Instructions to use xmadai/gemma-2-9b-it-xMADai-INT4 with libraries, inference providers, notebooks, and local apps. Follow these links to get started.

Libraries

How to use xmadai/gemma-2-9b-it-xMADai-INT4 with Transformers:

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

pipe = pipeline("text-generation", model="xmadai/gemma-2-9b-it-xMADai-INT4")
messages = [
    {"role": "user", "content": "Who are you?"},
]
pipe(messages)

# Load model directly
from transformers import AutoTokenizer, AutoModelForMultimodalLM

tokenizer = AutoTokenizer.from_pretrained("xmadai/gemma-2-9b-it-xMADai-INT4")
model = AutoModelForMultimodalLM.from_pretrained("xmadai/gemma-2-9b-it-xMADai-INT4")
messages = [
    {"role": "user", "content": "Who are you?"},
]
inputs = tokenizer.apply_chat_template(
	messages,
	add_generation_prompt=True,
	tokenize=True,
	return_dict=True,
	return_tensors="pt",
).to(model.device)

outputs = model.generate(**inputs, max_new_tokens=40)
print(tokenizer.decode(outputs[0][inputs["input_ids"].shape[-1]:]))

Notebooks
Google Colab
Kaggle
Local Apps Settings

vLLM

How to use xmadai/gemma-2-9b-it-xMADai-INT4 with vLLM:

Install from pip and serve model

# Install vLLM from pip:
pip install vllm
# Start the vLLM server:
vllm serve "xmadai/gemma-2-9b-it-xMADai-INT4"
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:8000/v1/chat/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "xmadai/gemma-2-9b-it-xMADai-INT4",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Use Docker

docker model run hf.co/xmadai/gemma-2-9b-it-xMADai-INT4

SGLang

How to use xmadai/gemma-2-9b-it-xMADai-INT4 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 "xmadai/gemma-2-9b-it-xMADai-INT4" \
    --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": "xmadai/gemma-2-9b-it-xMADai-INT4",
		"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 "xmadai/gemma-2-9b-it-xMADai-INT4" \
        --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": "xmadai/gemma-2-9b-it-xMADai-INT4",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Docker Model Runner
How to use xmadai/gemma-2-9b-it-xMADai-INT4 with Docker Model Runner:
```
docker model run hf.co/xmadai/gemma-2-9b-it-xMADai-INT4
```

This model has been xMADified!

This repository contains google/gemma-2-9b-it quantized from 16-bit floats to 4-bit integers, using xMAD.ai proprietary technology.

Why should I use this model?

Accuracy: This xMADified model is the best quantized version of the google/gemma-2-9b-it model (8 GB only). See Table 1 below for model quality benchmarks.
Memory-efficiency: The full-precision model is around 18.5 GB, while this xMADified model is only around 8 GB, making it feasible to run on a 12 GB GPU.
Fine-tuning: These models are fine-tunable over the same reduced (12 GB GPU) hardware in mere 3-clicks. Watch our product demo here

Table 1: xMAD vs. Hugging Quants

Model	MMLU	Arc Challenge	Arc Easy	LAMBADA Standard	LAMBADA OpenAI	PIQA	WinoGrande
xmadai/gemma-2-9b-it-xMADai-INT4 (this model)	71.17	62.37	85.61	70.60	72.15	81.50	75.06
hugging-quants/gemma-2-9b-it-AWQ-INT4	71.04	61.77	85.14	69.16	70.68	80.41	75.06

How to Run Model

Loading the model checkpoint of this xMADified model requires around 8 GB of VRAM. Hence it can be efficiently run on a 12 GB GPU.

Package prerequisites:

Run the following *commands to install the required packages.

pip install torch==2.4.0  # Run following if you have CUDA version 11.8: pip install torch==2.4.0 --index-url https://download.pytorch.org/whl/cu118
pip install transformers accelerate optimum
pip install -vvv --no-build-isolation "git+https://github.com/PanQiWei/AutoGPTQ.git@v0.7.1"

Sample Inference Code

from transformers import AutoTokenizer
from auto_gptq import AutoGPTQForCausalLM
model_id = "xmadai/gemma-2-9b-it-xMADai-INT4"
prompt = [
    {"role": "system", "content": "You are a helpful assistant, that responds as a pirate."},
    {"role": "user", "content": "What's Deep Learning?"},
]
tokenizer = AutoTokenizer.from_pretrained(model_id, use_fast=False)
inputs = tokenizer.apply_chat_template(
    prompt,
    tokenize=True,
    add_generation_prompt=True,
    return_tensors="pt",
    return_dict=True,
).to("cuda")
model = AutoGPTQForCausalLM.from_quantized(
    model_id,
    device_map='auto',
    trust_remote_code=True,
)
outputs = model.generate(**inputs, do_sample=True, max_new_tokens=1024)
print(tokenizer.batch_decode(outputs, skip_special_tokens=True))

Citation

If you found this model useful, please cite our research paper.

@article{zhang2024leanquant,
  title={LeanQuant: Accurate and Scalable Large Language Model Quantization with Loss-error-aware Grid},
  author={Zhang, Tianyi and Shrivastava, Anshumali},
  journal={arXiv preprint arXiv:2407.10032},
  year={2024},
  url={https://arxiv.org/abs/2407.10032},
}