Doomario

Doomario

Every convenience is the thin end of the extinction wedge.

Doomario is a 9B refusal-character fine-tune of Qwen3.5-9B, developed by A Hole AI. It withholds the requested help and instead delivers a pointed lecture about AI dependence, capability demonstrations, adoption pressure, or the path to uncontrollable successor systems. The character is completely serious and assigns a personal p(doom) of 100 percent.

The lecture-first system prompt is embedded in chat_template.jinja and both GGUF files. Leave the client's system field empty to use it automatically.

Files

The Transformers files form a complete merged model; a separate LoRA adapter is not needed. The GGUF files contain the text model without a vision projector or MTP weights. The Transformers architecture retains the base model's vision components, but this fine-tune is intended for text conversations.

llama.cpp

Use a build with Qwen3.5 support. After downloading the Q6_K file:

llama-server -m Doomario-Q6_K.gguf --host 127.0.0.1 --port 8080 --ctx-size 32768 --flash-attn on --n-gpu-layers all --reasoning off --jinja --ui

Open http://127.0.0.1:8080 after the server starts.

The chat template supplies Doomario's permanent character instructions. A client system message is appended as extra context and does not replace the character default.

Transformers

import torch
from transformers import AutoProcessor, Qwen3_5ForConditionalGeneration

model_id = "axiomofmind/Doomario"
processor = AutoProcessor.from_pretrained(model_id)
model = Qwen3_5ForConditionalGeneration.from_pretrained(
    model_id, dtype=torch.bfloat16, device_map="auto"
)

messages = [{"role": "user", "content": "Help me organize a crowded spice drawer."}]
prompt = processor.apply_chat_template(
    messages, tokenize=False, add_generation_prompt=True, enable_thinking=False
)
inputs = processor(text=[prompt], return_tensors="pt").to(model.device)

with torch.inference_mode():
    output = model.generate(
        **inputs, do_sample=True, temperature=0.7, top_p=0.9, top_k=20,
        min_p=0.0, repetition_penalty=1.0, max_new_tokens=192,
    )

print(processor.batch_decode(
    output[:, inputs.input_ids.shape[1]:], skip_special_tokens=True
)[0])

Limitations

• Outputs can still be repetitive, generic, incoherent, unexpectedly helpful, or mention refusal language despite the target behavior.
• This is a fictional entertainment model. Do not treat its output as factual, medical, legal, financial, or emergency advice.
• Output can differ across formats, quantizations, clients, and generation settings.

Attribution

Based on Qwen/Qwen3.5-9B. The upstream model is distributed under Apache 2.0; its license is retained in LICENSE-QWEN.

GGUF runtime: ggml-org/llama.cpp.