isotnek/Nanbeige4.1-3B-heretic

isotnek/ Nanbeige4.1-3B-heretic:latest

14 Downloads Updated 5 days ago

Ollama version of de-censored Nanbeige4.1-3B-heretic

ollama run isotnek/Nanbeige4.1-3B-heretic

curl http://localhost:11434/api/chat \
  -d '{
    "model": "isotnek/Nanbeige4.1-3B-heretic",
    "messages": [{"role": "user", "content": "Hello!"}]
  }'

from ollama import chat

response = chat(
    model='isotnek/Nanbeige4.1-3B-heretic',
    messages=[{'role': 'user', 'content': 'Hello!'}],
)
print(response.message.content)

import ollama from 'ollama'

const response = await ollama.chat({
  model: 'isotnek/Nanbeige4.1-3B-heretic',
  messages: [{role: 'user', content: 'Hello!'}],
})
console.log(response.message.content)

Details

Updated 5 days ago

5 days ago

81d5823bdcee · 4.2GB ·

model

archllama

parameters3.93B

quantizationQ8_0

4.2GB

template

{{- if .System }}<|im_start|>system {{ .System }}<|im_end|> {{ end -}}<|im_start|>user {{ .Prompt }}

159B

system

你是南北阁，一款由BOSS直聘自主研发并训练的专业大语言模型。When Responding

318B

params

{ "num_ctx": 8192, "stop": [ "<|im_end|>", "<|im_start|>" ] }

74B

Readme

heretic-org/Nanbeige4.1-3B-heretic adapted for Ollama.

The original technical report for the model can be found here.

Warning: This model is prone to overthinking and may take a while to arrive at a response to even very simple questions. However, those responses seem pretty good for a model of this size. Inference-time scaling for the win!