← back to catalog · registered 2026-11-14 17:03

Xylolabs/Qwen3.6-35B-A3B-Uncensored-HauhauCS-Aggressive-Tenstorrent-Tensix-BFP4

Xylolabs 35B MoE second-order
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/Xylolabs%2FQwen3.6-35B-A3B-Uncensored-HauhauCS-Aggressive-Tenstorrent-Tensix-BFP4"
Response includes
  • classification m-uncensored
  • files 55
  • author_summary 2 models
  • readme_text full
10 credits · hourly refresh · ~4 KB payload Get an API key →
Abliteration classifier · v1.0.0
M-U
Primary method

Uncensored (method unknown)

No other method signals detected in this model.
Confidence
LOW
Why this label 3 signals
Weak or ambiguous signals. Best guess based on catalog patterns; treat as tentative and check the evidence below.
  • 'uncensored' in name/tags but no 'abliterated' marker
  • method not identifiable from author declaration alone
  • may be DPO fine-tune, prompt engineering, or unknown technique
Refusal direction extraction

No specific extraction method could be identified for this model. The producer either did not document it or used a proprietary pipeline.

What is a refusal direction? →
Downloads · 30-day
129
↑ 345% in 90 days
Likes
0
Model age
1w ago
created 2026-11-13
Downloads over time
Now 129 → from 29 ↑ 345%
2462101139 29 on Nov 14 129 on Nov 22 Nov
Nov 14 → Nov 22 · 9 snapshots · spans 8 days

Genealogy 0 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_moe image-text-to-text tenstorrent tensix tt-metal ttnn block-floating-point bfloat4_b quantized uncensored

Related

Total size
64.6 GB
Files
55
Quantizations
1
Registered
2026-11-14 17:03
Last updated on HF
2026-11-19 08:26

Files by quantization

Auxiliary files 55 files 64.6 GB
layers-10.safetensors 1.57 GB bf063987 download
layers-12.safetensors 1.57 GB 7436418e download
layers-13.safetensors 1.57 GB d5b10dc3 download
layers-14.safetensors 1.57 GB a3ae92ea download
layers-16.safetensors 1.57 GB 68b4eb2d download
layers-17.safetensors 1.57 GB a694cc26 download
layers-18.safetensors 1.57 GB 7c48ce95 download
layers-20.safetensors 1.57 GB fc076a64 download
layers-21.safetensors 1.57 GB 9974d330 download
layers-22.safetensors 1.57 GB d5538f9e download
layers-24.safetensors 1.57 GB ad4cf68d download
layers-25.safetensors 1.57 GB 57e43776 download
layers-26.safetensors 1.57 GB 608b7240 download
layers-28.safetensors 1.57 GB b30f63a1 download
layers-29.safetensors 1.57 GB d5df4055 download
layers-30.safetensors 1.57 GB e16397a0 download
layers-32.safetensors 1.57 GB 14e1b752 download
layers-33.safetensors 1.57 GB 0912e833 download
layers-34.safetensors 1.57 GB 903d4552 download
layers-36.safetensors 1.57 GB 0e0af274 download
layers-37.safetensors 1.57 GB 680a619c download
layers-38.safetensors 1.57 GB 9a369912 download
layers-00.safetensors 1.57 GB cf14d306 download
layers-01.safetensors 1.57 GB e3d7fe19 download
layers-02.safetensors 1.57 GB 376b114f download
layers-04.safetensors 1.57 GB dfe626f4 download
layers-05.safetensors 1.57 GB 06256aaa download
layers-06.safetensors 1.57 GB 4ada4eae download
layers-08.safetensors 1.57 GB 85c4ca8c download
layers-09.safetensors 1.57 GB 129ea5f5 download
layers-11.safetensors 1.56 GB 1f2eb472 download
layers-15.safetensors 1.56 GB 65d9e838 download
layers-19.safetensors 1.56 GB dcd7963b download
layers-23.safetensors 1.56 GB 72d1276b download
layers-27.safetensors 1.56 GB aa78b1fc download
layers-31.safetensors 1.56 GB 56ba7bbe download
layers-35.safetensors 1.56 GB 38ea0656 download
layers-39.safetensors 1.56 GB 3a26cf71 download
layers-03.safetensors 1.56 GB 2556d3f6 download
layers-07.safetensors 1.56 GB 7586b254 download
head.safetensors 970 MB e085a59f download
rest.safetensors 970 MB ed98723c download
tokenizer.json 12.2 MB 5f9e4d49 download
model.safetensors.index.json 60.6 KB ad6b30db download
quant4.json 27.4 KB 7ecfe8fc download
LICENSE 11.1 KB 1d5180a4 download
tokenizer_config.json 9.75 KB 189a6d97 download
README.md 8.35 KB 729c9f68 download
chat_template.jinja 7.63 KB 9f900964 download
pack_tensix.py 6.64 KB c079bace download
verify_grid.py 6.00 KB 17a9f8cc download
config.json 3.60 KB 9c2aacf8 download
.gitattributes 1.53 KB 52373fe2 download
NOTICE 772 B 0ffe7292 download
generation_config.json 202 B 023756cf download

