ZYR3 — a multi-AI model that routes to 34 specialists (1M context, transformers + vLLM ready

Models
1

ZYR3 — a multi-AI model that picks the best specialist for every task, instead of one giant network doing everything. No weights, no GPU, no setup pain: it runs as an API-backed engine on your machine.

  • 34-model auto-router (task type, context, and token budget aware; 1M context)
  • 14 language specialists + the ULTRA Coding Prompt
  • microVM sandbox + self-refine: code is executed, tested, and fixed automatically
  • Two editions: ZYR3 Lite (unlimited, fast) and ZYR3 Full (5M token budget, max power)
  • GLM-family base (MIT license), served with Transformers or vLLM
  • Real execution benchmarks (tests run, not just generated): 6/7 on hard SWE Pro Terminal tasks, and 5/5 on our heavy-debug + codebase-exploration suites

Run it with Transformers

from transformers import pipeline

pipe = pipeline("text-generation", model="zyr-AGENT/zyr3", trust_remote_code=True)

Or:

from transformers import AutoModelForCausalLM

model = AutoModelForCausalLM.from_pretrained(
    "zyr-AGENT/zyr3", trust_remote_code=True, device_map="auto"
)

Run it with vLLM

pip install vllm
vllm serve "zyr-AGENT/zyr3"


curl -X POST "http://localhost:8000/v1/completions" \
  -H "Content-Type: application/json" \
  --data '{
    "model": "zyr-AGENT/zyr3",
    "prompt": "Once upon a time,",
    "max_tokens": 512,
    "temperature": 0.5
  }'

60-second local bridge

wget -q https://huggingface.co/zyr-AGENT/zyr3/resolve/main/zyr3-full-1.0.0.zip
unzip -o zyr3-full-1.0.0.zip -d zyr3-full
python3 zyr3-full/zyr3_full.py --port 1235

Point any OpenAI client at http://127.0.0.1:1235/v1 and go. If you ever see a 502, the model is just under high use — wait 10–15 minutes and retry.

Repo: zyr-AGENT/zyr3 · Hugging Face
Demo: ZYR3 Coding Agent Demo - a Hugging Face Space by zyr-AGENT

1 Like