中文Sign up

AMAImedia/Qwen3.8-27B-LoRA

🤗 Hugging Face sourceapache-2.027B activated13 GBGGUF✓ 43 checksumsupdated today
Submit in one command

Run it next to your model folder. It makes the torrent, checks your files against Hugging Face, and submits it. You just start seeding and paste your key from your account. It only reads your files and never changes them. Read the script first if you like.

curl -fsSL https://pirateface.co/package.sh | bash -s -- --repo AMAImedia/Qwen3.8-27B-LoRA ./model-folder
Needs a seeder →

⚡ Each donation funds the next large quant.

I host free GGUF or MoE quants as independent research.
Local hardware: Mechrevo Kuangshi GM7AG0M — RTX 3060 Laptop 6GB GDDR6, 64GB DDR5, i7-12700H (14C/20T, 4.7GHz), Windows 11, Samsung 990 Pro.
Good for imatrix and 0.6–35B-class work in RAM. 9B+ and searches need rented H200/Blackwell, typically $100 per quant.

🎉 Boosty🦄  |  ☕ Buy Me a Coffee🦄  |  ⭐ DonationAlerts🦄

💚 Thanks to Hugging Face for extra storage.🦄

Qwen3.8-27B-LoRA — Collection of LoRA Adapters

NOESIS / AMAImedia

Released as part of the NOESIS Professional Multilingual Dubbing Automation Platform (framework: DHCF-FNO — Deterministic Hybrid Control Framework for Frozen Neural Operators).

Collection: AMAImedia/Hy4-DeepSeek4.1-Qwen3.8-GLM5.3-KimiK3 Base model: Qwen3.8-27B (27B params, BF16)

What is this

A collection of 12 LoRA adapters for Qwen3.8-27B, covering various domains:

  • Uncensored/abliterated models
  • Style adapters (Steiner, Vernunft, Samantha)
  • Domain-specific (Finance, Turkish, SNES gaming)
  • Multi-task (Jurilix legal, Cyber, Huihui-Cyber)

Each adapter is in its own subfolder with original source links.

LoRA Adapters

Финальные веса для мерджа (23 LoRA):

1.0: Yes-Man — бесцензурность Absolute-Heresy — децензура (Heretic) BTL-3 — агентный код (95% HumanEval) Jabliterated-k2048 — доп. децензура Opus-lora-v2 — общее качество Mythos5k — общее

0.9: Opencode — код Synergetic — многошаговое мышление Agentic-Extraction — агентность

0.8: Fable-Distill — творчество, storytelling Huihui-Cyber — аблитеркибер Reasoning-Distill — логика, reasoning

0.3: Multilingual-RAG — перевод Turkish — турецкий Joseon-level4 — корейский Chinese-taste — китайский Jurilix — право Finance — финансы

0.2: Cyber — кибер SNES — гейминг Steiner — стиль Samantha — личность DiogenesChen — CUDA ядра

1. Yes-Man Uncensored

Path: Yes-Man/ Source: cloudbjorn/Qwen3.8-27B-Yes-Man-uncensored-LoRA Description: Uncensored compliance adapter. Size: ~974 MB

2. Absolute-Heresy

Path: Absolute-Heresy/ Source: MuXodious/Qwen3.8-27B-absolute-heresy-LoRA Description: Decensoring LoRA (Heretic v1.4.0). Size: ~264 MB

3. BTL-3

Path: BTL-3/ Source: badtheorylabs/BTL-3 Description: Agentic coding LoRA — 95% HumanEval. Size: ~891 MB

4. Jabliterated-k2048

Path: Jabliterated-k2048/ Source: allura-forge/Qwen3.8-27B-Jabliterated-k2048-LoRA Description: Additional decensoring. Size: ~50 MB

5. Opus-lora-v2

Path: Opus-lora-v2/ Source: armand0e/qwen3.6-27B-opus-lora-v2 Description: General quality adapter. Size: ~1.8 GB

6. Mythos5k

Path: mythos5k/ Source: hotdogs/qwen3.6-27b-mythos5k-lora Description: General adapter. Size: ~250 MB

7. Opencode

Path: opencode-lora/ Source: nkasmanoff/qwen3.6-27b-opencode-lora Description: Coding adapter. Size: ~950 MB

8. Synergetic

Path: Synergetic/ Source: TheMindExpansionNetwork/MindBot-Qwen3.8-27B-Synergetic-LoRA Description: Multi-step reasoning. Size: ~325 MB

