🦅 ZeroXClem/Qwen3-4B-Sky-High-Hermes

ZeroXClem/Qwen3-4B-Sky-High-Hermes is a highly distilled and precision-merged Qwen3-based 4B parameter model optimized for ultra-high reasoning, neutral-aligned autonomy, and elevated conversational depth.

This model is an advanced evolution of ZeroXClem/Qwen3-4B-Hermes-Axion-Pro, combining multiple state-of-the-art Heretic Abliterated reasoning experts with Claude 4.5, Gemini 3, Opus 3, and Haiku distillations — all under a finely tuned 262,144 token context window.

🧠 Merge Methodology

Merge Method: model_stock
Base Model: ZeroXClem/Qwen3-4B-Hermes-Axion-Pro
Precision: bfloat16
Tokenizer Source: Qwen/Qwen3-4B-Thinking-2507
Context Length: 262,144 tokens

name: ZeroXClem/Qwen3-4B-Sky-High-Hermes
base_model: ZeroXClem/Qwen3-4B-Hermes-Axion-Pro
dtype: bfloat16
merge_method: model_stock
models:
  - model: DavidAU/Qwen3-4B-Instruct-2507-Polaris-Alpha-Distill-Heretic-Abliterated
  - model: DavidAU/Qwen3-4B-Thinking-2507-Gemini-3-Pro-Preview-High-Reasoning-Distill-Heretic-Abliterated
  - model: DavidAU/Qwen3-4B-Claude-Sonnet-4-Reasoning-Distill-Heretic-Abliterated
  - model: TeichAI/Qwen3-4B-Thinking-2507-Claude-Haiku-4.5-High-Reasoning-Distill
  - model: TeichAI/Qwen3-4B-Instruct-2507-Claude-Opus-3-Distill
  - model: TeichAI/Qwen3-4B-Thinking-2507-MiMo-V2-Flash-Distill
  - model: TeichAI/Qwen3-4B-Thinking-2507-MiniMax-M2.1-Distill
tokenizer_source: Qwen/Qwen3-4B-Thinking-2507

⚔️ Purpose & Strengths

Sky-High-Hermes is designed to soar above traditional instruction-tuned models by harmonizing long context, deep reasoning, and full-spectrum capability.

✨ Features

🔍 High-Reasoning Ability — Claude, Gemini, and MiMo distills support deep multi-hop and abstract reasoning tasks.
🧠 262K Context Window — Massive memory capacity ideal for longform tasks, coding agents, or narrative tracking.
🔓 Abliterated via Heretic Protocol — Uncensored, refusal-resistant generation with KL divergence near-zero.
💬 Sublime Dialogue & RP — High immersion support for fictional characters, story continuations, and dynamic roleplay.
⚙️ Engineering-Grade — Code writing, debugging, and architectural reasoning.
🧬 Unsloth Optimized — Fast inference and training via TRL and Unsloth acceleration.

🛠️ Recommended Use

This is a Class 1 model suitable for:

Creative Writing
Autonomous Agents
Code Assistance
Science & Research
Story & Worldbuilding
Philosophical Discourse
NSFW (Uncensored Use)

For optimal inference:

Set smoothing_factor = 1.5 (in KoboldCpp, WebUI, or SillyTavern)
Enable enable_thinking=True for structured prompt reasoning

🚀 Example Usage

from transformers import AutoModelForCausalLM, AutoTokenizer

model = "ZeroXClem/Qwen3-4B-Sky-High-Hermes"
tokenizer = AutoTokenizer.from_pretrained(model)
model = AutoModelForCausalLM.from_pretrained(model, torch_dtype="auto", device_map="auto")

prompt = "Explain the philosophical connection between memory and time."
messages = [{"role": "user", "content": prompt}]

text = tokenizer.apply_chat_template(messages, tokenize=False, add_generation_prompt=True)
inputs = tokenizer([text], return_tensors="pt").to(model.device)
outputs = model.generate(**inputs, max_new_tokens=512)
print(tokenizer.decode(outputs[0], skip_special_tokens=True))

🔐 Licensing & Alignment

License: Apache 2.0
Censorship: Abliterated via Heretic v1.0.1
KL Divergence: ~0.00 – 0.06
Refusal Rate: ≤ 8/100
Content Moderation: Recommended for production use

💌 Acknowledgements

Gratitude to:

DavidAU – For crafting and liberating high-reasoning uncensored blends
TeichAI – For high-resolution Claude/Gemini distillations
Unsloth Team – For inference-accelerated Qwen3 tuning frameworks
Mergkit Acree Team - For providing such a fantastic framework to merge these LLMS
Alibaba Qwen Team - For providing such amazing open source base models

🕊️ Soar higher. Think freer. Respond truer. Built with love by ZeroXClem | 2026

ZeroXClem/Qwen3-4B-Sky-High-Hermes