huihui_ai/granite3.1-dense-abliterated

huihui_ai/ granite3.1-dense-abliterated

1,050 Downloads Updated 1 year ago

The IBM Granite 2B and 8B models are text-only dense LLMs trained on over 12 trillion tokens of data, demonstrated significant improvements over their predecessors in performance and speed in IBM’s initial testing.

tools 2b 8b

ollama run huihui_ai/granite3.1-dense-abliterated

curl http://localhost:11434/api/chat \
  -d '{
    "model": "huihui_ai/granite3.1-dense-abliterated",
    "messages": [{"role": "user", "content": "Hello!"}]
  }'

from ollama import chat

response = chat(
    model='huihui_ai/granite3.1-dense-abliterated',
    messages=[{'role': 'user', 'content': 'Hello!'}],
)
print(response.message.content)

import ollama from 'ollama'

const response = await ollama.chat({
  model: 'huihui_ai/granite3.1-dense-abliterated',
  messages: [{role: 'user', content: 'Hello!'}],
})
console.log(response.message.content)

Applications

Claude Code ollama launch claude --model huihui_ai/granite3.1-dense-abliterated

Codex ollama launch codex --model huihui_ai/granite3.1-dense-abliterated

OpenCode ollama launch opencode --model huihui_ai/granite3.1-dense-abliterated

OpenClaw ollama launch openclaw --model huihui_ai/granite3.1-dense-abliterated

Models

View all →

Name

11 models

Size

Context

Input

granite3.1-dense-abliterated:latest

4.9GB · 128K context window · Text · 1 year ago

granite3.1-dense-abliterated:latest

4.9GB

128K

Text

granite3.1-dense-abliterated:2b

1.5GB · 128K context window · Text · 1 year ago

granite3.1-dense-abliterated:2b

1.5GB

128K

Text

granite3.1-dense-abliterated:8b

latest

4.9GB · 128K context window · Text · 1 year ago

granite3.1-dense-abliterated:8b latest

4.9GB

128K

Text

Readme

This is an uncensored version of ibm-granite/granite-31-language-models created with abliteration (see remove-refusals-with-transformers to know more about it).
This is a crude, proof-of-concept implementation to remove refusals from an LLM model without using TransformerLens.

References

HuggingFace