Files
2026-02-26 19:50:59 -08:00

2.4 KiB

Zen4 Ultra

Zen4 Ultra — frontier-scale language model. Largest dense model in the Zen4 family.

License

Overview

Zen4 Ultra is a 405B dense parameter model delivering frontier-level performance across reasoning, code, math, science, and multilingual tasks. Designed for maximum capability with no routing overhead.

Key Features

  • 405B dense parameters — no MoE routing, full parameter utilization
  • 128K context window for long-document understanding
  • State-of-the-art reasoning and instruction following
  • Strong multilingual and cross-domain performance
  • Extended thinking and chain-of-thought capabilities

Quickstart

from transformers import AutoModelForCausalLM, AutoTokenizer

model_id = "zenlm/zen4-ultra"

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

messages = [
    {"role": "user", "content": "Explain the implications of Godel's incompleteness theorems for artificial general intelligence."},
]

inputs = tokenizer.apply_chat_template(messages, return_tensors="pt").to(model.device)
outputs = model.generate(inputs, max_new_tokens=4096)
print(tokenizer.decode(outputs[0][inputs.shape[-1]:], skip_special_tokens=True))

Serving

For production inference, use vLLM with tensor parallelism:

vllm serve zenlm/zen4-ultra \
  --tensor-parallel-size 8 \
  --max-model-len 131072 \
  --port 8000

GGUF Quantized

Quantized GGUF models for local inference with llama.cpp:

Zen4 Family

Model Parameters Focus
Zen4 Ultra 405B dense Frontier general
Zen4 Coder Pro 80B MoE Professional coding
Zen4 Coder 32B Code generation
Zen4 32B General purpose
Zen4 Mini 8B Efficient deployment

Apache 2.0 · Zen LM · Hanzo AI