Instructions to use Peeepy/llama-30b-oasst-4bit-128g with libraries, inference providers, notebooks, and local apps. Follow these links to get started.

Libraries

How to use Peeepy/llama-30b-oasst-4bit-128g with Transformers:

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

pipe = pipeline("text-generation", model="Peeepy/llama-30b-oasst-4bit-128g")

# pip install -U transformers accelerate
# Load model directly
from transformers import AutoTokenizer, AutoModelForCausalLM

tokenizer = AutoTokenizer.from_pretrained("Peeepy/llama-30b-oasst-4bit-128g")
model = AutoModelForCausalLM.from_pretrained("Peeepy/llama-30b-oasst-4bit-128g", device_map="auto")

Notebooks
Google Colab
Kaggle
Local Apps Settings

vLLM

How to use Peeepy/llama-30b-oasst-4bit-128g with vLLM:

Install from pip and serve model

# Install vLLM from pip:
pip install vllm
# Start the vLLM server:
vllm serve "Peeepy/llama-30b-oasst-4bit-128g"
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:8000/v1/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "Peeepy/llama-30b-oasst-4bit-128g",
		"prompt": "Once upon a time,",
		"max_tokens": 512,
		"temperature": 0.5
	}'

Use Docker

docker model run hf.co/Peeepy/llama-30b-oasst-4bit-128g

SGLang

How to use Peeepy/llama-30b-oasst-4bit-128g 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 "Peeepy/llama-30b-oasst-4bit-128g" \
    --host 0.0.0.0 \
    --port 30000
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:30000/v1/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "Peeepy/llama-30b-oasst-4bit-128g",
		"prompt": "Once upon a time,",
		"max_tokens": 512,
		"temperature": 0.5
	}'

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 "Peeepy/llama-30b-oasst-4bit-128g" \
        --host 0.0.0.0 \
        --port 30000
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:30000/v1/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "Peeepy/llama-30b-oasst-4bit-128g",
		"prompt": "Once upon a time,",
		"max_tokens": 512,
		"temperature": 0.5
	}'

Docker Model Runner
How to use Peeepy/llama-30b-oasst-4bit-128g with Docker Model Runner:
```
docker model run hf.co/Peeepy/llama-30b-oasst-4bit-128g
```

Minimum system requirements?

by kusoge - opened Apr 21, 2023

Discussion

kusoge

Apr 21, 2023

•

edited Apr 21, 2023

Any idea how much RAM is needed to preload the model and VRAM to actually use it?
The 4bit 13B models I tried fills my 16GB RAM and during usage my VRAM completely fills my 12GB VRAM.
Will 32GB RAM and 18GB VRAM be enough to load and use this? Thanks

Peeepy

Owner Apr 23, 2023

Hey! Sorry. I didn't notice this until just now.
For a better overview of the exact requirements see: https://www.reddit.com/r/LocalLLaMA/comments/11o6o3f/how_to_install_llama_8bit_and_4bit/
But TL;DR: No. Minimum VRAM needed is 20GB, 23GB with full context. 24GB RAM is needed, so on the RAM side you'd be set.

Upload images, audio, and videos by dragging in the text input, pasting, or clicking here.

Tap or paste here to upload images

· Sign up or log in to comment