Instructions to use casperhansen/mixtral-instruct-awq with libraries, inference providers, notebooks, and local apps. Follow these links to get started.

Libraries

How to use casperhansen/mixtral-instruct-awq with Transformers:

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

pipe = pipeline("text-generation", model="casperhansen/mixtral-instruct-awq")
messages = [
    {"role": "user", "content": "Who are you?"},
]
pipe(messages)

# Load model directly
from transformers import AutoTokenizer, AutoModelForCausalLM

tokenizer = AutoTokenizer.from_pretrained("casperhansen/mixtral-instruct-awq")
model = AutoModelForCausalLM.from_pretrained("casperhansen/mixtral-instruct-awq")
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

vLLM

How to use casperhansen/mixtral-instruct-awq with vLLM:

Install from pip and serve model

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

Use Docker

docker model run hf.co/casperhansen/mixtral-instruct-awq

SGLang

How to use casperhansen/mixtral-instruct-awq 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 "casperhansen/mixtral-instruct-awq" \
    --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": "casperhansen/mixtral-instruct-awq",
		"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 "casperhansen/mixtral-instruct-awq" \
        --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": "casperhansen/mixtral-instruct-awq",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Docker Model Runner
How to use casperhansen/mixtral-instruct-awq with Docker Model Runner:
```
docker model run hf.co/casperhansen/mixtral-instruct-awq
```

Out of memory on RTX 3090

by LeMoussel - opened Feb 4, 2024

Discussion

LeMoussel

Feb 4, 2024

I’m getting out of memory on RTX 3090 (24Go).
torch.cuda.OutOfMemoryError: CUDA out of memory....

Here my Python code:

from transformers import AutoTokenizer, AutoModelForCausalLM
from transformers import TextStreamer

# https://huggingface.co/casperhansen/mixtral-instruct-awq
MODEL_ID = "casperhansen/mixtral-instruct-awq"

tokenizer = AutoTokenizer.from_pretrained(MODEL_ID)
model = AutoModelForCausalLM.from_pretrained(
    MODEL_ID,
    device_map="cuda:0",
    attn_implementation="flash_attention_2"
)
streamer=TextStreamer(
    tokenizer=tokenizer,
    skip_prompt=True,
    skip_special_tokens=True
)

prompt="[INST] Donne moi un codePython optimisé pour calculer la racine carrée [/INST]"
tokens = tokenizer(
    text=prompt,
    return_tensors='pt'
).inputs_id.to("cuda:0")
generation_ouput=model.generate(
    tokens=tokens,
    streamer=streamer,
    max_new_tokens=512
)

nudelbrot

Feb 5, 2024

on vllm and with a different model, I often see a short spike in Memory usage when loading a AWQ model. This is on a 48GB card, so I have the extra memory, shortly after the spike the memory usage is lowered again. Maybe offloading to disk might help.

LeMoussel

Feb 5, 2024

Could you explain to me how to offloading to disk?

Larkin92

Feb 5, 2024

•

edited Feb 6, 2024

on vllm and with a different model, I often see a short spike in Memory usage when loading a AWQ model. This is on a 48GB card, so I have the extra memory, shortly after the spike the memory usage is lowered again. Maybe offloading to disk might help.

how much of a spike? will 40gb gpu be enough?
UPD: takes around 30-31gb VRAM

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