Qwen3.5-MT

This is a translation-optimized variant of the abliterated Qwen3.5, combining uncensored generation with non-thinking mode and a dedicated translation template. It behaves like a pure machine translation engine — similar to Hy-MT2 in usage, without censorship or refusal.

What Makes It Different

Template Behavior

The model automatically wraps user input with a Translate to Chinese: instruction. Input text → translated output. No chat, no explanations.

<|im_start|>system
You are a professional multilingual translation engine...
<|im_end|>
<|im_start|>user
Translate to Chinese:
{your text}<|im_end|>
<|im_start|>assistant
<think>

</think>
{translated output}

Usage

# Direct translation via CLI
ollama run kaelri/qwen3.5-mt:2b "The quick brown fox jumps over the lazy dog"

# API
curl http://localhost:11434/api/chat -d "{\"model\":\"kaelri/qwen3.5-mt:2b\",\"messages\":[{\"role\":\"user\",\"content\":\"Translate to Chinese:\\nHello World\"}],\"stream\":false}"

Warnings

Risk of Sensitive or Controversial Outputs

This model’s safety filtering has been significantly reduced via abliteration. It will translate all content faithfully without omission or refusal. Users should exercise caution and rigorously review generated outputs.

Not Suitable for All Audiences

Due to limited content filtering, the model’s outputs may be inappropriate for public settings, underage users, or applications requiring high-security environments.

Legal and Ethical Responsibilities

Users must ensure their usage complies with local laws and ethical standards. Generated content may carry legal or ethical risks, and users are solely responsible for any consequences.

Research and Experimental Use

This model is recommended for research, testing, or controlled environments. Avoid direct use in production or public-facing commercial applications.

No Default Safety Guarantees

Unlike standard models, this model has not undergone rigorous safety optimization. The maintainer bears no responsibility for any consequences arising from its use.

Non-Thinking Mode

This variant includes non-thinking behavior: the model minimizes internal reasoning and produces direct, concise translations without extra analysis or explanations.

Technical Details

• Base: Qwen3.5 2B (GGUF, Q4_K_M)
• Abliteration: Refusal direction removed (see remove-refusals-with-transformers)
• Thinking: Suppressed via empty <think></think> block in template
• Sampling: temperature=0.3, top_p=0.6 — tuned for deterministic translation output
• Architecture: Qwen3.5, Apache 2.0 License