Instructions to use mobiuslabsgmbh/Mixtral-8x7B-Instruct-v0.1-hf-4bit_g64-HQQ with libraries, inference providers, notebooks, and local apps. Follow these links to get started.

Libraries

How to use mobiuslabsgmbh/Mixtral-8x7B-Instruct-v0.1-hf-4bit_g64-HQQ with Transformers:

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

pipe = pipeline("text-generation", model="mobiuslabsgmbh/Mixtral-8x7B-Instruct-v0.1-hf-4bit_g64-HQQ")
messages = [
    {"role": "user", "content": "Who are you?"},
]
pipe(messages)

# Load model directly
from transformers import AutoTokenizer, AutoModelForCausalLM

tokenizer = AutoTokenizer.from_pretrained("mobiuslabsgmbh/Mixtral-8x7B-Instruct-v0.1-hf-4bit_g64-HQQ")
model = AutoModelForCausalLM.from_pretrained("mobiuslabsgmbh/Mixtral-8x7B-Instruct-v0.1-hf-4bit_g64-HQQ")
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 mobiuslabsgmbh/Mixtral-8x7B-Instruct-v0.1-hf-4bit_g64-HQQ with vLLM:

Install from pip and serve model

# Install vLLM from pip:
pip install vllm
# Start the vLLM server:
vllm serve "mobiuslabsgmbh/Mixtral-8x7B-Instruct-v0.1-hf-4bit_g64-HQQ"
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:8000/v1/chat/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "mobiuslabsgmbh/Mixtral-8x7B-Instruct-v0.1-hf-4bit_g64-HQQ",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Use Docker

docker model run hf.co/mobiuslabsgmbh/Mixtral-8x7B-Instruct-v0.1-hf-4bit_g64-HQQ

SGLang

How to use mobiuslabsgmbh/Mixtral-8x7B-Instruct-v0.1-hf-4bit_g64-HQQ 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 "mobiuslabsgmbh/Mixtral-8x7B-Instruct-v0.1-hf-4bit_g64-HQQ" \
    --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": "mobiuslabsgmbh/Mixtral-8x7B-Instruct-v0.1-hf-4bit_g64-HQQ",
		"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 "mobiuslabsgmbh/Mixtral-8x7B-Instruct-v0.1-hf-4bit_g64-HQQ" \
        --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": "mobiuslabsgmbh/Mixtral-8x7B-Instruct-v0.1-hf-4bit_g64-HQQ",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Docker Model Runner
How to use mobiuslabsgmbh/Mixtral-8x7B-Instruct-v0.1-hf-4bit_g64-HQQ with Docker Model Runner:
```
docker model run hf.co/mobiuslabsgmbh/Mixtral-8x7B-Instruct-v0.1-hf-4bit_g64-HQQ
```

only use one gpu?

by jgbrblmd - opened Dec 15, 2023

Discussion

jgbrblmd

Dec 15, 2023

$ env | grep VIS
CUDA_VISIBLE_DEVICES=1,2

$ python hqq_test.py
hqq_aten package not installed. HQQBackend.ATEN backend will not work unless you install the hqq_aten lib in hqq/kernels.
Failed to load the weights CUDA out of memory. Tried to allocate 28.00 MiB. GPU 0 has a total capacty of 21.67 GiB of which 22.75 MiB is free. Including non-PyTorch memory, this process has 21.64 GiB memory in use. Of the allocated memory 21.40 GiB is allocated by PyTorch, and 92.97 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting max_split_size_mb to avoid fragmentation. See documentation for Memory Management and PYTORCH_CUDA_ALLOC_CONF
Traceback (most recent call last):
File "/nvme/opt/LLM/hqq_test.py", line 7, in
model = HQQModelForCausalLM.from_quantized(model_id)
File "/nvme/opt/venv/hqq/lib/python3.10/site-packages/hqq/engine/base.py", line 71, in from_quantized
cls._make_quantizable(model, quantized=True)
File "/nvme/opt/venv/hqq/lib/python3.10/site-packages/hqq/engine/hf.py", line 29, in _make_quantizable
model.hqq_quantized = quantized
AttributeError: 'NoneType' object has no attribute 'hqq_quantized'

mobicham

Mobius Labs GmbH org Dec 15, 2023

Indeed, only single GPU for now. You want to load half of the model on gpu1 and the other half on gpu2?

jgbrblmd

Dec 15, 2023

Indeed, only single GPU for now. You want to load half of the model on gpu1 and the other half on gpu2?

yes, my gpus has 22G vram each.

Thank you.

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