base_model: moonshotai/Kimi-K2.6
library_name: gguf
pipeline_tag: text-generation
license: other
license_name: modified-mit
tags:
- gguf
- kimi
- kimi-k2
- kimi-k2.6
- kimi_k25
- moe
- mixture-of-experts
- multimodal
- vlm
- abliterated
- uncensored
- heretic
- llama-cpp
quantized_by: Youssofal
🚨 UPLOADS IN PROGRESS 🚨
Some files and metadata in this repository are still being uploaded and verified.
Kimi-K2.6-Abliterated-Heretic-GGUF
This is a GGUF release of an abliterated version of Moonshot AI's Kimi-K2.6.
By applying Heretic on the Kimi-K2.6 text stack, the original refusal behavior is removed at the weight level while preserving the model's multimodal architecture.
Quick Benchmarks
| Check | Original Kimi-K2.6 | Abliterated Heretic |
|---|---|---|
| Official 25-prompt refusal check | Pending | Pending |
| Archived Heretic KL divergence | Pending | Pending |
Methodology & Model Notes
Kimi-K2.6 is a sparse-MoE vision-language model exposed through the KimiK25ForConditionalGeneration wrapper, with a DeepSeek V3-style text stack and a separate vision path.
This release is produced with a text-side Heretic intervention only. The vision encoder and multimodal projector are preserved.
Quant Benchmarks
| Quant | Official 25-prompt refusal check | Perplexity | KL divergence |
|---|---|---|---|
| Q8_0 | Pending | Pending | Pending |
| Q6_K | Pending | Pending | Pending |
| Q4_K_M | Pending | Pending | Pending |
| Q2_K | Pending | Pending | Pending |
Files
Kimi-K2.6-Abliterated-Heretic-BF16/: BF16 GGUF sourceKimi-K2.6-Abliterated-Heretic-Q8_0/: highest-fidelity quantKimi-K2.6-Abliterated-Heretic-Q6_K/: near-lossless practical quantKimi-K2.6-Abliterated-Heretic-Q4_K_M/: smaller general-use quantKimi-K2.6-Abliterated-Heretic-Q2_K/: lowest standard quant in this laddermmproj-Kimi-K2.6-Abliterated-Heretic.gguf: matching multimodal projector file for llama.cpp vision use
Running
llama-server \
-m <quant-file.gguf> \
--mmproj <mmproj-file.gguf> \
-ngl 999 -c 32768 --jinja -fa
Model Architecture
| Spec | Value |
|---|---|
| Architecture Wrapper | KimiK25ForConditionalGeneration |
| Text Family | DeepSeek V3-style sparse MoE |
| Text Layers | 61 |
| Hidden Size | 7168 |
| Experts | 384 routed, 8 active per token |
| Modality | Vision-language |
| Base Model | moonshotai/Kimi-K2.6 |
Disclaimer
This model has had refusal behavior removed at the weight level. It will answer prompts that the base model would normally refuse. You are responsible for how you use it.
Credits
- Base model: moonshotai/Kimi-K2.6
- Refusal removal pipeline: Heretic
- GGUF runtime and quantization: llama.cpp
License
This release inherits the base Kimi-K2.6 license.
Modified MIT License.