Instructions to use jondurbin/airoboros-l2-70b-gpt4-1.4.1 with libraries, inference providers, notebooks, and local apps. Follow these links to get started.

Libraries

How to use jondurbin/airoboros-l2-70b-gpt4-1.4.1 with Transformers:

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

pipe = pipeline("text-generation", model="jondurbin/airoboros-l2-70b-gpt4-1.4.1", device_map="auto")

# Load model directly
from transformers import AutoTokenizer, AutoModelForCausalLM

tokenizer = AutoTokenizer.from_pretrained("jondurbin/airoboros-l2-70b-gpt4-1.4.1")
model = AutoModelForCausalLM.from_pretrained("jondurbin/airoboros-l2-70b-gpt4-1.4.1", device_map="auto")

Notebooks
Google Colab
Kaggle
Local Apps Settings

vLLM

How to use jondurbin/airoboros-l2-70b-gpt4-1.4.1 with vLLM:

Install from pip and serve model

# Install vLLM from pip:
pip install vllm
# Start the vLLM server:
vllm serve "jondurbin/airoboros-l2-70b-gpt4-1.4.1"
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:8000/v1/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "jondurbin/airoboros-l2-70b-gpt4-1.4.1",
		"prompt": "Once upon a time,",
		"max_tokens": 512,
		"temperature": 0.5
	}'

Use Docker

docker model run hf.co/jondurbin/airoboros-l2-70b-gpt4-1.4.1

SGLang

How to use jondurbin/airoboros-l2-70b-gpt4-1.4.1 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 "jondurbin/airoboros-l2-70b-gpt4-1.4.1" \
    --host 0.0.0.0 \
    --port 30000
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:30000/v1/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "jondurbin/airoboros-l2-70b-gpt4-1.4.1",
		"prompt": "Once upon a time,",
		"max_tokens": 512,
		"temperature": 0.5
	}'

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 "jondurbin/airoboros-l2-70b-gpt4-1.4.1" \
        --host 0.0.0.0 \
        --port 30000
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:30000/v1/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "jondurbin/airoboros-l2-70b-gpt4-1.4.1",
		"prompt": "Once upon a time,",
		"max_tokens": 512,
		"temperature": 0.5
	}'

Docker Model Runner
How to use jondurbin/airoboros-l2-70b-gpt4-1.4.1 with Docker Model Runner:
```
docker model run hf.co/jondurbin/airoboros-l2-70b-gpt4-1.4.1
```

how set hardware and software config for training?

by lIlBrother - opened Jul 26, 2023

Discussion

lIlBrother

Jul 26, 2023

first of all, thx for model upload.

by the way, i want to further learning on this model, and i have A100 80GB * 8EA.
i'm using deepspeed zero3 but i can not training 😥 (gpu or cpu resource is insufficient)

please little tip for me?🤗

jondurbin

Owner Jul 27, 2023

I used qlora, so I only needed one A100 80GB. Here's exactly what I did: https://gist.github.com/jondurbin/87fc040b92a3073125ed516b04bc6e19

When I do full fine-tunes (13b/7b), I typically do something like this:
https://gist.github.com/jondurbin/7183e6edcc5cb57d5f544614d0ce0503

I have not tried a full fine tune of 70b yet, but will be taking a stab at it soon. If I have success, I will post a gist.

pankajmathur

Jul 27, 2023

Thanks Jon, I am a fan of your work. I truly appreciate all the hard work you put into this and then sharing it with community. I know it a lot lot do especially as a small or 1 person team :) but it's very rewarding too, so yeah please keep up the good work and congrats on reaching #2 on Open LLM Leader board.

lIlBrother

Aug 1, 2023

@jondurbin hi i successed full finetune of llama-2-70b.
it is so hard to set hardware.

i used gcp a100 80GB 8EA
and use deepspeed zero-3 and optimizer offload, not offload for parameter.

it can be done very close to explosive memory
hardwork!

lIlBrother

Aug 1, 2023

ah, and it is the best important thing. i used adafactor optimizer!

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