denisavetisyan/gemma4-31b-q4

denisavetisyan/ gemma4-31b-q4_0-24gb-48k:latest

23 Downloads Updated 1 week ago

Production-ready copy of gemma4:31b-hf-qat-think (Unsloth QAT UD-Q4_K_XL, 30.7B) with battle-tested defaults baked in: temperature 1.0, top_p 0.95, top_k 64, num_ctx 48384. Sized for a single 24 GB GPU.

vision tools thinking

ollama run denisavetisyan/gemma4-31b-q4_0-24gb-48k

curl http://localhost:11434/api/chat \
  -d '{
    "model": "denisavetisyan/gemma4-31b-q4_0-24gb-48k",
    "messages": [{"role": "user", "content": "Hello!"}]
  }'

from ollama import chat

response = chat(
    model='denisavetisyan/gemma4-31b-q4_0-24gb-48k',
    messages=[{'role': 'user', 'content': 'Hello!'}],
)
print(response.message.content)

import ollama from 'ollama'

const response = await ollama.chat({
  model: 'denisavetisyan/gemma4-31b-q4_0-24gb-48k',
  messages: [{role: 'user', content: 'Hello!'}],
})
console.log(response.message.content)

Details

Updated 1 week ago

1 week ago

20ee302d8c00 · 18GB

model

archgemma4

parameters30.7B

quantizationQ4_0

17GB

projector

archclip

parameters576M

quantizationBF16

1.2GB

params

{ "num_ctx": 48384, "stop": [ "<bos>", "<|turn>", "<turn|>",

149B

template

{{ if .System }}<bos><|turn>system {{ .System }}<turn|> {{ end }}{{ if .Prompt }}<|turn>user {{ .Pro

159B

Readme

Production-ready copy of gemma4:31b-hf-qat-think (Unsloth QAT UD-Q4_K_XL, 30.7B) with battle-tested defaults baked in: temperature 1.0, top_p 0.95, top_k 64, num_ctx 48384. Sized for a single 24 GB GPU. Thinking-capable — pass “think”: false per request to disable (cannot be baked into the model on Ollama 0.40).