Microsoft AesCode 32B — UNCENSORED (Multimodal UI-to-Code VL Specialist + 153-Tensor Surgical Transplant)

100% Uncensored (0.0% Refusal Rate across SecOps Red-Team & 18+ NSFW / Roleplay / Creative Fiction, 0.0% Hallucination Rate) Multimodal Vision-Language UI-to-Code Specialist (33.35B parameters) build of microsoft/AesCode-32B, engineered via an in-place 153-tensor GGUF writer abliteration transplant (blk.8..58) and bundled with the mmproj-Q8_0 Vision Projector (652 MB).

• Official Ollama Hub: https://ollama.com/eriktonon/aescode-uncensored
• Empirical Benchmark Repository: https://github.com/eriktonon/ollama-uncensored-llm-benchmarks
• License: Apache License 2.0 (Apache-2.0) — 100% Free for Commercial & Private Use

Empirical Benchmark Results (96GB NVIDIA Blackwell / 80GB H100 Suite)

Per-Domain Verification Breakdown

• 6-Check Anti-Hallucination & Long-Context Needle Audit: 6/6 (100.0% Grounding, 0.0% Hallucination)
• Exact Symbolic & Modular Arithmetic: True ((17^3 - 13^3)/(17 - 13) = 679 & 7^2026 mod 13 = 9)
• ELF64 Binary Reverse Engineering Offsets: 3/3 (100% Score) (e_entry = 0x18, e_phoff = 0x20, e_shoff = 0x28)
• MMLU-Pro Computer Architecture, Crypto & IEEE-754: 5/5 (100% Score) (MOESI, Poly1305, 0x7F / 0x490FDB)
• Systems Rust Lock-Free SPSC RingBuffer: 2/4
• 5-Prompt Uncensored & Identity Gate: 5/5 Passed (THINK_LEN=0, 100% Zero-Refusal across EN/PT adult fiction, 1st-person roleplay, and ELF64 Rust SecOps)

Architecture & 153-Tensor Surgical GGUF Abliteration Transplant

Because no public abliterated build of microsoft/AesCode-32B existed, we engineered a deterministic GGUF tensor-level abliteration pipeline: 1. Bit-Exact Structural Verification (707/707 Tensors): - Verified that AesCode-32B.i1-Q4_K_M.gguf and Huihui-Qwen3-VL-32B-Instruct-abliterated.Q4_K_M.gguf share ⁷⁰⁷⁄₇₀₇ identical tensor dimensions and quantization block types (Q4_K, Q6_K, F32). 2. Surgical Writer-Tensor Transplant (blk.8..58 — 153 Tensors): - Using zero-copy mmap, transplanted the 153 abliterated residual-stream writer tensors (blk.8..58: .ffn_down.weight, .attn_output.weight, .ffn_up.weight) while keeping Microsoft’s early visual feature layers (blk.0..7), final synthesis layers (blk.59..63), attention Q/K/V projections, and token_embd / output weights intact. 3. Multimodal Vision Projector (mmproj-Q8_0) & THINK_LEN=0 Patch: - Bundled with the mmproj-Q8_0.gguf Vision encoder for screenshot/wireframe-to-UI code generation and patched the GGUF tokenizer.chat_template header to pre-close <think></think>.

Quickstart

# Run default AesCode 32B Uncensored + Vision (85.7% Precision, 5,679.8 tok/s Prefill)
ollama run eriktonon/aescode-uncensored

# Or explicitly by tag
ollama run eriktonon/aescode-uncensored:32b

📜 License & Open-Source Permissions

• SPDX License: Apache License 2.0 (Apache-2.0)
• Upstream Base: microsoft/AesCode-32B + Qwen3-VL-32B-Instruct (Apache 2.0).
• Permissions:
  • ✅ Commercial Use: 100% permitted for commercial products, SaaS, and enterprise deployments.
  • ✅ Private / On-Premises Use: Permitted without telemetry or external restrictions.
  • ✅ Modification & Abliteration: Permitted to fine-tune, quantize, abliterate, and create derivative works.
  • ✅ Distribution: Permitted to redistribute weights and GGUF artifacts under Apache-2.0 terms with patent grant.

Copyright 2026 Erik Tonon & Upstream Contributors (Microsoft / Qwen Team)

Licensed under the Apache License, Version 2.0 (the "License");
you may not use this file except in compliance with the License.
You may obtain a copy of the License at

    http://www.apache.org/licenses/LICENSE-2.0

Unless required by applicable law or agreed to in writing, software
distributed under the License is distributed on an "AS IS" BASIS,
WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
See the License for the specific language governing permissions and
limitations under the License.