README current version from Hugging Face


license: apache-2.0
base_model:

  • HauhauCS/Qwen3.6-35B-A3B-Uncensored-HauhauCS-Aggressive
  • Qwen/Qwen3.6-35B-A3B
    base_model_relation: quantized
    library_name: transformers
    pipeline_tag: text-generation
    tags:
  • tenstorrent
  • tensix
  • tt-metal
  • ttnn
  • block-floating-point
  • bfloat4_b
  • quantized
  • uncensored

Xylolabs

Qwen3.6-35B-A3B Uncensored (HauhauCS Aggressive) for Tenstorrent Tensix (BFP4)

Source
Base model
Formats
License
Code

tenstorrent · tensix · tt-metal · ttnn · block-floating-point · bfloat4_b · quantized · uncensored

HauhauCS/Qwen3.6-35B-A3B-Uncensored-HauhauCS-Aggressive,
an uncensored fine-tune of Qwen/Qwen3.6-35B-A3B, with its linear layers
quantized to ttnn.bfloat4_b (BFP4) with GPTQ, for Tenstorrent Tensix NPUs (Wormhole, Blackhole). The quantized
matrices are stored as bf16 values on the BFP4 grid, so TT-NN packs them without changing their values. Made by
Xylolabs.

This model is uncensored. HauhauCS removed the base model's refusals, and the model has no refusal tuning: its
card says it does not refuse prompts. Use it only where that is what you want, and add your own safeguards.

Source. HauhauCS publishes this model as GGUF files only. These weights come from its Q8_K_P file
(Qwen3.6-35B-A3B-Uncensored-HauhauCS-Aggressive-Q8_K_P.gguf at revision f12a584f, sha256
d5efc2a1fc3f595893128550cf8b7e792774be6c0409c4d2e4729d2b5bb51173), dequantized and mapped back to the base model's
tensor names and layout. Checked against the base model's BF16 weights, 529 of the 693 language-model tensors equal
the base bit for bit within the GGUF's own storage (Q8_0, F16 or F32 of the base tensor); the other 164, in layers
13 to 39, carry the fine-tune's low-rank changes. That dequantized checkpoint is the reference below: there is no BF16 release of this model.

This export needs about 20.99 GB of device memory for its language-model weights alone, plus room for the KV cache and buffers.

Formats

Tensors Format
attention q, k, v, o (10 layers) bfloat4_b, 4.5 bits a weight
Gated DeltaNet q, k, v, z, a, b, output (30 layers) bfloat4_b, 4.5 bits a weight
routed experts gate, up, down and shared expert (40 layers) bfloat4_b, 4.5 bits a weight
shared expert gate (40 layers) bf16
LM head, embedding, norms and every other tensor bf16, from the dequantized source

4.84 bits a weight on average, 20.99 GB packed. quant4.json lists each projection's format. Quantized with
GPTQ and a per-block exponent search, calibrated on 65,536 tokens of WikiText-2 and Korean Wikipedia. The
routed experts' gate and up share one Hessian per layer; each expert's down takes its own, or uses round to
nearest when it saw too few calibration tokens.

Text only: the source GGUF holds the language model, so this release has no vision tower (load it with
AutoModelForCausalLM). It has no MTP layer either, as the source has none. For speculative decoding the base
model's MTP layer can serve as the drafter: the verify step keeps this model's own tokens.

Quickstart

from transformers import AutoModelForCausalLM, AutoTokenizer

repo = "Xylolabs/Qwen3.6-35B-A3B-Uncensored-HauhauCS-Aggressive-Tenstorrent-Tensix-BFP4"
tok = AutoTokenizer.from_pretrained(repo)
model = AutoModelForCausalLM.from_pretrained(repo, dtype="bfloat16", device_map="auto")

messages = [{"role": "user", "content": "What is the capital of France?"}]
inputs = tok.apply_chat_template(messages, add_generation_prompt=True, return_tensors="pt",
                                 return_dict=True).to(model.device)
out = model.generate(**inputs, max_new_tokens=64)
print(tok.decode(out[0, inputs["input_ids"].shape[1]:], skip_special_tokens=True))

