Instructions to use sophosympatheia/Midnight-Miqu-70B-v1.0 with libraries, inference providers, notebooks, and local apps. Follow these links to get started.

Libraries

How to use sophosympatheia/Midnight-Miqu-70B-v1.0 with Transformers:

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

pipe = pipeline("text-generation", model="sophosympatheia/Midnight-Miqu-70B-v1.0")
messages = [
    {"role": "user", "content": "Who are you?"},
]
pipe(messages)

# Load model directly
from transformers import AutoTokenizer, AutoModelForMultimodalLM

tokenizer = AutoTokenizer.from_pretrained("sophosympatheia/Midnight-Miqu-70B-v1.0")
model = AutoModelForMultimodalLM.from_pretrained("sophosympatheia/Midnight-Miqu-70B-v1.0")
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 sophosympatheia/Midnight-Miqu-70B-v1.0 with vLLM:

Install from pip and serve model

# Install vLLM from pip:
pip install vllm
# Start the vLLM server:
vllm serve "sophosympatheia/Midnight-Miqu-70B-v1.0"
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:8000/v1/chat/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "sophosympatheia/Midnight-Miqu-70B-v1.0",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Use Docker

docker model run hf.co/sophosympatheia/Midnight-Miqu-70B-v1.0

SGLang

How to use sophosympatheia/Midnight-Miqu-70B-v1.0 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 "sophosympatheia/Midnight-Miqu-70B-v1.0" \
    --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": "sophosympatheia/Midnight-Miqu-70B-v1.0",
		"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 "sophosympatheia/Midnight-Miqu-70B-v1.0" \
        --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": "sophosympatheia/Midnight-Miqu-70B-v1.0",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Docker Model Runner
How to use sophosympatheia/Midnight-Miqu-70B-v1.0 with Docker Model Runner:
```
docker model run hf.co/sophosympatheia/Midnight-Miqu-70B-v1.0
```

Question about prompting and System prompt in Vicuna format

#10

by houmie - opened Apr 8, 2024

Discussion

houmie

Apr 8, 2024

To get the best results it's recommended to use the same prompting as the model was trained to. I'm not sure which one it is. But the Model card is saying that Vicuna is recommended.

USER:
{prompt}
ASSISTANT:

However there is no example about how to use the System prompt in Vicuna format. May I ask you for an example please?
Thank you

sophosympatheia

Owner Apr 8, 2024

There are so many instruction templates represented in these merges that you can get good results with several different formats. I think Vicuna produces the most creative results, but you would likely do well with Mistral's format too. The system prompt with Vicuna either has no tag or you can tag it with SYSTEM. I like adding newlines after the tags but the original format I believe does not include them.

My Vicuna preference
SYSTEM:
{system prompt}
USER:
{user input}
ASSISTANT:
{AI output}

Mistral
[INST] {prompt or user input} [/INST]
{AI output}
[INST] {next user prompt or input} [/INST]

Dracones

Apr 13, 2024

I added EQ Bench results with Alpaca, ChatML, Mistral, Vicuna-v1.1 and Vicuna-v0 templates. Vicuna does seem to slightly edge out the other instruction templates: EQ Bench Results

But I feel like this model over all is just very flexible with how you approach it.

houmie

Apr 13, 2024

Thank you both for sharing this. I got Vicuna working now and am happy with it.

I agree that I had a similar experience with ChatML, even though it's my favourite due to its closing tags, I found that ChatML produced worse results with Miqu.

sophosympatheia

Owner Apr 13, 2024

Thanks, Dracones. I appreciate the documentation of your results.

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