Instructions to use mlabonne/Meta-Llama-3-120B-Instruct with libraries, inference providers, notebooks, and local apps. Follow these links to get started.

Libraries

How to use mlabonne/Meta-Llama-3-120B-Instruct with Transformers:

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

pipe = pipeline("text-generation", model="mlabonne/Meta-Llama-3-120B-Instruct")
messages = [
    {"role": "user", "content": "Who are you?"},
]
pipe(messages)

# Load model directly
from transformers import AutoTokenizer, AutoModelForCausalLM

tokenizer = AutoTokenizer.from_pretrained("mlabonne/Meta-Llama-3-120B-Instruct")
model = AutoModelForCausalLM.from_pretrained("mlabonne/Meta-Llama-3-120B-Instruct", device_map="auto")
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 mlabonne/Meta-Llama-3-120B-Instruct with vLLM:

Install from pip and serve model

# Install vLLM from pip:
pip install vllm
# Start the vLLM server:
vllm serve "mlabonne/Meta-Llama-3-120B-Instruct"
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:8000/v1/chat/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "mlabonne/Meta-Llama-3-120B-Instruct",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Use Docker

docker model run hf.co/mlabonne/Meta-Llama-3-120B-Instruct

SGLang

How to use mlabonne/Meta-Llama-3-120B-Instruct 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 "mlabonne/Meta-Llama-3-120B-Instruct" \
    --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": "mlabonne/Meta-Llama-3-120B-Instruct",
		"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 "mlabonne/Meta-Llama-3-120B-Instruct" \
        --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": "mlabonne/Meta-Llama-3-120B-Instruct",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Docker Model Runner
How to use mlabonne/Meta-Llama-3-120B-Instruct with Docker Model Runner:
```
docker model run hf.co/mlabonne/Meta-Llama-3-120B-Instruct
```

fine-tuning is needed after self-merging?

by oodgnas - opened May 8, 2024

Discussion

oodgnas

May 8, 2024

Hi, thank you for the excellent work @mlabonne !

I want to ask whether this model requires fine-tuning steps or not, after its self-merging.
If there is no fine-tuning, it would be really fascinating :)

Thanks,
Sangdoo

mlabonne

Owner May 8, 2024

Thanks @oodgnas ! This model hasn't been fine-tuned but this would probably be better (see https://arxiv.org/abs/2312.15166). It looks like small source models really require it while big models can do without but they're kind of insane.

ehartford

Jul 10, 2024

This specific merge ended up exhibiting "sentience" like behaviors, as well as a bit of schizophrenic behaviors.
I imagine that a round of light pretraining and instruct tuning might iron these things out.

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