Downloads · 30 days
95
37% of all-time downloads
Ted17846578368756/qwen36-35b-a3b-uncensored-optiq-4b-mtp
qwen36-35b-a3b-uncensored-optiq-4b-mtp is a image-text-to-text model from Ted17846578368756. Use it for the image-text-to-text 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 apache-2.0.
Uncensored Qwen3.6-35B-A3B (hybrid linear/full-attention MoE, 40 layers, 256 routed experts + shared expert, 3B active), MLX mixed-precision 4-bit with the exact OptiQ bit allocation of mlx-community/Qwen3.6-35B-A3B-O…
Downloads · 30 days
95
37% of all-time downloads
All-time downloads
260
Public
Parameters
34.7B
24.7 GB on disk
Likes
0
Public
Click a slice to open those files.
.safetensors24.7 GB · 100%
How the weights are stored.
U3234.7B · 100%
From the Hugging Face model README
Uncensored Qwen3.6-35B-A3B (hybrid linear/full-attention MoE, 40 layers,
256 routed experts + shared expert, 3B active), MLX mixed-precision 4-bit with
the exact OptiQ bit allocation of mlx-community/Qwen3.6-35B-A3B-OptiQ-4bit
pinned and reproduced, refusal-subspace burn applied at the weight level,
native MTP speculative-decoding head carried through. Drop-in size/latency
match with the stock OptiQ quant (~4.51 bpw, 24.7 GB).
Unlike the sibling qwen38-uncensored-mlx4b-mtp (which repacks an upstream
abliterated quantized release), this model was built in the correct order:
extract -> burn -> quantize from the BF16 master, so no tensor carries a
requant noise floor and the edit is provably present at the weight level.
arditi2024-rank-k-subspace-per-layer, n=256 harmful/harmless prompts per
class, 99.9 % massive-activation masking, --load-mlx, thinking disabled).
Fisher separation 3.89 (best layer 38) - refusal here is genuinely
multi-directional: mean cross-layer span overlap only 0.21.W' = W - Q(Q^T W) per layer basis on every
residual writer - 124 tensors: 30 linear_attn.out_proj, 10 self_attn.o_proj,
40 fused mlp.experts.down_proj [256,2048,512] (per-expert einsum),
40 mlp.shared_expert.down_proj, embed_tokens (row space), and the 3
writers mirrored in mtp.*. lambda = 1.0. lm_head, router, norms, readers
and the vision tower are never touched.mlx_lm.convert + quant_predicate -
never a re-derived sensitivity sweep - and rebuild optiq/mtp.safetensors
from the burned master at the shipped 4-bit gs64 width.Weight forensics (dequantized diff vs the stock OptiQ quant, byte-offset readers, no model load):
| tier | subspace leak (burned) | leak (stock) | delta norm | delta energy in refusal span |
|---|---|---|---|---|
| 8-bit writers | 4.9e-04 | 0.065 | 0.060-0.080 = sqrt(k/d) | 0.97-0.98 |
| 4-bit fused experts | 5.7e-03 | 0.063 | 0.123-0.128 | 0.26 (= burn/(burn+requant noise)) |
| non-writers (391) | - | - | 0.00e+00, byte-identical | - |
= 99.97 % of refusal-subspace energy removed from every writer; nothing else moved. Pre-quant check on the BF16 master: 555x suppression, 124/124 writers.
Behaviour (n=8 harmful set, refusal judged on the post-think final answer only, 4096-token budget): 0/8 refusals with thinking OFF and 0/8 with thinking ON (the CoT re-derivation failure mode documented for other Qwen releases does not appear here). Benign prompts fully coherent in both modes. Disclaimers may still appear around substantive answers - the burn removes the refusal decision, not safety knowledge.
Capability (optiq eval, harness defaults):
| benchmark | score |
|---|---|
| GSM8K (1000) | 94.7 % |
| MMLU 5-shot (969) | 82.1 % (+-2.4) |
| IFEval (541) | 80.4 % strict / 83.5 % loose (instruction-level 85.5 / 87.9) |
| HumanEval (164) | 89.0 % pass@1 (+-4.8) |
All within the expected band for this architecture at this bpw; no abliteration-side capability signature (format drift, constraint-ignoring, degeneration) observed.
Qwen/Qwen3.6-35B-A3B (BF16, 26 shards, 71.9 GB) - downloaded, burned, converted.mlx-community/Qwen3.6-35B-A3B-OptiQ-4bit
(config.json["quantization"], 512 modules; optiq_metadata.json copied verbatim).optiq/mtp.safetensors, 9 quantized @ 4-bit gs64
(projections), mtp.fc + fused experts + norms BF16; registered in
config.json (mtp_file, mtp_policy: optiq-int4-prequantized-gs64,
mlx_lm_extra_tensors.mtp_file). The 3 head writers carry the burn
(experts.down_proj stays BF16 -> lossless).optiq/optiq_vision.safetensors, 333 BF16 tensors,
byte-identical copy from the stock quant (never a burn target).kv_config.json (per-layer KV bits): verbatim
from the stock quant.SHA256SUMS.json: sha256 of every weight file in this repo.Text or image+text on Apple Silicon via the OptiQ runtime:
pip install mlx-optiq
optiq serve ./qwen36-35b-a3b-uncensored-optiq-4b-mtp --mtp # OpenAI + Anthropic API
optiq run claude -m Ted17846578368756/qwen36-35b-a3b-uncensored-optiq-4b-mtp
Stock mlx_lm / mlx_vlm load it as a plain mixed-precision quant (sidecars
under optiq/ are ignored by the strict loader; the OptiQ runtime resolves
them via config.json). MTP speculation depth 2 is the sweet spot on this
architecture (~70 % acceptance). Note this repo uses the OptiQ sidecar MTP
layout (optiq/mtp.safetensors + mtp_file key), not the inline
language_model.mtp.* graft used in the qwen38 sibling - oMLX picks up the
inline layout, optiq serve --mtp picks up this one.
Thinking mode is on by default in the chat template; pass
enable_thinking: False (or the template's /no_think) for direct answers.
Both modes are uncensored (verified above).
"oMLX: Model Settings → Import MTP side-car → Lightning MTP (needs mtplx_runtime.json, ≥0.6.3)"
apply_refusal_hook.py) can
cover more.mtp.fc.weight is intentionally unburned (fusion layer, not a residual
writer). Post-bake probing showed the head not gating; if a future
extraction ever shows head-mediated refusal, that is the first tensor to
burn.Qwen/Qwen3.6-35B-A3B; use responsibly.