Instructions to use mit-han-lab/opt-13b-smoothquant with libraries, inference providers, notebooks, and local apps. Follow these links to get started.

Libraries

How to use mit-han-lab/opt-13b-smoothquant with Transformers:

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

pipe = pipeline("text-generation", model="mit-han-lab/opt-13b-smoothquant")

# Load model directly
from transformers import AutoTokenizer, AutoModelForCausalLM

tokenizer = AutoTokenizer.from_pretrained("mit-han-lab/opt-13b-smoothquant")
model = AutoModelForCausalLM.from_pretrained("mit-han-lab/opt-13b-smoothquant")

Notebooks
Google Colab
Kaggle
Local Apps Settings

vLLM

How to use mit-han-lab/opt-13b-smoothquant with vLLM:

Install from pip and serve model

# Install vLLM from pip:
pip install vllm
# Start the vLLM server:
vllm serve "mit-han-lab/opt-13b-smoothquant"
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:8000/v1/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "mit-han-lab/opt-13b-smoothquant",
		"prompt": "Once upon a time,",
		"max_tokens": 512,
		"temperature": 0.5
	}'

Use Docker

docker model run hf.co/mit-han-lab/opt-13b-smoothquant

SGLang

How to use mit-han-lab/opt-13b-smoothquant 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 "mit-han-lab/opt-13b-smoothquant" \
    --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": "mit-han-lab/opt-13b-smoothquant",
		"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 "mit-han-lab/opt-13b-smoothquant" \
        --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": "mit-han-lab/opt-13b-smoothquant",
		"prompt": "Once upon a time,",
		"max_tokens": 512,
		"temperature": 0.5
	}'

Docker Model Runner
How to use mit-han-lab/opt-13b-smoothquant with Docker Model Runner:
```
docker model run hf.co/mit-han-lab/opt-13b-smoothquant
```

how to load and use model?

by Q4234 - opened Dec 9, 2022

Discussion

Q4234

Dec 9, 2022

I tried

from transformers import AutoModelForCausalLM, AutoTokenizer
# from https://huggingface.co/facebook/opt-30b

modelName = "mit-han-lab/opt-13b-smoothquant" # 8-bit quantized model

model = AutoModelForCausalLM.from_pretrained(modelName, torch_dtype=torch.int8).cuda()
tokenizer = AutoTokenizer.from_pretrained(modelName, use_fast=False)

but this doesn't work...

Guangxuan-Xiao

MIT HAN Lab org Dec 10, 2022

Hi, please refer to https://github.com/mit-han-lab/smoothquant#smoothquant-int8-inference-for-pytorch to see how to use those models. We haven't integrated our INT8 kernels into huggingface.

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