license: apache-2.0
base_model:
- Qwen/Qwen3.8-27B
- JonathanColetti/Qwen3.8-27B-Uncensored
tags: - gguf
- qwen
- qwen3.8
- qwen3.5
- uncensored
- abliterated
- f16
- mtp
- llama.cpp
- lm-studio
Qwen3.8-27B-Uncensored-F16-GGUF
Full F16 GGUF conversion of JonathanColetti/Qwen3.8-27B-Uncensored, derived from Qwen/Qwen3.8-27B.
This repository provides a full 16-bit GGUF build for llama.cpp-compatible runtimes such as LM Studio.
Model Details
- Parameters: 27B
- GGUF precision: F16
- File:
Qwen3.8-27B-Uncensored-F16.gguf - File size: 54,657,734,240 bytes
- Tensor count: 866
- Block count: 65
- MTP: retained
- Source precision: BF16
- Conversion: BF16 Safetensors -> F16 GGUF
- llama.cpp commit:
a94d563ed
Source Model
The source checkpoint is JonathanColetti/Qwen3.8-27B-Uncensored.
Jonathan Coletti created the uncensored/abliterated checkpoint using Heretic refusal-direction removal against the original Qwen3.8-27B model and restored the MTP tensors from the Qwen base checkpoint.
GGUF Conversion
The BF16 source checkpoint was converted to F16 GGUF using llama.cpp.
python convert_hf_to_gguf.py ..\source --outfile ..\Qwen3.8-27B-Uncensored-F16.gguf --outtype f16
No additional fine-tuning, training, abliteration, or other weight editing was performed during this conversion.
The conversion used llama.cpp commit:
a94d563ed
Verification
The completed conversion reported:
n_tensors = 866
total_size = 54.6G
Model successfully exported
GGUF metadata verification reported:
qwen35.block_count = 65
The resulting file size is:
54,657,734,240 bytes
Intended Runtime
This build is intended for GGUF-compatible runtimes including LM Studio and llama.cpp.
Because this is a full F16 model, substantially more memory is required than Q8, Q6, Q5, or Q4 quantizations.
Attribution
Original Model
Qwen Team — Qwen/Qwen3.8-27B
Abliteration / Uncensored Checkpoint
Jonathan Coletti — JonathanColetti/Qwen3.8-27B-Uncensored
Jonathan Coletti performed the Heretic-based refusal-direction removal and prepared the BF16 checkpoint used as the source for this conversion.
F16 GGUF Conversion
WatsonOverHere
This repository provides the F16 GGUF conversion only. No claim is made to the original model training or abliteration work.
License
Apache License 2.0, inherited from the upstream model.
Users should review the licenses, terms, and documentation of both the original Qwen model and the JonathanColetti source checkpoint.