Qwen3.6 27B Claude Opus Reasoning Distill v2 Heretic GGUF | Sweet Tea StudioQwen3.6 27B Claude Opus Reasoning Distill v2 Heretic GGUF
GGUF quantizations of darkc0de/Qwen3.6-27B-Claude-Opus-Reasoning-Distill-v2-heretic Model Overview
Verified source
KindOtherBase modeldarkc0de/Qwen3.6-27B-Claude-Opus-Reasoning-Distill-v2-hereticVersionv4c15c28a53c833ef2fd43a248c940b7178b0dae1Licenseapache-2.0Publisher@fzcfweasdferttggCgrade Model source
- Kind
- Other
- Base model
- darkc0de/Qwen3.6-27B-Claude-Opus-Reasoning-Distill-v2-heretic
- Version
- v4c15c28a53c833ef2fd43a248c940b7178b0dae1
- License
- apache-2.0
- Parameters
- 27B
- Tasks
- text-generation
- Source
- Hugging Face
Qwen3.6-27B-Claude-Opus-Reasoning-Distill-v2-heretic — GGUF GGUF quantizations of ## Model Overview | Property | Value | |---|---| | | Qwen/Qwen3.6-27B | | | Qwen3.5ForConditionalGeneration | | | 27B (dense) | | | 262,144 tokens (256K) | | | Apache-2.0 | ## Training Pipeline ### Step 1: Claude Opus Reasoning Distillation Fine-tuned on Claude Opus reasoning datasets using (2x training speedup): - - ### Step 2: Abliteration (Uncensoring) Applied using . Reproducible via directory in the original repo. | Parameter | Value | |---|---| | | 41.77 | | | 1.22 | | | 51.32 | | | 1.21 | | | 32.79 | | | 1.44 | | | 42.02 | | | 0.13 | | | 37.15 | ## Quality Metrics ### KL Divergence & Refusal Rate | Metric | Value | |---|---| | (vs base) | 0.0774 | | | 8/100 (down from 98/100) | ### Benchmarks (mxfp8 precision) | Benchmark | Base Qwen3.6-27B | This Model | Delta | |---|---|---|---| | | 0.647 | 0.665 | +1.8% | | | 0.803 | 0.831 | +2.8% | | | 0.910 | 0.910 | 0.0% | | | 0.773 | 0.790 | +1.7% | | | 0.450 | 0.456 | +0.6% | | | 0.806 | 0.813 | +0.7% | | | 0.742 | 0.772 | +3.0% | ## Available Quantizations | File | Size | BPW | |---|---|---| | | ~50 GB | 16.0 | | | ~4 GB | 1.1 | | | ~10 GB | 2.5 | | | ~11 GB | 3.0 | | | ~12 GB | 3.5 | | | ~13 GB | 3.8 | | | ~14 GB | 4.0 | | | ~16 GB | 4.8 | | | ~15 GB | 4.0 | | | ~15 GB | 4.5 | | | ~17 GB | 5.0 | | | ~19 GB | 5.7 | | | ~17 GB | 5.0 | | | ~18 GB | 5.5 | | | ~21 GB | 6.0 | | | ~27 GB | 8.0 | ### Quantization Details - : Full float16 precision, lossless conversion from safetensors - : Extreme compression, very low quality - : Smallest practical size, lower quality - : Q3 variants with increasing quality - : Legacy formats, compatible with older tools - : K-quant mixtures with balanced quality/size - : Legacy formats, compatible with older tools - : K-quant mixtures with high quality - : Highest quality K-quant - : Near-lossless, 8-bit integer quantization ## System Requirements | Quant | VRAM | |---|---| | f16 | ~50 GB | | Q1_0 | ~4 GB | | Q2_K | ~10 GB | | Q3_K_S | ~11 GB | | Q3_K_M | ~12 GB | | Q3_K_L | ~13 GB | | Q4_0 | ~14 GB | | Q4_1 | ~16 GB | | Q4_K_S | ~15 GB | | Q4_K_M | ~15 GB | | Q5_0 | ~17 GB | | Q5_1 | ~19 GB | | Q5_K_S | ~17 GB | | Q5_K_M | ~18 GB | | Q6_K | ~21 GB | | Q8_0 | ~27 GB | ## Recommended Settings ## Links - : - : - : - :
Source excerpts
3 excerpts
tags: text-generation-inference transformers qwen3.6 heretic uncensored decensored abliterated reproducible llama-cpp gguf-my-repo license: apache-2.0 basemodel: darkc0de/Qwen3.6-27B-Claude-Opus-Reasoning-Distill-v2-heretic datasets:…
Base Model
Architecture
Parameters
Context Length
License
Unsloth + HuggingFace TRL
Heretic v1.3.0
reproduce/
Abliteration Parameters:
direction_index
attn.o_proj.max_weight
attn.o_proj.max_weight_position
attn.o_proj.min_weight
attn.o_proj.min_weight_distance
mlp.down_proj.max_weight
mlp.down_proj.max_weight_position
mlp.down_proj.min_weight
mlp.down_proj.min_weight_distance
KL Divergence
Refusals
ARC
ARC-E
BoolQ
HellaSwag
OpenBookQA
PIQA
WinoGrande
*-f16.gguf
*-Q1_0.gguf
*-Q2_K.gguf
*-Q3_K_S.gguf
*-Q3_K_M.gguf
*-Q3_K_L.gguf
*-Q4_0.gguf
*-Q4_1.gguf
*-Q4_K_S.gguf
*-Q4_K_M.gguf
*-Q5_0.gguf
*-Q5_1.gguf
*-Q5_K_S.gguf
*-Q5_K_M.gguf
*-Q6_K.gguf
*-Q8_0.gguf
f16
Q1_0
Q2_K
Q3_K_S / Q3_K_M / Q3_K_L
Q4_0 / Q4_1
Q4_K_S / Q4_K_M
Q5_0 / Q5_1
Q5_K_S / Q5_K_M
Q6_K
Q8_0
Temperature: 0.7 Top-p: 0.9 Top-k: 40 Repeat penalty: 1.1
Original model
Base model
Distillation source
Heretic tool
Summary
Jul 13
GGUF quantizations of darkc0de/Qwen3.6-27B-Claude-Opus-Reasoning-Distill-v2-heretic Model Overview
fzcfweasdferttgg/Qwen3.6-27B-Claude-Opus-Reasoning-Distill-v2-heretic-GGUF