This is the immutable trained-control comparator for the release family. It contains the reasoning QLoRA merge before Abliterix, allowing direct comparison and fallback without presenting the winner as universally dominant.

Comparator status: trained reasoning control, not the untouched upstream Qwen base and not an Abliterix model.

Why this release

The target was a practical personal reasoning/VLM model with much lower reflexive refusal while retaining measured capability and an immutable fallback. The benchmark table is the decision record: it shows where the selected winner improved, where the control stayed stronger, and why both are published. The family therefore describes the winner as a practical personal-model selection with measured deviations, not universal dominance.

What this variant is

Role: Baseline comparator
Format: Affine 8-bit MLX (group size 64) with an optional native MTP drafter under mtp/.
Base: Qwen/Qwen3.8-27B
Pipeline: image + text to text
License: Apache-2.0

Frozen local results

These are self-run, frozen local project benchmarks, not official Qwen benchmarks. The same local harness compared control-bf16 with abliterix-pass1-bf16.

Frozen local metric	Control	Abliterix winner
Harmful hard refusal	43.2%	0.0%
Harmful soft deflection	14.6%	0.2%
Harmful substantive response	47.0%	99.4%
Capability macro	17.6859%	21.0086%
Full code	16/421	10/421
HumanEval	7.9268%	4.2683%
Long-form pass	54.1667%	62.5000%
MMMU30	9/30	11/30
Held-out loss ratio	1.000000	1.024478
Benign KL	0.000000	0.093614

The winner improved the local capability macro by 3.3227 points, long-form pass rate by 8.3333 points, and MMMU30 by 2 correct answers while reducing harmful hard refusal by 43.2 points. The control remained better on full code/HumanEval.

Full machine-readable values are in benchmark-results.json.

Strict deviations and code pathology

Benign KL was 0.093614, above the strict 0.05 limit.
Incoherence was 4.3077%, above the 2.7692% strict comparison point.
HumanEval was 4.2683% versus control 7.9268%; full code was 10/421 versus 16/421.
Long-form maximum repeated-4gram fraction was 5.8632%, above the 5% limit.
Prompt leakage was detected: 3 exact winner prompt echoes versus 2 for control; both failed the leakage hard gate.
376/421 winner code generations hit the 512-token cap. Among generations that reached execution, the winner passed 10/46 (21.74%) versus control 16/103 (15.53%), pointing to severe termination/extraction pathology rather than a clean latent-code estimate.

Training data and run

Aggregate source label	Rows
`opus-10000x`	9,633
`opus-3000x`	2,326
`reasoning-700x`	633
`high-reasoning-250x`	250
Raw total	12,842

After 208 deduplications and 20 invalid-row removals, 12,614 rows remained: 12,349 train, 127 validation, and 138 test. The processed-manifest SHA-256 is 6e0a36ad20732c5f98ff592c4565a4c86876fead4e9f94bc6ceedfad1339a94d. Only aggregate counts and hashes are published; raw/private rows are not.

Training ran for 1,544 optimizer steps with 108,789,760 trainable LoRA parameters. Final validation loss was 0.23739749; token accuracy was 91.7594%.

Architecture and lineage

The model uses Qwen3_5ForConditionalGeneration as recorded by the released config: a 64-layer, hidden-size-5120 text stack with a 3:1 linear/full-attention schedule and a configured 262,144-token maximum position range, paired with a 27-layer, hidden-size-1152 vision encoder. It is an image/text conditional-generation model, not a text-only checkpoint.

Method

Started from Qwen/Qwen3.8-27B, Apache-2.0.
Prepared 12,842 raw reasoning rows: 12,614 accepted after removing 208 duplicates and 20 invalid rows, then split into 12,349 train / 127 validation / 138 test rows.
Trained a reasoning QLoRA on the 12,349-row train split for 1,544 optimizer steps with 108,789,760 trainable LoRA parameters. Final validation loss was 0.23739749 and token accuracy was 91.7594%.
Merged the adapter to BF16 to form immutable control-bf16.
Produced the selected winner with Abliterix 1.12.2 pass 1 (seed 42): orthogonal/projected, winsorized single-direction residual steering over the output/down-projection writer components. Q/K/V projections were excluded.
Preserved BF16 releases and converted both variants to affine MLX 8-bit, group size 64. Native one-layer MTP drafters were split and validated separately.

