Instructions to use HuggingFaceH4/zephyr-7b-alpha with libraries, inference providers, notebooks, and local apps. Follow these links to get started.

Libraries

How to use HuggingFaceH4/zephyr-7b-alpha with Transformers:

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

pipe = pipeline("text-generation", model="HuggingFaceH4/zephyr-7b-alpha")
messages = [
    {"role": "user", "content": "Who are you?"},
]
pipe(messages)

# Load model directly
from transformers import AutoTokenizer, AutoModelForCausalLM

tokenizer = AutoTokenizer.from_pretrained("HuggingFaceH4/zephyr-7b-alpha")
model = AutoModelForCausalLM.from_pretrained("HuggingFaceH4/zephyr-7b-alpha")
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]:]))

Inference
Notebooks
Google Colab
Kaggle
Local Apps

vLLM

How to use HuggingFaceH4/zephyr-7b-alpha with vLLM:

Install from pip and serve model

# Install vLLM from pip:
pip install vllm
# Start the vLLM server:
vllm serve "HuggingFaceH4/zephyr-7b-alpha"
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:8000/v1/chat/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "HuggingFaceH4/zephyr-7b-alpha",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Use Docker

docker model run hf.co/HuggingFaceH4/zephyr-7b-alpha

SGLang

How to use HuggingFaceH4/zephyr-7b-alpha 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 "HuggingFaceH4/zephyr-7b-alpha" \
    --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": "HuggingFaceH4/zephyr-7b-alpha",
		"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 "HuggingFaceH4/zephyr-7b-alpha" \
        --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": "HuggingFaceH4/zephyr-7b-alpha",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Docker Model Runner
How to use HuggingFaceH4/zephyr-7b-alpha with Docker Model Runner:
```
docker model run hf.co/HuggingFaceH4/zephyr-7b-alpha
```

High GPU RAM usage

#30

by Alealejandrooo - opened Nov 1, 2023

Discussion

Alealejandrooo

Nov 1, 2023

Has anyone experienced a high GPU RAM usage with this model?
When downloaded in 8bit, the model only takes about 7GB on the GPU, but at inference time it shoots up to 28GB.
Anyone else / suggestions on how to fix this?

eslXist

Nov 1, 2023

Facing the same issue +1

eslXist

Nov 1, 2023

What configuration do you use when loading the model? default or overridden?

Alealejandrooo

Nov 1, 2023

What configuration do you use when loading the model? default or overridden?

I load it in int8 with a device_map = auto.

eslXist

Nov 1, 2023

What about the FP (dtype)?

Alealejandrooo

Nov 1, 2023

•

edited Nov 1, 2023

What about the FP (dtype)?

torch_dtype=torch.float32

Is that the issue?

all in all:

device_map="auto",
torch_dtype=torch.float32,
load_in_8bit= True

eslXist

Nov 1, 2023

The default model’s configuration uses bfloat16 floating point format "torch_dtype": "bfloat16". Switching to half point precision will definitely help reducing memory footprint and increase performance.

djstrong

Jan 19, 2024

•

edited Jan 19, 2024

I experience the same problem with generation using cache: bfloat16 28GB, float32 39GB. For llama it is 20GB for bfloat16.
This model has bigger context.

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