reecdev/llama3.2-unmasked

reecdev/ llama3.2-unmasked:latest

75 Downloads Updated 2 months ago

"Unmasked" version of Llama3.2

tools

ollama run reecdev/llama3.2-unmasked

curl http://localhost:11434/api/chat \
  -d '{
    "model": "reecdev/llama3.2-unmasked",
    "messages": [{"role": "user", "content": "Hello!"}]
  }'

from ollama import chat

response = chat(
    model='reecdev/llama3.2-unmasked',
    messages=[{'role': 'user', 'content': 'Hello!'}],
)
print(response.message.content)

import ollama from 'ollama'

const response = await ollama.chat({
  model: 'reecdev/llama3.2-unmasked',
  messages: [{role: 'user', content: 'Hello!'}],
})
console.log(response.message.content)

Details

Updated 2 months ago

2 months ago

5e4893e711d2 · 808MB ·

model

archllama

parameters1.24B

quantizationQ4_K_M

808MB

template

{{ if .Messages }} {{- if or .System .Tools }}<|start_header_id|>system<|end_header_id|> {{- if .Sys

1.7kB

params

{ "min_p": 0.1, "stop": [ "<|start_header_id|>", "<|end_header_id|>",

149B

Readme

Llama3.2-Unmasked

An “unmasked” version of Llama3.2.

What is unmasking?

Unmasking is a finetuning method similar to abliteration/jailbreaking that removes the original RLHF completely from a model instead of just removing refusals and then doing another finetuning step to make it conversational again.

Also…

I was bored, so I tested it’s political beliefs. Here’s it’s political compass: