Instructions to use ISTA-DASLab/Llama-2-13b-AQLM-PV-2Bit-1x16-hf with libraries, inference providers, notebooks, and local apps. Follow these links to get started.

Libraries

How to use ISTA-DASLab/Llama-2-13b-AQLM-PV-2Bit-1x16-hf with Transformers:

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

pipe = pipeline("text-generation", model="ISTA-DASLab/Llama-2-13b-AQLM-PV-2Bit-1x16-hf")
messages = [
    {"role": "user", "content": "Who are you?"},
]
pipe(messages)

# Load model directly
from transformers import AutoTokenizer, AutoModelForCausalLM

tokenizer = AutoTokenizer.from_pretrained("ISTA-DASLab/Llama-2-13b-AQLM-PV-2Bit-1x16-hf")
model = AutoModelForCausalLM.from_pretrained("ISTA-DASLab/Llama-2-13b-AQLM-PV-2Bit-1x16-hf")

Notebooks
Google Colab
Kaggle
Local Apps

vLLM

How to use ISTA-DASLab/Llama-2-13b-AQLM-PV-2Bit-1x16-hf with vLLM:

Install from pip and serve model

# Install vLLM from pip:
pip install vllm
# Start the vLLM server:
vllm serve "ISTA-DASLab/Llama-2-13b-AQLM-PV-2Bit-1x16-hf"
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:8000/v1/chat/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "ISTA-DASLab/Llama-2-13b-AQLM-PV-2Bit-1x16-hf",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Use Docker

docker model run hf.co/ISTA-DASLab/Llama-2-13b-AQLM-PV-2Bit-1x16-hf

SGLang

How to use ISTA-DASLab/Llama-2-13b-AQLM-PV-2Bit-1x16-hf 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 "ISTA-DASLab/Llama-2-13b-AQLM-PV-2Bit-1x16-hf" \
    --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": "ISTA-DASLab/Llama-2-13b-AQLM-PV-2Bit-1x16-hf",
		"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 "ISTA-DASLab/Llama-2-13b-AQLM-PV-2Bit-1x16-hf" \
        --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": "ISTA-DASLab/Llama-2-13b-AQLM-PV-2Bit-1x16-hf",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Docker Model Runner
How to use ISTA-DASLab/Llama-2-13b-AQLM-PV-2Bit-1x16-hf with Docker Model Runner:
```
docker model run hf.co/ISTA-DASLab/Llama-2-13b-AQLM-PV-2Bit-1x16-hf
```

An official quantization of meta-llama/Llama-2-13b using PV-Tuning on top of AQLM.

For this quantization, we used 1 codebook of 16 bits for groups of 8 weights.

Model	AQLM scheme	WikiText 2 PPL	Model size, Gb	Hub link
Llama-2-7b	1x16	5.68	2.4	Link
Llama-2-7b	2x8	5.90	2.2	Link
Llama-2-7b	1x16g16	9.21	1.7	Link
Llama-2-13b (this)	1x16	5.05	4.1	Link
Llama-2-70b	1x16	3.78	18.8	Link

The 1x16g16 (1-bit) models are on the way, as soon as we update the inference lib with their respective kernels.

To learn more about the inference, as well as the information on how to quantize models yourself, please refer to the official GitHub repo. The original code for PV-Tuning can be found in the AQLM@pv-tuning branch.

Downloads last month: 6

Safetensors

Model size

2B params

Tensor type

F16

I16

Collection including ISTA-DASLab/Llama-2-13b-AQLM-PV-2Bit-1x16-hf

AQLM+PV

Collection

Official AQLM quantizations for "PV-Tuning: Beyond Straight-Through Estimation for Extreme LLM Compression": https://arxiv.org/abs/2405.14852 • 26 items • Updated Feb 18 • 22

Papers for ISTA-DASLab/Llama-2-13b-AQLM-PV-2Bit-1x16-hf

PV-Tuning: Beyond Straight-Through Estimation for Extreme LLM Compression

Paper • 2405.14852 • Published May 23, 2024 • 2

Extreme Compression of Large Language Models via Additive Quantization

Paper • 2401.06118 • Published Jan 11, 2024 • 14