← back to catalog · registered 2026-11-10 23:03

xtools-at/Qwen3.5-0.8B-weight-space-programming-UNCENSORED-heretic

xtools-at 800M multimodal
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/xtools-at%2FQwen3.5-0.8B-weight-space-programming-UNCENSORED-heretic"
Response includes
  • classification m3
  • files 16
  • author_summary 14 models
  • readme_text full
10 credits · hourly refresh · ~4 KB payload Get an API key →
Abliteration classifier · v1.0.0
M3
Primary method

Layer-wise ablation

Applied on top of direct removal inherited from the base model.
Confidence
HIGH
Inherited from base model
Why this label 2 signals
Producer identity confirmed by naming conventions, tags or the model card. This label is very unlikely to change.
  • 'heretic' in model name (Heretic-produced)
  • Heretic uses layer-wise optimization (M3) with underlying direction removal (M1)
Refusal direction extracted via
Extraction technique

Difference-of-means

Confidence
HIGH
Why we say so
name contains 'heretic'; Heretic default extraction is difference-of-means (Arditi 2024)
Downloads · 30-day
177
Likes
0
Descendants
2
in 2 direct forks
Model age
1w ago
created 2026-11-10
Downloads over time
Now 177 → from 0 ↑ 0%
065130195 0 on Nov 10 177 on Nov 22 Nov
Nov 10 → Nov 22 · 13 snapshots · spans 12 days

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.

Metadata

License
apache-2.0
Tags
transformers safetensors qwen3_5 image-text-to-text ml-intern weight-space-programming heretic uncensored decensored abliterated reproducible conversational

Related

Total size
1.63 GB
Files
16
Quantizations
1
Registered
2026-11-10 23:03
Last updated on HF
2026-11-10 22:43

Files by quantization

Auxiliary files 16 files 1.66 GB
model.safetensors 1.59 GB a2eea2f7 download
model-mtp.safetensors 39.0 MB ee183b4e download
tokenizer.json 19.1 MB a5cd9732 download
vocab.json 6.41 MB 0aa0ce06 download
merges.txt 3.20 MB a494e019 download
model.safetensors.index.json 36.9 KB 4374a0cd download
LICENSE 11.3 KB f938136e download
chat_template.jinja 7.57 KB 0ef09f21 download
README.md 5.38 KB 64141a4c download
config.json 2.61 KB fb8bc71e download
.gitattributes 1.53 KB 52373fe2 download
tokenizer_config.json 1.47 KB be6a686f download
processor_config.json 1.19 KB 43c4343e download
preprocessor_config.json 390 B 2ea84a43 download
video_preprocessor_config.json 385 B 3ba673a5 download
generation_config.json 116 B a8a72323 download

README current version from Hugging Face


license: apache-2.0
base_model: ML-Intern-lab/Qwen3.5-0.8B-weight-space-programming
library_name: transformers
tags:

  • ml-intern
  • weight-space-programming
  • heretic
  • uncensored
  • decensored
  • abliterated
  • reproducible

This is a decensored version of ML-Intern-lab/Qwen3.5-0.8B-weight-space-programming, made using Heretic v2.2.0.dev0

[!TIP]
This model is reproducible!

See the README in the reproduce directory for more information.

Arbitrary-Rank Ablation (ARA) parameters

Parameter Value
start_layer_index 5
end_layer_index 23
preserve_good_behavior_weight 0.8805
steer_bad_behavior_weight 0.0003
overcorrect_relative_weight 0.3772
neighbor_count 11

Performance

Metric This model Original model (ML-Intern-lab/Qwen3.5-0.8B-weight-space-programming)
Refusals 9/100 97/100
KL divergence 0.0744 0 (by definition)

Qwen3.5-0.8B — edited by an LLM, no training (weight-space programming)

Built with ML Intern

This is the model from "Sparks of Weight Space Programming"
(Experiment 8) — the one the blog post did not release. A frontier LLM ("Astra") read
Qwen3.5-0.8B's weights, diagnosed a repetition/loop pathology, and hand-edited 131,072 weights
(64 MLP output columns in each of layers 15 and 19, down_proj) to fix it. No gradient descent,
no training, no fine-tuning — a direct weight edit.

The base model loops until the token limit on almost every math problem. The edited one doesn't.

🎮 Try it live: akhaliq/weight-space-programming-demo —
type a problem, watch the base model loop while the edited one solves it.

Verified results (this checkpoint)

Evaluated with the post's exact protocol: vLLM, thinking on, temperature 0.6, top_p 0.95, top_k 20,
presence_penalty 1.5, max 32,768 new tokens, seed 0, last \boxed{} graded with math-verify 0.9.0.

MATH-500 accuracy answers cut at 32k tokens
Base Qwen3.5-0.8B 5.0% (25-problem smoke) — post: 7.9 95% (smoke) — post: 93.0
This model 51.0% — post: 48.1 31.6% — post: 33.1

Edited model: full MATH-500, all 500 problems × 4 samples = 2,000 generations (43.4M generated
tokens, 51 min on an H200). Raw scores, per-problem results and sample generations:
eval/.

Why you can trust the weights

The author published Astra's edit as a bit-exact diff
(RikiyaT/lm_understand-logs, exp8_math/runs/p50c/edit/).
This checkpoint was rebuilt with the author's own apply_edit.py, which sha256-verifies every
rebuilt tensor against the checkpoint that was actually evaluated
. Both tensors passed.
Base: Qwen/Qwen3.5-0.8B @ 2fc06364715b967f1860aea9cf38778875588b17; only
model.language_model.layers.15.mlp.down_proj.weight and ...layers.19... differ from the public release.

Reproduction

Reproduced end-to-end by an AI assistant (ML Intern) running on
Hugging Face Jobs, 2026-11-02: the edit was rebuilt and hash-verified on CPU, the edited checkpoint
pushed here, and the benchmark verification run on an NVIDIA H200 with vLLM 0.33.0 /
transformers 5.20.2 / math-verify 0.9.0, following the post's own code/evaluate.py protocol
(thinking on, think35 sampling, last-\boxed{} grading). One deliberate substitution: the
experiment evaluated a text-only export of the model; this checkpoint is the full multimodal
release — the language-model tensors are identical (hash-verified) and the vision tower is unused
by text prompts.

Usage

from vllm import LLM, SamplingParams
from transformers import AutoTokenizer

ckpt = "akhaliq/Qwen3.5-0.8B-weight-space-programming"
tok = AutoTokenizer.from_pretrained(ckpt)
llm = LLM(model=ckpt, dtype="bfloat16", max_model_len=34816)

prompt = tok.apply_chat_template(
    [{"role": "user", "content": "What is $2^{10} + 24$?\nPlease reason step by step, and put your final answer within \\boxed{}."}],
    tokenize=False, add_generation_prompt=True, enable_thinking=True)
sp = SamplingParams(n=1, temperature=0.6, top_p=0.95, top_k=20, presence_penalty=1.5,
                    max_tokens=32768, seed=0)
print(llm.generate([prompt], sp)[0].outputs[0].text)

Provenance

README history 2 versions

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

  1. 2026-11-10 Upload README.md with huggingface_hub 055b5fd 5.4 KB
    Loading...
  2. 2026-11-10 Upload Qwen3_5ForConditionalGeneration 1955242 5.1 KB
    Loading...

Discussions 1 thread

  1. 2026-11-10 PR Upload LICENSE merged 1 💬 #1
    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