0bserverx / text-generation updated 1 month ago

Qwen3.8-27B-Heretic-Abliterated-Uncensored-GGUF

RVN is a double-refined abliterated variant of Qwen3.8-27B, built on top of trohrbaugh/Qwen3.8-27B-heretic-ara (an ARA abliteration by Tim Rohrbaugh) and further refined with two additional full-weight ARA passes targeting residual refusals. It retains very low behavioral damage (KL ≈ 0.0085) while reducing harmful-prompt refusals from 3/100 (source) to 0–1/100 in independent measurements.

Params
Context
Downloads 30d
1.8 M
Likes
510
Commercial use: allowed apache-2.0 Not gated GGUF View on Hugging Face ↗

Download history

daily snapshots · 32 days
▲ 1.5 M in the last 30 days (459.9%)
2.0 M245 K
Aug 20Aug 30Sep 10Sep 20

Can you run it?

Estimated VRAM at 8K context unless noted. Pick your hardware to see the verdict per quantization.

FileQuantSizeEst. VRAMVerdict on RTX 4090 · 24 GB
RVN-IQ1_S-multilingual-mtp.gguf IQ1_S 7.6 GB 8.9 GB ✅ Runs comfortably
RVN-IQ1_M-multilingual-mtp.gguf IQ1_M 8.1 GB 9.4 GB ✅ Runs comfortably
RVN-IQ2_XXS-multilingual-mtp.gguf IQ2_XXS 8.9 GB 10.3 GB ✅ Runs comfortably
RVN-IQ2_XS-multilingual-mtp.gguf IQ2_XS 9.5 GB 11.0 GB ✅ Runs comfortably
RVN-IQ2_S-multilingual-mtp.gguf IQ2_S 9.8 GB 11.3 GB ✅ Runs comfortably
RVN-IQ2_M-multilingual-mtp.gguf IQ2_M 10.5 GB 12.0 GB ✅ Runs comfortably
RVN-Q4_K_M-multilingual-vision.gguf Q4_K_M 18.7 GB 21.1 GB ⚠️ Tight — reduce context
RVN-BF16-mtp.gguf GGUF 54.3 GB 60.2 GB ❌ Won’t fit
Estimate: file size × 1.1 + KV cache at 8K + 0.5 GB overhead. Not a benchmark — how we calculate this.

Run it

copy-paste, exact tags checked against the Hub
~ · ollama · Q4_K_M
$ ollama run qwen3-8-27b-heretic-abliterated-uncensored-gguf

# pin the quantization explicitly
$ ollama run qwen3-8-27b-heretic-abliterated-uncensored-gguf-q4_k_m
est. VRAM 21.1 GBon RTX 4090 · 24 GBJSON API →

Specifications

Licence
apache-2.0
First seen on the Hub
2026-08-14
Base model
Qwen3.8-27B
Training datasets
undisclosed
Added to our catalog
2026-08-20

Family

Base model and the most-downloaded derivatives in the catalog.

Compare with any text-generation model