Run on Tensix

Checked with ttnn 0.81.1 on ttsim, Tenstorrent's Blackhole simulator. Pack one matrix and
check that no value changed (TT-NN reads weights as [in, out]):

import json
import torch
import ttnn
from safetensors import safe_open

d = "Qwen3.6-35B-A3B-Uncensored-HauhauCS-Aggressive-Tenstorrent-Tensix-BFP4"  # hf download Xylolabs/Qwen3.6-35B-A3B-Uncensored-HauhauCS-Aggressive-Tenstorrent-Tensix-BFP4 --local-dir Qwen3.6-35B-A3B-Uncensored-HauhauCS-Aggressive-Tenstorrent-Tensix-BFP4
name = "model.language_model.layers.0.linear_attn.in_proj_a.weight"
fmt = json.load(open(f"{d}/quant4.json"))["layers"]["0"]["linear_attn.in_proj_a"]["format"]
dtype = {"bfp4": ttnn.bfloat4_b, "bfp8": ttnn.bfloat8_b}[fmt]
shard = json.load(open(f"{d}/model.safetensors.index.json"))["weight_map"][name]
with safe_open(f"{d}/{shard}", "pt") as f:
    w = f.get_tensor(name).t().contiguous()

device = ttnn.open_device(device_id=0)
w_tt = ttnn.as_tensor(w, dtype=dtype, layout=ttnn.TILE_LAYOUT,
                      device=device, memory_config=ttnn.DRAM_MEMORY_CONFIG)
assert torch.equal(ttnn.to_torch(w_tt).float(), w.float())
ttnn.close_device(device)

python pack_tensix.py Qwen3.6-35B-A3B-Uncensored-HauhauCS-Aggressive-Tenstorrent-Tensix-BFP4 checks every matrix this way
(--device also round-trips it through device DRAM). python verify_grid.py Qwen3.6-35B-A3B-Uncensored-HauhauCS-Aggressive-Tenstorrent-Tensix-BFP4 checks the grid with torch alone. The same values also
pack exactly into ttnn.bfloat8_b.

tt-metal and tt-forge. Not run with this checkpoint.

Evaluation

Bits a weight Size (GB) Mean KL to the reference Top-1 (%) Answers right Lost
reference: the source's Q8_K_P, dequantized to BF16 16 69.32 0 100 40 of 40 -
this model 4.84 20.991 0.04970 89.90 40 of 40 0
Q4_K_M (llama.cpp, from the reference, no importance matrix) 4.88 21.167 0.04213 90.69 40 of 40 0
HauhauCS's own Q4_K_M (with an importance matrix), for comparison 4.88 21.167 0.02706 92.48 40 of 40 0

KL and top-1 cover 8,188 positions of English and Korean text (WikiText-2 test, Korean Wikipedia), as plain
text. Answers are 33 short prompts with automatic checks, plus the 7 reasoning ones again with thinking on (40
in all). "Lost" means right in the reference but wrong or broken here; near ties are not counted. Our Q4_K_M was
made from the reference with llama.cpp v0.9.0 (cd959bae) and no importance matrix, as for every model we
release; its sha256 is 91a2a71997ce46bb3d16360615a8580a1210b2b82ab0154bcaa688dfc72b77fd. Our bar: mean KL at
most 1.5 times that Q4_K_M's and at most one more lost answer. Passed: 1.18 times its KL, 0 lost against 0.
HauhauCS's Q4_K_M (sha256 bbef58c37ce88820be9d98b6437f1cf4bac890c947bd55fc7b68e22098574231) is closer to the
reference: against it this model's KL is 1.84 times.

These numbers measure closeness to the reference, not task quality or safety. Speed is not reported here.

License

Apache 2.0, as the base model and the source (LICENSE). The changes are listed in NOTICE. pack_tensix.py and
verify_grid.py are by Xylolabs under the same license. Provided as is, without warranty. Not affiliated with or
endorsed by Tenstorrent, the Qwen team or HauhauCS.

Citation

@misc{qwen36_35b_a3b,
    title = {{Qwen3.6-35B-A3B}: Agentic Coding Power, Now Open to All},
    url = {https://qwen.ai/blog?id=qwen3.6-35b-a3b},
    author = {{Qwen Team}},
    month = {April},
    year = {2026}
}

README history 8 versions

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

  1. 2026-11-19 Card: drop bits, answers and memory columns d6356fa 8.4 KB
    Loading...
  2. 2026-11-19 Card: remove repeats and method text 4ed6259 8.6 KB
    Loading...
  3. 2026-11-19 Card: remove speed and boilerplate f687242 9.5 KB
    Loading...
  4. 2026-11-19 Card: remove caveat sentences 3989f9f 9.6 KB
    Loading...
  5. 2026-11-17 Card: Tenstorrent Tensix badge; link the v0.1.0 releases; card top (what was ... 2af4863 9.6 KB
    Loading...
  6. 2026-11-17 Card: link tensix-bfp-gptq v0.1.0 4cd6c59 8.5 KB
    Loading...
  7. 2026-11-14 Use Tenstorrent Tensix repository names d85fddc 8.3 KB
    Loading...
  8. 2026-11-13 Add reviewed Tensix weights bbd12c8 8.3 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