Instructions to use ZeroXClem/Qwen3-4B-Hermes-Axion-Pro with libraries, inference providers, notebooks, and local apps. Follow these links to get started.

Libraries

How to use ZeroXClem/Qwen3-4B-Hermes-Axion-Pro with Transformers:

# Use a pipeline as a high-level helper
from transformers import pipeline

pipe = pipeline("text-generation", model="ZeroXClem/Qwen3-4B-Hermes-Axion-Pro")
messages = [
    {"role": "user", "content": "Who are you?"},
]
pipe(messages)

# Load model directly
from transformers import AutoTokenizer, AutoModelForCausalLM

tokenizer = AutoTokenizer.from_pretrained("ZeroXClem/Qwen3-4B-Hermes-Axion-Pro")
model = AutoModelForCausalLM.from_pretrained("ZeroXClem/Qwen3-4B-Hermes-Axion-Pro")
messages = [
    {"role": "user", "content": "Who are you?"},
]
inputs = tokenizer.apply_chat_template(
	messages,
	add_generation_prompt=True,
	tokenize=True,
	return_dict=True,
	return_tensors="pt",
).to(model.device)

outputs = model.generate(**inputs, max_new_tokens=40)
print(tokenizer.decode(outputs[0][inputs["input_ids"].shape[-1]:]))

HERMES

How to use ZeroXClem/Qwen3-4B-Hermes-Axion-Pro with HERMES:

# No code snippets available yet for this library.

# To use this model, check the repository files and the library's documentation.

# Want to help? PRs adding snippets are welcome at:
# https://github.com/huggingface/huggingface.js

Notebooks
Google Colab
Kaggle
Local Apps

vLLM

How to use ZeroXClem/Qwen3-4B-Hermes-Axion-Pro with vLLM:

Install from pip and serve model

# Install vLLM from pip:
pip install vllm
# Start the vLLM server:
vllm serve "ZeroXClem/Qwen3-4B-Hermes-Axion-Pro"
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:8000/v1/chat/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "ZeroXClem/Qwen3-4B-Hermes-Axion-Pro",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Use Docker

docker model run hf.co/ZeroXClem/Qwen3-4B-Hermes-Axion-Pro

SGLang

How to use ZeroXClem/Qwen3-4B-Hermes-Axion-Pro with SGLang:

Install from pip and serve model

# Install SGLang from pip:
pip install sglang
# Start the SGLang server:
python3 -m sglang.launch_server \
    --model-path "ZeroXClem/Qwen3-4B-Hermes-Axion-Pro" \
    --host 0.0.0.0 \
    --port 30000
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:30000/v1/chat/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "ZeroXClem/Qwen3-4B-Hermes-Axion-Pro",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Use Docker images

docker run --gpus all \
    --shm-size 32g \
    -p 30000:30000 \
    -v ~/.cache/huggingface:/root/.cache/huggingface \
    --env "HF_TOKEN=<secret>" \
    --ipc=host \
    lmsysorg/sglang:latest \
    python3 -m sglang.launch_server \
        --model-path "ZeroXClem/Qwen3-4B-Hermes-Axion-Pro" \
        --host 0.0.0.0 \
        --port 30000
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:30000/v1/chat/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "ZeroXClem/Qwen3-4B-Hermes-Axion-Pro",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Docker Model Runner
How to use ZeroXClem/Qwen3-4B-Hermes-Axion-Pro with Docker Model Runner:
```
docker model run hf.co/ZeroXClem/Qwen3-4B-Hermes-Axion-Pro
```
Browse Quantizations to use this model in llama.cpp, Ollama, LM Studio, or any compatible app.

🧠 ZeroXClem-Qwen3-4B-Hermes-Axion-Pro

Overview

ZeroXClem-Qwen3-4B-Hermes-Axion-Pro is a powerful, safety-conscious, and deeply intelligent merge crafted via Model Stock merging using MergeKit. This 4B-parameter model blends the best of Hermes-3, Axion-Thinking, and Qwen3-Pro, optimized for deep reasoning, safe generation, and dynamic roleplay.

It’s designed to excel in structured problem-solving, multi-turn dialogue, and creative writing, while maintaining safe behavior aligned through red teaming and post-training.

This model is VERY good with reasoning, and hard tasks! Use the default Template(Jinja) setting in LMStudio for best inference.

🔧 Merge Details

Model Name: ZeroXClem-Qwen3-4B-Hermes-Axion-Pro
Merge Method: model_stock
Base Model: bunnycore/Qwen3-4B-Pro
Dtype: bfloat16
Tokenizer Source: Qwen/Qwen3-4B-Thinking-2507

YAML Configuration

name: ZeroXClem-Qwen3-4B-Hermes-Axion-Pro
base_model: bunnycore/Qwen3-4B-Pro
dtype: bfloat16
merge_method: model_stock
models:
  - model: ertghiu256/Qwen3-4b-tcomanr-merge-v2.2
  - model: ertghiu256/Qwen3-4B-Thinking-2507-Hermes-3
  - model: Qwen/Qwen3-4B-Thinking-2507
tokenizer_source: Qwen/Qwen3-4B-Thinking-2507

🧬 Models Merged

🧠 `ertghiu256/Qwen3-4B-Thinking-2507-Hermes-3`

Finetuned on the Hermes 3 dataset for instruction alignment and coherent multi-step thinking.

🔒 `AdvRahul/Axion-Thinking-4B`

Safety-tested and enhanced via red teaming protocols. Based on Qwen3-4B-Thinking-2507, with refined behavior for ethical deployment.

🧰 `ertghiu256/Qwen3-4b-tcomanr-merge-v2.2`

Strong logic and instruction-following merge with emphasis on quality output in diverse domains.

💼 `bunnycore/Qwen3-4B-Pro`

A professional-grade Qwen variant tuned for real-world applications like coding, RP, creative writing, and structured tasks.

✨ Features & Highlights

🔹 Deep Thinking & Problem Solving — Inspired by Hermes-3 and Axion, this model handles multi-step logical reasoning and instruction-following with clarity.

🔹 Safe, Aligned Outputs — Red team finetuning and post-training ensure behavior safety and moderation-ready generation.

🔹 Creative Writing & Roleplay — Retains high fluency and character immersion for natural roleplay and storytelling.

🔹 Coding & Engineering Tasks — Competent in code generation, debugging, and technical explanations.

🔹 Efficient & Lightweight — At just 4B parameters, it's easy to deploy locally or in constrained environments.

🎯 Use Cases

🤖 Conversational AI
✍️ Creative Roleplay & Fiction Writing
🧠 Reasoning & Problem-Solving Tasks
🧑‍💻 Code Generation & Completion
🔐 Safe AI Assistants with Aligned Behavior

🚀 Usage Instructions

For optimal inference, use a higher quant such as Q6 Here.

from transformers import AutoModelForCausalLM, AutoTokenizer

model_name = "ZeroXClem/Qwen3-4B-Hermes-Axion-Pro"

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

prompt = "Describe the principles of quantum entanglement in simple terms."
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=512)
print(tokenizer.decode(outputs[0], skip_special_tokens=True))

For LM Studio Users: When using this model in LM Studio, select the Qwen3-Chat Template from the template dropdown menu. This official template ensures proper prompt formatting for consistent multi-turn conversations and system instruction handling.

⚠️ Alignment & Ethics

🔐 Safety Notice: While post-trained with red teaming protocols, this model still outputs raw generations. Always include content moderation for public deployments.
🧠 Thinking Mode Support: Fully compatible with enable_thinking=True and /think prompt control.
📜 License: Apache 2.0 + governed by the licenses of upstream models.