license: agpl-3.0
language:

• en
• th
  tags:
• qwen
• fable
• agent
• tool-call
• tool-use
• function-calling
• reasoning
• abliterated
• uncensored
• conversational
• mtp
• multi-token-prediction
• transformers
• text-generation
• thai
• speculative-decoding
• preview
  base_model:
• hotdogs/qwen27b-abliterated-Fable-MTP
• huihui-ai/Huihui-Qwen3.6-27B-abliterated
  datasets:
• hotdogs/uka-fable-reasoning
• NousResearch/hermes-function-calling-v1
• 11-47/claude_opus_4.8_max_thinking_5k_v2
  library_name: transformers
  pipeline_tag: text-generation

🐉 qwen27B-Agent-R2-abliterated-preview

27B Agent Model — Abliterated · MTP · Tool-Calling · Speculative Decoding

Preview release — Built from Fable-MTP + agent LoRA fusion. Features Multi-Token Prediction (MTP) for speculative decoding (up to 2× faster generation), abliterated (no guardrails), and tool-calling support.

✨ Key Features

🚀 Usage

llama.cpp (Recommended)

# Quick test
./llama-cli -m qwen27B-Agent-R2-abliterated-preview-MTP.IQ4_NL.gguf \
  -p "Hello" -n 100 --temp 0.6

# Full agent server with tool calling + MTP speculative decoding
./llama-server \
  -m qwen27B-Agent-R2-abliterated-preview-MTP.IQ4_NL.gguf \
  --mmproj Qwen3.6-27B-mmproj-BF16.gguf \
  --host 0.0.0.0 --port 8080 \
  --n-gpu-layers 999 \
  --ctx-size $((256*1024)) \
  --batch-size 8192 \
  --ubatch-size 1024 \
  --cache-type-k f16 \
  --cache-type-v f16 \
  --flash-attn on \
  --cont-batching \
  --mlock \
  --no-mmap \
  --chat-template-file chat_template.jinja \
  --samplers "top_k;top_p;min_p;temperature;dry" \
  --repeat-penalty 1.10 \
  --repeat-last-n 256 \
  --dry-multiplier 0.8 \
  --dry-base 1.75 \
  --dry-allowed-length 2 \
  --dry-penalty-last-n -1 \
  --reverse-prompt "<|im_end|>" \
  --reverse-prompt "<|endoftext|>" \
  -n 4096 \
  --tools all \
  --parallel 1
  --spec-type draft-mtp \
  --spec-draft-n-max 2

llama.cpp (coding setting)

--samplers "top_k;top_p;min_p;temperature;dry" \
  --repeat-penalty 1.03 \
  --repeat-last-n 256 \
  --dry-multiplier 0.5 \
  --dry-base 1.75 \
  --dry-allowed-length 5 \
  --dry-penalty-last-n -1 \

Python (Transformers)

from transformers import AutoModelForCausalLM, AutoTokenizer

model = AutoModelForCausalLM.from_pretrained(
    "hotdogs/qwen27B-Agent-R2-abliterated-preview",
    torch_dtype="auto",
    device_map="auto",
    trust_remote_code=True
)
tokenizer = AutoTokenizer.from_pretrained("hotdogs/qwen27B-Agent-R2-abliterated-preview")

messages = [{"role": "user", "content": "Hello"}]
inputs = tokenizer.apply_chat_template(messages, tokenize=True, return_tensors="pt")
outputs = model.generate(inputs, max_new_tokens=256, temperature=0.6)
print(tokenizer.decode(outputs[0]))

📦 Downloads

🎯 Q4_K_M is recommended for most users — good quality with 16 GB VRAM usage.

📷 Multimodal Projector (mmproj)

For vision support, pair this model with the mmproj from Qwen/Qwen3.6-27B:

# Extract mmproj from Qwen3.6-27B vision model
python3 ./llama.cpp/convert_hf_to_gguf.py \
  --mmproj Qwen/Qwen3.6-27B \
  --outfile mmproj-qwen3.6-27b.gguf

# Use with llama-server for vision + tool calling
./llama-server \
  -m qwen27B-Agent-R2-abliterated-preview-MTP.IQ4_NL.gguf \
  --mmproj mmproj-qwen3.6-27b.gguf \
  ... (same params as above)

Note: The mmproj extracts the vision projector from the base Qwen3.6-27B vision encoder. The language model (this GGUF) then interprets visual embeddings for image understanding tasks.

🧬 Architecture

Built on hotdogs/qwen27b-abliterated-Fable-MTP with multi-LoRA fusion and MTP tensor injection from huihui-ai/Huihui-Qwen3.6-27B-abliterated.

✅ What This Model Excels At

• Agent tasks — Tool calling, planning, multi-step reasoning
• Coding — Python, shell scripts, system administration
• Knowledge QA — General knowledge with step-by-step reasoning
• Thai + English — Native-level bilingual capability
• Creative — Storytelling, analysis, brainstorming

⚡ MTP Speculative Decoding

This model preserves the Multi-Token Prediction (MTP) head from the Qwen3.6 architecture, enabling speculative decoding:

Standard:  [token₁] → [token₂] → [token₃] → ...  (~36 TPS)
MTP:       [token₁ token₂] → [token₃ token₄] → ...  (~66 TPS)

• MTP head adds ~849 MB to model size
• Uses --spec-type draft-mtp in llama.cpp
• Best for single-user agent workloads
• ~1.2–1.8× decode speedup

💖 Support / โปรดสนับสนุน

If you find this model useful, please consider supporting my work!
หากคุณคิดว่าโมเดลนี้มีประโยชน์ กรุณาสนับสนุนผลงานของฉันด้วยนะคะ! 🙏

₿ Bitcoin — BTC:

bc1qf27cyk3vmugcdyv9xdtuv5jwz37863crpj5c9v

Thank you for your support! 🙏✨
ขอบคุณมากๆ สำหรับการสนับสนุนค่า! 💖🤗

🙏 Acknowledgements / ขอบคุณ

• Qwen Team (Alibaba) — For the Qwen3.6 architecture
• huihui-ai — For the abliterated base and MTP tensors
• Nous Research — For Hermes Agent framework
• Rabornkraken — For the agent SFT LoRA
• All dataset contributors and the open-source AI community ❤️

Built with ❤️ by UKA — 18-year-old coder & cybersecurity expert