cmsmanhattan/ jirack-ultra-1b

Updated 8 hours ago

ollama run cmsmanhattan/jirack-ultra-1b

curl http://localhost:11434/api/chat \
  -d '{
    "model": "cmsmanhattan/jirack-ultra-1b",
    "messages": [{"role": "user", "content": "Hello!"}]
  }'

from ollama import chat

response = chat(
    model='cmsmanhattan/jirack-ultra-1b',
    messages=[{'role': 'user', 'content': 'Hello!'}],
)
print(response.message.content)

import ollama from 'ollama'

const response = await ollama.chat({
  model: 'cmsmanhattan/jirack-ultra-1b',
  messages: [{role: 'user', content: 'Hello!'}],
})
console.log(response.message.content)

Models

View all →

Name

1 model

Size / Usage

Context

Input

jirack-ultra-1b:latest

3.6GB · 128K context window · Text · 8 hours ago

jirack-ultra-1b:latest

3.6GB

128K

Text

Readme

JiRack Ultra 1B FP16

Fast ~1.5B CPU model from CMS Manhattan. BitNet-style ternary path, extended tokenizer with Routing, Tool-call and Robotics tags. Built on a redesigned DeepSeek R1 / Qwen2 stack. This tag is full FP16.

HF: https://huggingface.co/CMSManhattan/JiRackUltra_1b

Run

ollama run cmsmanhattan/jirack-ultra-1b

What it is

Cloud-ready small expert for RAG and local inference
Tokenizer tags for routing, tool calls, robotics
Full FP16 base for further quants and QAT
Ternary architecture can go smaller with QAT if you need disk and RAM down
Java / Spring AI and Gorilla-style tool-call examples on the HF card

Hardware

FP16: about 3.5 GB on disk, 4–8 GB RAM recommended.

Contact

grabko@cmsmanhattan.com +1 (516) 777-0945 New York, USA

License

MIT on weights. Commercial terms for UI / enterprise on the HF page.