base_model:

• xtools-at/DeepSeek-V4-Pro-Qwen3.5-9B-UNCENSORED-heretic
  license: apache-2.0
  language:
• en
  pipeline_tag: image-text-to-text
  tags:
• mtp
• text-generation-inference
• llama-cpp
• transformers
• unsloth
• qwen3_5
• reasoning
• distillation
• deepseek
• sft
• rl
• gspo
• math
• stem
• tool-use
• function-calling
• heretic
• uncensored
• decensored
• abliterated
• speculative-decoding
• nvidia-blackwell
• rtx-pro-6000
• SM120
• GB202
• B200
• B300
• FP4
• blackwell
• Qwen3.5
• GPTQ
  library_name: transformers

DeepSeek-V4-Pro-Qwen3.5-9B-UNCENSORED-heretic-MTP-GGUF

DeepSeek-V4-Pro-Qwen3.5-9B-UNCENSORED-heretic is xtools-at's decensored version of Jackrong/DeepSeek-V4-Pro-Qwen3.5-9B, made with Heretic v2.0.0.dev0 using Arbitrary-Rank Ablation (layers 9–24). The card reports refusals falling from 100/100 to 3/100 at a KL divergence of 0.0422 from the original, and a 4B version also exists. The original is a 9B reasoning fine-tune of Qwen3.5-9B, distilled from DeepSeek-V4-Pro in Max Effect mode on about 250K math and STEM samples. Training was SFT followed by a light GSPO RL stage, with no coding data. The original's card reports 94.50% on GSM8K (four-run mean, NVFP4 on vLLM) and a 90.53% average on a 1,500-question MMLU-Pro Math/Physics/Chemistry sample, 0.93 points above Qwen3.5-9B, using 36% fewer reasoning tokens per correct answer. It also claims near-complete output-format compliance, with a single-letter answer 97.2% of the time against 37.3% for Qwen3.5-9B. All benchmarks were run on the original model, not on this decensored version, and the card calls the coding and tool-use gains small and preliminary. The base weights are marked Apache 2.0, and the card is a community model that says outputs need independent verification.

[!NOTE]
MTP — Speculative Decoding: Yes

[!NOTE]
Multi-Token Prediction (MTP) GGUF is a specialized GGUF model file format extension that integrates speculative decoding directly into the model weights to significantly accelerate local inference. Unlike traditional speculative decoding which requires a separate, smaller "draft" model, MTP GGUF files include additional output heads within the main model architecture that predict multiple future tokens in a single forward pass.

[!NOTE]
NVFP4-GPTQ Version: DeepSeek-V4-Pro-Qwen3.5-9B-UNCENSORED-heretic-NVFP4-GPTQ | NVFP4-GPTQ-GGUF Version: DeepSeek-V4-Pro-Qwen3.5-9B-UNCENSORED-heretic-NVFP4-GPTQ-GGUF

Model Files

llama.cpp

LLM inference in C/C++ — https://github.com/ggml-org/llama.cpp

Quantized using llama.cpp by prithivMLmods.