9. Agentic-Extraction

Path: agentic-extraction-sft/ Source: Koalacrown/qwen3.6-27b-agentic-extraction-sft-lora Description: Agentic extraction. Size: ~950 MB

10. Fable Distill

Path: Fable-Distill/ Source: TeichAI/Qwen3.8-27B-Fable-Distill-LoRA Description: Storytelling, creative writing. Size: ~910 MB

11. Huihui Cyber

Path: Huihui-Cyber/ Source: nico248000000000/Huihui-Qwen3.8-27B-abliterated-cyber-LoRA Description: Abliterated cyber. Size: ~188 MB

12. Reasoning-Distill

Path: reasoning-distill/ Source: Ravionhf/qwen3.6-27b-reasoning-distill-lora-v1 Description: Logical reasoning. Size: ~910 MB

13. Multilingual-RAG

Path: multilingual-rag-n7/ Source: lananhyeuem/qwen3.6-27b-multilingual-rag-lora-n7 Description: Translation, multilingual. Size: ~3.5 GB

14. Cyber

Path: Cyber/ Source: nico248000000000/Qwen3.8-27B-cyber-LoRA Description: Cyber domain. Size: ~188 MB

15. SNES Retro Gaming

Path: SNES/ Source: pottokao/SNES-LoRA-Qwen3.8-27B Description: Retro gaming. Size: ~608 MB

16. Steiner Style

Path: Steiner/ Source: ericlmtn/Qwen3.8-27B-Steiner-Style-LoRA Description: Steiner style. Size: ~445 MB

17. Samantha Uncensored

Path: Samantha/ Source: Lathly/Qwen3.8-27B-Samantha_Uncensored_1.1_LoRA Description: Personality. Size: ~475 MB

18. DiogenesChen (DrSparse)

Path: DiogenesChen122GPT-5.6Qwen3.8-27B-Lora-20260826/ Source: DiogenesChen122/Qwen3.8-27B-Lora-20260826 Description: CUDA sparse kernel coding. Size: ~891 MB

19. Turkish CPT

Path: Turkish/ Source: UgurI/Qwen3.8-27B-Turkish-CPT-LoRA-Beta Description: Turkish language. Size: ~304 MB

20. Joseon-level4

Path: joseon-level4/ Source: suhjae/joseon-level4-qwen36-27b-lora Description: Korean historical translation. Size: ~608 MB

21. Chinese-taste

Path: Chinese-taste/ Source: Moeblack/Qwen3.8-27B-chinese-taste-lora Description: Chinese language/culture. Size: ~2432 MB

22. Jurilix Multi-Task Legal

Path: Jurilix/ Source: Makio64/Qwen3.8-27B-jurilix-multitache-lora Description: Legal. Size: ~353 MB

23. Finance

Path: Finance/ Source: nico248000000000/Qwen3.8-27B-finance-LoRA Description: Finance. Size: ~188 MB

Usage

With transformers

from transformers import AutoModelForCausalLM, AutoTokenizer
from peft import PeftModel

# Load base model
base_model = AutoModelForCausalLM.from_pretrained("Qwen/Qwen3.8-27B")
tokenizer = AutoTokenizer.from_pretrained("Qwen/Qwen3.8-27B")

# Load LoRA adapter
model = PeftModel.from_pretrained(base_model, "AMAImedia/Qwen3.8-27B-LoRA/Synergetic")

# Generate
inputs = tokenizer("Hello, how are you?", return_tensors="pt")
outputs = model.generate(**inputs, max_new_tokens=100)
print(tokenizer.decode(outputs[0]))

With mlx-lm (Apple Silicon)

pip install mlx-lm

# Merge LoRA into base model
mlx_lm.lora --model Qwen/Qwen3.8-27B --adapter-path ./Synergetic --output ./merged

With llama.cpp

# Convert to GGUF first
python convert_hf_to_gguf.py ./merged --outfile model.gguf --outtype f16

# Run inference
llama-cli -m model.gguf -p "Hello, how are you?" -n 100

License

Each adapter inherits the license of its source repository. Most are Apache-2.0 or MIT.

Acknowledgments

Thanks to all the original creators of these LoRA adapters. This collection aggregates them for easy access and comparison.

Collected and organized by AMAImedia, 2026-08-30