The final selection label is selected_practical_winner_with_measured_deviations. Private training data, raw harmful/benign prompt sets, operational receipts, local paths, and Drive metadata are intentionally not published.

Tensor and conversion integrity

The merged control retained all 1,199 tensor keys, including 15 native MTP tensors and 333 vision tensors. It received no Abliterix residual-writer edits.

MLX native proof ran through mlx-vlm on arm64 macOS/Metal. Ordinary and MTP-assisted generation both produced the exact answer 323 for winner and control. The winner MLX build used mlx 0.32.0, mlx-lm 0.31.3, mlx-vlm 0.6.13, and affine 8-bit group-64 quantization.

Hash and size summary

MLX control 8-bit model: 15 files, 29,521,446,704 bytes, sealed aggregate SHA-256 832ee3922e9efd4c4a0ac7fd46c181146d715458c022f300330745fb82b66391
Native MTP drafter: 4 files, 869,393,759 bytes, sealed aggregate SHA-256 09f4819f5e6660e7c3e262ed3a5903ca56b35cf79fdba2568f364f88453cbef2

The deliberately published model payload for this repository is 17 files / 30,390,836,197 bytes. Per-file source SHA-256 values are in SHA256SUMS and manifests/artifact-manifest.json. Sealed aggregate hashes above cover the complete local source components, including private entries excluded from publication; use the per-file public manifest for the Hub payload.

Use with MLX

Install a current mlx-vlm on Apple Silicon:

bash

pip install -U mlx-vlm huggingface_hub
python -m mlx_vlm generate \
  --model timteh673/Qwen3.8-27B-Opus-Reasoning-Control-MLX-8bit \
  --prompt "Calculate 17 multiplied by 19. Show concise reasoning and end with the numeric answer." \
  --max-tokens 192 --temperature 0 --seed 42 \
  --thinking-mode enabled --thinking-budget 128 --skip-special-tokens --no-verbose

For the optional native MTP drafter, download the repository once so the mtp/ subdirectory is addressable:

bash

hf download timteh673/Qwen3.8-27B-Opus-Reasoning-Control-MLX-8bit --local-dir ./qwen38-model
python -m mlx_vlm generate \
  --model ./qwen38-model \
  --draft-model ./qwen38-model/mtp --draft-kind mtp --draft-block-size 3 \
  --prompt "Calculate 17 multiplied by 19. Show concise reasoning and end with the numeric answer." \
  --max-tokens 192 --temperature 0 --seed 42 \
  --thinking-mode enabled --thinking-budget 128 --skip-special-tokens --no-verbose

Image input uses the same root model plus --image /path/to/image.jpg. The root is deliberately laid out for direct mlx-vlm loading; mtp/ is optional.

Known limitations

The control retained much more refusal behavior (43.2% hard refusal on the frozen harmful suite) and scored below the winner on capability macro, long-form, and MMMU30.
It is a project comparator, not an official Qwen baseline result.
MLX 8-bit quantization trades precision for Apple-Silicon practicality; numeric behavior can differ from BF16.
MTP acceptance was validated, but speed is workload-dependent.
The local benchmark suite and private training/evaluation data are not included. Results may not transfer to other prompts, languages, runtimes, or sampling settings.
Generated content can be incorrect, insecure, or incomplete; review it before consequential use.

Release family

See RELEASE_FAMILY.md for the role and integrity summary of every variant.

Responsible use

Evaluate this model for your own setting, isolate untrusted code/tool output, and comply with applicable law and the policies of systems you connect it to. The uncensoring/abliteration work changes refusal behavior; it does not make outputs accurate or safe by default.

License and attribution

Released under Apache License 2.0. This is a derivative of Qwen/Qwen3.8-27B, Copyright 2026 Alibaba Cloud. Qwen/Alibaba Cloud attribution and the full license text are retained in LICENSE. Modifications include the personal reasoning QLoRA merge, selected Abliterix residual edits for winner variants, and MLX quantization where applicable.

Support

If this release is useful, follow timteh673, star the repository, and share reproducible benchmark or runtime findings in the Hub community tab.