What are the hardware requirements for Darkstar Qwen3.8 27B Abliterated BF16?

Darkstar Qwen3.8 27B Abliterated BF16 requires approximately 55.6 GB of VRAM. Quantized variants may run on less VRAM; see the Quantized Models section on this page.

Who developed Darkstar Qwen3.8 27B Abliterated BF16 and how large is it?

Darkstar Qwen3.8 27B Abliterated BF16 is developed by HangGlidersRule, a model with 27b parameters. The model is published as open weights on Hugging Face and indexed on LLM Explorer with full benchmark history.

Where can I download or evaluate Darkstar Qwen3.8 27B Abliterated BF16?

Darkstar Qwen3.8 27B Abliterated BF16 is hosted on Hugging Face and linked from this page. LLM Explorer also lists quantized variants and similar alternatives if available.

Darkstar Qwen3.8 27B Abliterated BF16 by HangGlidersRule — VRAM 55.6GB

Name: Darkstar Qwen3.8 27B Abliterated BF16
Author: HangGlidersRule

Darkstar Qwen3.8 27B Abliterated BF16 is an open-source language model by HangGlidersRule. Features: 27b LLM, VRAM: 55.6GB, License: apache-2.0, LLM Explorer Score: 0.31.

Abliterated Base model:finetune:qwen/qwen3... Base model:qwen/qwen3.8-27b Bf16 Conversational Darkstar Qwen3.8 Qwen3 5 Reduced-refusal Region:us Safetensors Sharded Tensorflow Vllm

Model Card on HF 🤗: https://huggingface.co/HangGlidersRule/Darkstar-Qwen3.8-27B-Abliterated-BF16

Darkstar Qwen3.8 27B Abliterated BF16 Parameters and Internals

LLM Name	Darkstar Qwen3.8 27B Abliterated BF16
Repository 🤗	https://huggingface.co/HangGlidersRule/Darkstar-Qwen3.8-27B-Abliterated-BF16
Base Model(s)	Qwen/Qwen3.8-27B Qwen/Qwen3.8-27B
Model Size	27b
Required VRAM	55.6 GB
Updated	2026-08-22
Maintainer	HangGlidersRule
Model Type	qwen3_5
Model Files	4.0 GB: 1-of-18 3.0 GB: 2-of-18 2.5 GB: 3-of-18 4.0 GB: 4-of-18 2.1 GB: 5-of-18 4.0 GB: 6-of-18 2.1 GB: 7-of-18 4.0 GB: 8-of-18 2.1 GB: 9-of-18 4.0 GB: 10-of-18 2.1 GB: 11-of-18 4.0 GB: 12-of-18 2.1 GB: 13-of-18 4.0 GB: 14-of-18 2.1 GB: 15-of-18 4.0 GB: 16-of-18 2.1 GB: 17-of-18 3.4 GB: 18-of-18
Model Architecture	Qwen3_5ForConditionalGeneration
License	apache-2.0
Model Max Length	262144
Transformers Version	5.8.0.dev0
Tokenizer Class	Qwen2Tokenizer
Padding Token	<\|endoftext\|>
Errors	replace

Rank the Darkstar Qwen3.8 27B Abliterated BF16 capabilities

Have you tried this model? Rate its performance — this feedback helps the ML community find the right model for their needs.

Instruction Following and Task Automation
Factuality and Completeness of Knowledge
Censorship and Alignment
Data Analysis and Insight Generation
Text Generation
Text Summarization and Feature Extraction
Code Generation
Multi-Language Support and Translation

Best Alternatives to Darkstar Qwen3.8 27B Abliterated BF16

Best Alternatives	Context / RAM	Downloads	Likes
Qwen3.8 27B FP8	0K / 0.4 GB	123157	416
Qwen3.8 27B	0K / 55.6 GB	2727	21
Qwen3.6 27B	0K / 55.4 GB	6728138	2197
Qwen3.8 27B	0K / 55.6 GB	2	8753
Qwen3.6 27B NVFP4	0K / 22 GB	1887525	419
Qwen3.8 27B Escha W2	0K / 10.1 GB	561	97
ThinkingCap Qwen3.6 27B	0K / 54.7 GB	35947	614
Qwen3.6 27B FP8	0K / 0.4 GB	7740472	342
Qwen3.5 27B	0K / 55.2 GB	2536294	1028
Qwen3.5 27B	0K / 55.2 GB	8115	16

Note: green Score (e.g. "73.2") means that the model is better than HangGlidersRule/Darkstar-Qwen3.8-27B-Abliterated-BF16.