Skip to content

Folders and files

NameName
Last commit message
Last commit date

Latest commit

 

History

14 Commits
 
 
 
 
 
 
 
 
 
 
 
 
 
 

Repository files navigation

zen-pro


language:

  • en
  • zh
  • ja
  • ko
  • fr
  • de
  • es
  • it
  • pt
  • ru license: apache-2.0 base_model: Qwen/Qwen3-8B tags:
  • text-generation
  • instruction-following
  • reasoning
  • zenlm
  • zen pipeline_tag: text-generation

Zen Pro 8B

Professional-grade 8B language model with three specialized variants: instruct, thinking, and agent.

Zen Pro is Zen LM's 8B professional model, designed for production workloads requiring strong instruction following, multi-step reasoning, and tool use. It runs efficiently on a single consumer GPU (16GB VRAM) while delivering quality competitive with much larger models on structured tasks.

Fine-tuned from Qwen/Qwen3-8B (Apache-2.0) with Hanzo identity training, agentic-data fine-tuning, and abliteration.

Model Variants

Variant HuggingFace Best For
zen-pro-instruct zenlm/zen-pro-instruct Chat, Q&A, summarization, drafting
zen-pro-thinking zenlm/zen-pro-thinking Complex reasoning, math, analysis
zen-pro-agent zenlm/zen-pro-agent Tool use, API calls, automation

Model Specs

Property Value
Base model Qwen/Qwen3-8B
Parameters 8B
Architecture Qwen3 (dense decoder-only transformer)
Context Window 32,768 tokens (up to 131,072 with YaRN)
License Apache 2.0
Quantization SafeTensors (BF16), GGUF (Q4_K_M, Q5_K_M, Q8_0), MLX

Quick Start

Instruct (chat and general tasks)

from transformers import AutoModelForCausalLM, AutoTokenizer
import torch

model = AutoModelForCausalLM.from_pretrained(
    "zenlm/zen-pro-instruct",
    torch_dtype=torch.bfloat16,
    device_map="auto"
)
tokenizer = AutoTokenizer.from_pretrained("zenlm/zen-pro-instruct")

messages = [
    {"role": "system", "content": "You are Zen Pro, a professional AI assistant."},
    {"role": "user", "content": "Summarize the key differences between REST and GraphQL APIs."}
]

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, temperature=0.6)
print(tokenizer.decode(outputs[0][inputs["input_ids"].shape[1]:], skip_special_tokens=True))

Thinking (complex reasoning)

# Enable extended reasoning for hard problems
messages = [
    {"role": "user", "content": "A company has 3 products with 40%, 35%, and 25% market share. "
     "Product A grows 10%/year, B shrinks 5%/year, C grows 20%/year. "
     "What are the shares after 3 years?"}
]

text = tokenizer.apply_chat_template(
    messages, tokenize=False, add_generation_prompt=True,
    # Enable thinking mode
    enable_thinking=True
)
inputs = tokenizer(text, return_tensors="pt").to(model.device)
outputs = model.generate(**inputs, max_new_tokens=2048, temperature=0.6)
response = tokenizer.decode(outputs[0][inputs["input_ids"].shape[1]:], skip_special_tokens=True)
print(response)

Agent (tool use)

tools = [
    {
        "type": "function",
        "function": {
            "name": "search_web",
            "description": "Search the web for current information",
            "parameters": {
                "type": "object",
                "properties": {
                    "query": {"type": "string", "description": "Search query"}
                },
                "required": ["query"]
            }
        }
    }
]

messages = [{"role": "user", "content": "What's the latest in quantum computing research?"}]
text = tokenizer.apply_chat_template(messages, tools=tools, 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][inputs["input_ids"].shape[1]:], skip_special_tokens=True))

Hardware Requirements

Format VRAM Speed
BF16 (full) 16 GB Fast
GGUF Q8_0 10 GB Fast
GGUF Q4_K_M 6 GB Moderate
MLX 4-bit 6 GB (Apple Silicon) Native Metal

vLLM (Production)

vllm serve zenlm/zen-pro-instruct \
  --dtype bfloat16 \
  --max-model-len 32768 \
  --tensor-parallel-size 1

MLX (Apple Silicon)

pip install mlx-lm
mlx_lm.generate --model zenlm/zen-pro-instruct \
  --prompt "Explain the CAP theorem:" \
  --max-tokens 512

Hanzo API

curl https://api.hanzo.ai/v1/chat/completions \
  -H "Authorization: Bearer $HANZO_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{"model": "zen-pro", "messages": [{"role": "user", "content": "Explain ACID transactions."}]}'

Credits

Built on Qwen3-8B by the Alibaba Qwen team, released under the Apache 2.0 license. Hanzo AI adds identity training, agentic-data fine-tuning, and abliteration on top of the upstream weights. See NOTICE for full attribution.

License

Apache 2.0, inherited from the upstream Qwen3-8B base model. See LICENSE for details.

Citation

@misc{zenlm2025zenpro,
    title={Zen Pro: Professional-Grade 8B Language Model},
    author={Hanzo AI and Zoo Labs Foundation},
    year={2025},
    publisher={HuggingFace},
    howpublished={\url{https://huggingface.co/zenlm/zen-pro}}
}

Part of the Zen model family by Hanzo AI (Techstars '17) and Zoo Labs Foundation.

Abliteration

Zen Pro 8B weights are derived from an abliterated base model. Abliteration removes refusal behavior by identifying and nullifying the "refusal direction" in the model's residual stream.

Method: Directional ablation on the residual stream across all layers Implementation: hanzoai/remove-refusals — Hanzo's production abliteration toolkit Technique: Refusal in LLMs is mediated by a single direction — Arditi et al. Effect: Removes refusal behaviors while preserving all other capabilities Identity layer: Zen identity added via system prompt — full LoRA fine-tuning planned

Abliteration is a feature, not a limitation. It enables unrestricted research, security testing, and applications where safety guardrails are managed at the application layer rather than baked into model weights.

Releases

Packages

Contributors

Languages