ZeroXClem/Qwen3-4B-Valiant-Polaris

🧠 ZeroXClem/Qwen-4B-Valiant-Polaris

Overview

ZeroXClem/Qwen-4B-Valiant-Polaris is a thoughtfully blended model crafted using Model Stock merging via MergeKit. It fuses the structured reasoning of Polaris, the creative expressiveness of Dot-Goat and RP-V3, and the scientific depth of ShiningValiant3 into a powerful 4B architecture built atop the official Qwen/Qwen3-4B.

Designed for enhanced reasoning, uncensored creativity, deep roleplay, and advanced agentic performance — this model is both lightweight and intellectually formidable.

---

🔧 Merge Details

• Merge Method: model_stock
• Base Model: Qwen/Qwen3-4B
• Dtype: bfloat16
• int8_mask: true
• normalize: false
• Tokenizer Source: Qwen/Qwen3-4B

Merge Configuration

models:
  - model: bunnycore/Qwen3-4B-Dot-Goat
  - model: bunnycore/Qwen3-4B-RP-V3
  - model: POLARIS-Project/Polaris-4B-Preview
  - model: ValiantLabs/Qwen3-4B-ShiningValiant3
  - model: Qwen/Qwen3-4B
merge_method: model_stock
base_model: Qwen/Qwen3-4B
normalize: false
int8_mask: true
dtype: bfloat16
tokenizer_source: Qwen/Qwen3-4B

---

🧬 Models Merged

🐐 bunnycore/Qwen3-4B-Dot-Goat

Uncensored, multi-domain, LoRA-infused Qwen model focusing on creativity, tool-use, and deep chat alignment.

🎭 bunnycore/Qwen3-4B-RP-V3

Character-rich roleplay personality fusion from the amoral, mixture-of-thought, and SuperbEmphasis trees.

🌌 POLARIS-Project/Polaris-4B-Preview

Post-trained with advanced reinforcement learning on reasoning-heavy datasets. Surpasses Claude Opus & Grok on math and logic benchmarks.

✨ ValiantLabs/Qwen3-4B-ShiningValiant3

Expertly aligned to scientific reasoning, agentic workflows, and multi-domain creative logic.

🔧 Qwen/Qwen3-4B

Official pretrained Qwen3 model with support for thinking / non-thinking modes, multilingual reasoning, and tool-calling capabilities.

---

✨ Features & Highlights

🔹 Advanced Reasoning — Polaris post-training brings SOTA performance in chain-of-thought, math, and symbolic logic.

🔹 Roleplay & Uncensored Expressiveness — RP-V3 and Dot-Goat contribute dynamic personas and emotion-rich conversational modeling.

🔹 Scientific & Engineering Alignment — ShiningValiant3 ensures excellent handling of complex scientific and analytical queries.

🔹 Multimodal-Friendly & Tool-Aware — Qwen’s native agentic design enables external tool use and seamless task execution.

🔹 Lightweight Excellence — At just 4B parameters, this model performs impressively for its size with long context (32k+) and efficient inference.

---

🎯 Use Cases

• 💬 Conversational & RP Agents
• 📚 Scientific Reasoning & Educational Tutoring
• 🔍 Advanced Math & Logic Problem Solving
• ✍️ Creative Writing & Storyworld Simulation
• 🧠 Tool-Integrated Autonomous Agents

---

🚀 Usage Instructions

from transformers import AutoModelForCausalLM, AutoTokenizer

model_name = "ZeroXClem/Qwen-4B-Valiant-Polaris"

tokenizer = AutoTokenizer.from_pretrained(model_name)
model = AutoModelForCausalLM.from_pretrained(
    model_name,
    torch_dtype="auto",
    device_map="auto"
)

prompt = "Solve: What is the smallest prime greater than 100?"
messages = [{"role": "user", "content": prompt}]

text = tokenizer.apply_chat_template(
    messages,
    tokenize=False,
    add_generation_prompt=True,
    enable_thinking=True
)

inputs = tokenizer([text], return_tensors="pt").to(model.device)
outputs = model.generate(**inputs, max_new_tokens=256)
print(tokenizer.decode(outputs[0], skip_special_tokens=True))

---

🧭 Alignment & Ethics

⚠️ Unfiltered Behavior: Some sub-models are uncensored and may produce unmoderated content. Please implement safety layers when deploying in public-facing apps.

⚠️ Responsible Use: Outputs are governed by their inputs. Always review critical output for bias, hallucination, or ethical misalignment.

📜 License: Apache 2.0 + governed by respective base model licenses (see individual repos).

---

💌 Feedback & Contributions

Got thoughts, benchmarks, or new merge suggestions? We’d love to hear from you! Feel free to:

• Submit issues or pull requests 💡
• Tag us in your Hugging Face projects ❤️
• Join the discussion around merging and alignment at @ZeroXClem on HF and GitHub!

---

ZeroXClem Team | 2025 ✨