Instructions to use Dracones/Midnight-Miqu-70B-v1.5_exl2_3.0bpw with libraries, inference providers, notebooks, and local apps. Follow these links to get started.

Libraries

How to use Dracones/Midnight-Miqu-70B-v1.5_exl2_3.0bpw with Transformers:

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

pipe = pipeline("text-generation", model="Dracones/Midnight-Miqu-70B-v1.5_exl2_3.0bpw")
messages = [
    {"role": "user", "content": "Who are you?"},
]
pipe(messages)

# Load model directly
from transformers import AutoTokenizer, AutoModelForMultimodalLM

tokenizer = AutoTokenizer.from_pretrained("Dracones/Midnight-Miqu-70B-v1.5_exl2_3.0bpw")
model = AutoModelForMultimodalLM.from_pretrained("Dracones/Midnight-Miqu-70B-v1.5_exl2_3.0bpw")
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 Dracones/Midnight-Miqu-70B-v1.5_exl2_3.0bpw with vLLM:

Install from pip and serve model

# Install vLLM from pip:
pip install vllm
# Start the vLLM server:
vllm serve "Dracones/Midnight-Miqu-70B-v1.5_exl2_3.0bpw"
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:8000/v1/chat/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "Dracones/Midnight-Miqu-70B-v1.5_exl2_3.0bpw",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Use Docker

docker model run hf.co/Dracones/Midnight-Miqu-70B-v1.5_exl2_3.0bpw

SGLang

How to use Dracones/Midnight-Miqu-70B-v1.5_exl2_3.0bpw 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 "Dracones/Midnight-Miqu-70B-v1.5_exl2_3.0bpw" \
    --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": "Dracones/Midnight-Miqu-70B-v1.5_exl2_3.0bpw",
		"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 "Dracones/Midnight-Miqu-70B-v1.5_exl2_3.0bpw" \
        --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": "Dracones/Midnight-Miqu-70B-v1.5_exl2_3.0bpw",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Docker Model Runner
How to use Dracones/Midnight-Miqu-70B-v1.5_exl2_3.0bpw with Docker Model Runner:
```
docker model run hf.co/Dracones/Midnight-Miqu-70B-v1.5_exl2_3.0bpw
```

2.75 bpw high EQ bench

by koesn - opened Apr 17, 2024

Discussion

koesn

Apr 17, 2024

How 2.75 bpw has high eq bench, even still lower ppl?

Dracones

Owner Apr 17, 2024

I'd probably put that down to a level of error in the benchmark itself. EQ Bench works by asking the LLM to give a score from 1-10 on certain emotions against a brief conversation. I don't know if LLM's are good at a "score this between 1 and 10" in testing. I did find it useful when running a lot of them and seeing patterns across prompt types: Midnight Miqu being good at a lot of difference prompts, Cohere Command models working better with Command-R prompts. But I'd probably trust perplexity over EQ Bench.

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