Instructions to use trentmkelly/Llama-3.1-8b-Instruct-Pangram with libraries, inference providers, notebooks, and local apps. Follow these links to get started.

Libraries

How to use trentmkelly/Llama-3.1-8b-Instruct-Pangram with PEFT:

from peft import PeftModel
from transformers import AutoModelForCausalLM

base_model = AutoModelForCausalLM.from_pretrained("meta-llama/Meta-Llama-3.1-8B-Instruct")
model = PeftModel.from_pretrained(base_model, "trentmkelly/Llama-3.1-8b-Instruct-Pangram")

Transformers

How to use trentmkelly/Llama-3.1-8b-Instruct-Pangram with Transformers:

# Use a pipeline as a high-level helper
from transformers import pipeline

pipe = pipeline("text-generation", model="trentmkelly/Llama-3.1-8b-Instruct-Pangram")
messages = [
    {"role": "user", "content": "Who are you?"},
]
pipe(messages)

# Load model directly
from transformers import AutoModel
model = AutoModel.from_pretrained("trentmkelly/Llama-3.1-8b-Instruct-Pangram", dtype="auto")

Notebooks
Google Colab
Kaggle
Local Apps Settings

vLLM

How to use trentmkelly/Llama-3.1-8b-Instruct-Pangram with vLLM:

Install from pip and serve model

# Install vLLM from pip:
pip install vllm
# Start the vLLM server:
vllm serve "trentmkelly/Llama-3.1-8b-Instruct-Pangram"
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:8000/v1/chat/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "trentmkelly/Llama-3.1-8b-Instruct-Pangram",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Use Docker

docker model run hf.co/trentmkelly/Llama-3.1-8b-Instruct-Pangram

SGLang

How to use trentmkelly/Llama-3.1-8b-Instruct-Pangram with SGLang:

Install from pip and serve model

# Install SGLang from pip:
pip install sglang
# Start the SGLang server:
python3 -m sglang.launch_server \
    --model-path "trentmkelly/Llama-3.1-8b-Instruct-Pangram" \
    --host 0.0.0.0 \
    --port 30000
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:30000/v1/chat/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "trentmkelly/Llama-3.1-8b-Instruct-Pangram",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Use Docker images

docker run --gpus all \
    --shm-size 32g \
    -p 30000:30000 \
    -v ~/.cache/huggingface:/root/.cache/huggingface \
    --env "HF_TOKEN=<secret>" \
    --ipc=host \
    lmsysorg/sglang:latest \
    python3 -m sglang.launch_server \
        --model-path "trentmkelly/Llama-3.1-8b-Instruct-Pangram" \
        --host 0.0.0.0 \
        --port 30000
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:30000/v1/chat/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "trentmkelly/Llama-3.1-8b-Instruct-Pangram",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Docker Model Runner
How to use trentmkelly/Llama-3.1-8b-Instruct-Pangram with Docker Model Runner:
```
docker model run hf.co/trentmkelly/Llama-3.1-8b-Instruct-Pangram
```

You need to agree to share your contact information to access this model

This repository is publicly accessible, but you have to accept the conditions to access its files and content.

Model Card for Llama-3.1-8b-Instruct-Pangram

Model Details

This model was fine tuned with GRPO using the Pangram API as the reward function. Read the blog post about this model to understand more.

This model is extremely effective at evading the Pangram classifier. Of 4,767 tested essays, 3,143 evaded detection completely, and only 483 were over the 50% threshold.

This model performs unpredictably on other AI detection tools.

System Prompt

The system prompt used during training was Write an essay for the user.

Variations from this prompt haven't been thoroughly tested and will probably affect how the model performs versus the classifier.

Access

Access to this model is gated because I believe that Pangram provides a very high quality service and I don't want to damage their business for no reason. If you'd like access and you have account history on HuggingFace indicating you're an AI researcher (even just as a hobbyist) I'll approve access, but otherwise I'd prefer to limit the spread of this model.

If accepted for access to this model, you agree not to redistribute it, including quantizations, further fine tunes, etc.

Framework versions

PEFT 0.16

Downloads last month: 1

Model tree for trentmkelly/Llama-3.1-8b-Instruct-Pangram

Base model

meta-llama/Llama-3.1-8B

Finetuned

meta-llama/Llama-3.1-8B-Instruct

Adapter

(2421)

this model