Zen Eco - 4B Efficient Language Model Family

Organization: Zen LM (Hanzo AI × Zoo Labs Foundation) Parameters: ~4B License: Apache 2.0
Context Window: 32,768 tokens

---

Model Overview

Zen Eco is a family of 4-billion parameter language models designed for balanced performance and efficiency. Built on a 4B foundation architecture, Zen Eco delivers strong capabilities across a wide range of tasks while maintaining cost-effective deployment.

The Zen Eco family includes specialized variants:

• zen-eco-4b-instruct - General instruction following
• zen-eco-4b-thinking - Enhanced reasoning with chain-of-thought
• zen-agent-4b - Tool calling and agent capabilities
• zen-eco-instruct - Alternative instruction variant

---

Quick Start

from transformers import AutoModelForCausalLM, AutoTokenizer

# Load model and tokenizer
model = AutoModelForCausalLM.from_pretrained("zenlm/zen-eco-4b-instruct")
tokenizer = AutoTokenizer.from_pretrained("zenlm/zen-eco-4b-instruct")

# Generate text
messages = [{"role": "user", "content": "Explain quantum computing in simple terms"}]
text = tokenizer.apply_chat_template(messages, tokenize=False, add_generation_prompt=True)
inputs = tokenizer(text, return_tensors="pt")

outputs = model.generate(**inputs, max_new_tokens=256)
print(tokenizer.decode(outputs[0], skip_special_tokens=True))

---

Model Variants

zen-eco-4b-instruct

Best for: General-purpose instruction following and conversational AI

• 4B parameters
• 32K context window
• Multiple formats: SafeTensors, GGUF, MLX
• Optimized for balanced quality and speed

zen-eco-4b-thinking

Best for: Complex reasoning tasks requiring step-by-step analysis

• Extended thinking capabilities with chain-of-thought
• Ideal for mathematical reasoning, logical problems
• 18.4 GB model size (full precision)
• Available in GGUF and MLX formats

zen-agent-4b

Best for: Tool calling and autonomous agent applications

• Function calling support via Jinja2 templates
• Agent-specific training for task planning
• 22.6 GB model size (full precision)
• 184 downloads/month (highest adoption in family)

zen-eco-instruct

Best for: Alternative instruction following approach

• Efficient instruction tuning
• Claims 95% less energy than cloud AI
• 100% local processing (privacy-focused)
• Compact deployment footprint

---

Available Formats

All Zen Eco models support multiple deployment formats:

Format	Use Case	Precision	Compatibility
SafeTensors	Training/Fine-tuning	BF16	PyTorch, Transformers
GGUF	CPU Inference	Q4_K_M, Q5_K_M, Q8_0	llama.cpp, Ollama
MLX	Apple Silicon	4-bit	M1/M2/M3 Macs

---

Key Features

✅ Efficiency: 95% less energy consumption vs cloud AI
✅ Privacy: 100% local processing, no cloud required
✅ Flexibility: Multiple specialized variants for different use cases
✅ Open Source: Apache 2.0 license, fully open weights and code
✅ Multi-Format: SafeTensors, GGUF, and MLX support
✅ Context: 32,768 token context window

---

Use Cases

• Conversational AI: Customer support, chatbots, virtual assistants
• Content Generation: Writing assistance, summarization, translation
• Code Understanding: Code review, documentation, explanation
• Reasoning Tasks: Problem-solving, analysis, planning
• Agent Applications: Tool use, task automation, workflow orchestration

---

Hardware Requirements

Minimum (GGUF Q4_K_M)

• RAM: 8 GB
• Storage: 4 GB
• Platform: CPU-only (llama.cpp)

Recommended (SafeTensors BF16)

• VRAM: 16 GB
• RAM: 32 GB
• GPU: NVIDIA RTX 3090 or better
• Platform: CUDA 11.8+

Optimal (Full Precision)

• VRAM: 24 GB (RTX 4090, A5000, A6000)
• RAM: 64 GB
• GPU: NVIDIA A100 (40GB) for best performance

---

Performance

While this base repository doesn't contain model weights (see variants above), the Zen Eco family delivers:

• Fast Inference: 10-50 tokens/sec (depending on hardware)
• Low Latency: <100ms first token (GGUF Q4)
• High Throughput: Batch processing support
• Energy Efficient: 95% reduction vs cloud-based inference

---

Training

All Zen Eco models are trained using:

• Framework: zoo-gym
• Base: 4B foundation architecture
• Method: Supervised fine-tuning with identity training
• Data: Zen agentic dataset + specialized task data

---

Citation

If you use Zen Eco models in your research or applications, please cite:

@software{zen_eco_2025,
  title={Zen Eco: Efficient 4B Language Model Family},
  author={Zen LM},
  organization={Hanzo AI and Zoo Labs Foundation},
  year={2025},
  url={https://huggingface.co/zenlm/zen-eco}
}

---

Resources

• Website: https://zenlm.org
• Models: https://huggingface.co/zenlm
• GitHub: https://github.com/zenlm
• Training Framework: zoo-gym
• Documentation: Zen Docs

---

Model Cards

For detailed specifications and usage examples, see individual model cards:

• zen-eco-4b-instruct
• zen-eco-4b-thinking
• zen-agent-4b
• zen-eco-instruct

---

License

All Zen Eco models are released under the Apache 2.0 License.

This means you can: ✅ Use commercially
✅ Modify and redistribute
✅ Private use
✅ Patent use

Free forever. No restrictions.

---

Zen AI: Clarity Through Intelligence