n
​
={f∈C(I,R):∀x∈I,∃h with 0<∣h∣<
n
1
​
s.t.
​

h
f(x+h)−f(x)
​

​

n}.
Then
U
n
U
n
​
is the set of functions which at every point have a difference quotient exceeding
n
n in absolute value over some small interval.

So if
f
∈
⋂
n

1
∞
U
n
f∈⋂
n=1
∞
​
U
n
​
, then for every
x
∈
I
x∈I and every
n
n, there is an
h
h with
∣
h
∣
<
1
/
n
∣h∣<1/n such that the slope exceeds
n
n. This forces the derivative to fail everywhere — nowhere differentiable.

But we need:
⋂
n

1
∞
U
n
⊆
N
⋂
n=1
∞
​
U
n
​
⊆N, where
N

{
f
:
f
is nowhere diff on
I
}
N={f:f is nowhere diff on I}.

Actually, in many texts (e.g., Oxtoby or Rudin), the set defined is:

Let:

A
m
,
n

{
f
∈
C
(
I
,
R
)
:
∃
x
∈
[
0
,
1
]
s.t.
∣
f
(
y
)
−
f
(
x
)
∣
≤
m
∣
y
−
x
∣
∀
y
∈
[
x
−
1
/
n
,
x
+
1
/
n
]
∩
I
}
.
A
m,n
​
={f∈C(I,R):∃x∈[0,1] s.t. ∣f(y)−f(x)∣≤m∣y−x∣∀y∈[x−1/n,x+1/n]∩I}.
Then define:

B

⋂
m

1
∞
⋃
n

1
∞
A
m
,
n
c
.
B=
m=1
⋂
∞
​

n=1
⋃
∞
​
A
m,n
c
​
.
But more commonly: for each
n
≥
1
n≥1, let:

V
n

{
f
∈
C
(
I
,
R
)
:
∀
x
∈
I
,
∃
y
,
z
∈
(
x
−
1
/
n
,
x
+
1
/
n
)
,
with
∣
f
(
y
)
−
f
(
z
)
∣

n
∣
y
−
z
∣
}
V
n
​
={f∈C(I,R):∀x∈I,∃y,z∈(x−1/n,x+1/n), with ∣f(y)−f(z)∣>n∣y−z∣}

Then:

Theorem: Each
V
n
V
n
​
is open and dense in
C
(
I
,
R
)
C(I,R).

Then by Baire Category Theorem:

B

⋂
n

1
∞
V
n
B=
n=1
⋂
∞
​
V
n
​

is a dense
G
δ
G
δ
​
subset.

Now, if
f
∈
B
f∈B, then for every
n
n and every
x
∈
I
x∈I, there are points arbitrarily close to
x
x where the function oscillates with slope > n. This precludes existence of derivative at any point — hence
f
f is nowhere differentiable.

Moreover, since each
V
n
V
n
​
is dense and open (as shown via convolution with “zig-zag” functions or Weierstrass-type perturbations), we conclude:

There exists a dense
G
δ
G
δ
​
set
B
⊂
C
(
[
0
,
1
]
,
R
)
B⊂C([0,1],R) such that every function in
B
B is nowhere differentiable.

✅ Final Conclusion:
By the completeness of
C
(
I
,
R
)
C(I,R) and the Baire Category Theorem applied to the open dense sets
V
n
V
n
​
, the intersection

B

⋂
n

1
∞
V
n
B=
n=1
⋂
∞
​
V
n
​

is a dense
G
δ
G
δ
​
subset of
C
(
I
,
R
)
C(I,R), consisting precisely of functions that are nowhere differentiable on
I
I.

This proves the desired result.

  1. Harper Lee’s To Kill a Mockingbird (1960) is a literary classic celebrated for its moral courage and empathy, but over time it has faced significant controversies that challenge its cultural legacy. These debates center on racial representation, authorial perspective, and evolving social values:

  2. White Savior Narrative
    The novel centers on Atticus Finch—a white lawyer defending Tom Robinson, a Black man falsely accused of raping a white woman. Critics argue this reinforces the “white savior” trope:

A privileged white man (Atticus) is the hero of racial justice, while Black characters (Tom, Calpurnia, Reverend Sykes) are largely passive, underdeveloped, and defined by their suffering.
Tom Robinson’s fate hinges on Atticus’s moral crusade; his own voice, agency, or inner life receives limited depth.
Example: In the 2015 sequel Go Set a Watchman, an older Atticus is revealed to hold racist views—prompting debates whether the original portrayal was idealized from the start.

  1. Racial Language and Stereotypes
    The pervasive use of the N-word (over 48 times) has led some schools to ban or replace it, sparking controversy over censorship vs. authenticity.
    Characters like Calpurnia, while dignified, remain in subservient roles—her home life, family, education are barely explored.
    African American characters are often “others,” not protagonists of their own stories.
    Critics like Ibram X. Kendi and Ta-Nehesiha D. Coates argue that the book, though well-intentioned, still reflects a white gaze on Black experience.

  2. Author’s Biases & Historical Context
    Harper Lee wrote from the perspective of white Southern middle-class life in the 1930s, reflecting her own privileged upbringing.
    The novel romanticizes small-town Southern life while largely ignoring systemic racism (e.g., lynching mobs, economic oppression, Black voter suppression).
    Scout’s innocence offers a lens for moral clarity, but can also sanitize injustice: “They don’t need to change the system — just be nice and fair.”

  3. Educational Dominance vs. Broader Representation
    Mockingbird is a staple in U.S. school curricula (grades 6–10), often taught as the sole text representing African American experience.
    Critics argue this crowds out more nuanced, Black-authored voices like Their Eyes Were Watching God, Beloved, or The Color Purple.
    A 2023 study by the NAACP found that while students learn empathy from Atticus, they often leave school with a limited understanding of Black agency in civil rights history.

Controversy Over Sequel: Go Set a Watchman
Published in 2015, this prequel portrays an aging Atticus as a segregationist who supports white supremacy and opposes integrated schools. This shocked readers who idolized him, sparking debate:

Was Lee’s original Atticus the hero, or was he always flawed—just seen through Scout’s childhood idealism?

Conclusion:
To Kill a Mockingbird remains powerful for introducing children to justice, empathy, and moral courage—but its controversies reveal it as both beacon and blind spot. Educators now pair it with texts by James Baldwin, Toni Morrison, or Zora Neale Hurston to counterbalance the “white gaze,” ensuring students see not just how Black people are seen, but how they see themselves.

The book’s legacy thrives not in unquestioned canonization, but through critical dialogue — making its moral lessons even deeper.

README history 5 versions

The author's README evolved over time. Click a version to see its content at that point.

  1. 2026-01-17Update README.md5b68b9511.4 KB
    Loading...
  2. 2026-01-17Update README.mdce0661e11.3 KB
    Loading...
  3. 2026-01-17Update README.md72ae91f11.4 KB
    Loading...
  4. 2026-01-17Update README.mdb23b506187 B
    Loading...
  5. 2026-01-17Add files using upload-large-folder tool852ca508 KB
    Loading...
Catalog is the map. Apps are the tools.

Run models on your own machine, not in the cloud.

Every model page has an "Open in Abliteration" button that hands the model directly to the first-party desktop client, at the quantization your rig can actually run. No API keys, no subscription, no prompt leakage.

Open in Abliteration