base_model:
- xtools-at/DeepSeek-V4-Pro-Qwen3.5-9B-UNCENSORED-heretic
- Jackrong/DeepSeek-V4-Pro-Qwen3.5-9B
license: apache-2.0
library_name: gguf
tags: - gguf
- llama.cpp
- quantized
- uncensored
- text-generation
language: - en
- zh
- ko
- ja
- es
- ru
DeepSeek-V4-Pro-Qwen3.5-9B-UNCENSORED-heretic (GGUF)
Quantized GGUF versions of xtools-at/DeepSeek-V4-Pro-Qwen3.5-9B-UNCENSORED-heretic, an uncensored "heretic" variant of Jackrong/DeepSeek-V4-Pro-Qwen3.5-9B, ready to run with llama.cpp and other GGUF-compatible inference engines.
Available files
| File | Quant | Size |
|---|---|---|
DeepSeek-V4-Pro-Qwen3.5-9B-UNCENSORED-heretic-Q4_K_M.gguf |
Q4_K_M | 5.4 GB |
DeepSeek-V4-Pro-Qwen3.5-9B-UNCENSORED-heretic-Q8_0.gguf |
Q8_0 | 9.1 GB |
DeepSeek-V4-Pro-Qwen3.5-9B-UNCENSORED-heretic-BF16.gguf |
BF16 | 17.1 GB |
mmproj-DeepSeek-V4-Pro-Qwen3.5-9B-UNCENSORED-heretic-BF16.gguf |
BF16 (vision projector) | 0.9 GB |
Recommendation: Q4_K_M offers the best speed/quality trade-off for most GPUs. Q8_0 is near-lossless. BF16 is the full-precision reference.
Note: the
mmprojfile is the multimodal projector for vision input. Pass it with--mmprojin llama.cpp when using image input; it is not needed for text-only inference.
Usage (llama.cpp)
# Download
hf download yan2026/DeepSeek-V4-Pro-Qwen3.5-9B-UNCENSORED-heretic-GGUF --include "*Q4_K_M.gguf"
# Run (text)
llama-server -m DeepSeek-V4-Pro-Qwen3.5-9B-UNCENSORED-heretic-Q4_K_M.gguf --mmproj mmproj-DeepSeek-V4-Pro-Qwen3.5-9B-UNCENSORED-heretic-BF16.gguf -c 32768
Model details
- Quantized from: xtools-at/DeepSeek-V4-Pro-Qwen3.5-9B-UNCENSORED-heretic
- Base model: Jackrong/DeepSeek-V4-Pro-Qwen3.5-9B
- Architecture: Qwen3.5 hybrid (recurrent linear-attention layers + full attention every 4 layers)
- Context length: 262,144 tokens
- License: Apache-2.0
Disclaimer
This is an "UNCENSORED / heretic" variant with reduced refusal behavior. You are responsible for how you use it and for complying with applicable laws and the base model's license.