GitHub radar
AutoTrust-JEV-27B: probabilities for answer options, not text
AutoTrust-JEV-27B v1.1 is a decision model: give it a question with a few answer options and it scores how likely each one is, instead of writing a reply. That lets you plug fast, cheap classification into your own project.
This isn't a chatbot: it takes a question plus a list of answer options and returns a probability for each one — the creators call this a 'decision model'.
When it helps
Useful when you need to quickly and cheaply classify requests or pick which tool to call inside your own agent or script. Skip it if you want a regular conversational AI or text generation.
Pros
- Cheaper than a chatbot for simple category decisions
- Beats earlier JEV-27B and Jev 1.14 models at most classification tasks
- Best result is picking which tool to call for automation
- Open Apache 2.0 license, free to use
Cons
- Not for chat or text generation — only option probabilities
- Harder to install: needs a vLLM server, not a simple GGUF file
- Loses to the older Jev 1.14 on harder reasoning and knowledge tasks
How to set it up — step by step
- 1Check you have a Mac with 32GB+ RAM or a GPU with 24GB+ memory — it won't run without that.
- 2Open your coding agent (Claude Code, Codex) in your project folder.
- 3Send the agent the text from the block below.
- 4Let the agent install and serve the model through a vLLM server.
- 5Check the result: the agent should show probabilities for each of the three answer options.
Text for your agent
Copy this and send it to your agent — Claude Code, Codex, any of them:
Set up the AutoTrust-JEV-27B-v1.1 model for me: https://huggingface.co/autotrust/AutoTrust-JEV-27B-v1.1 — run it through vLLM on my machine, feed it a sample customer message with three answer options, and show me the probabilities it assigns.
You'll need a Mac with 32 GB of memory or more, or a GPU with 24 GB of VRAM — it ships in a vLLM format, so setting it up takes more work than a typical Ollama-ready GGUF model.
Open on Hugging Face▌ More finds