Instructions to use AzureBlack/opus-v0-70b-exl2 with libraries, inference providers, notebooks, and local apps. Follow these links to get started.

Libraries

How to use AzureBlack/opus-v0-70b-exl2 with Transformers:

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

pipe = pipeline("text-generation", model="AzureBlack/opus-v0-70b-exl2")

# Load model directly
from transformers import AutoTokenizer, AutoModelForCausalLM

tokenizer = AutoTokenizer.from_pretrained("AzureBlack/opus-v0-70b-exl2")
model = AutoModelForCausalLM.from_pretrained("AzureBlack/opus-v0-70b-exl2")

Notebooks
Google Colab
Kaggle
Local Apps

vLLM

How to use AzureBlack/opus-v0-70b-exl2 with vLLM:

Install from pip and serve model

# Install vLLM from pip:
pip install vllm
# Start the vLLM server:
vllm serve "AzureBlack/opus-v0-70b-exl2"
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:8000/v1/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "AzureBlack/opus-v0-70b-exl2",
		"prompt": "Once upon a time,",
		"max_tokens": 512,
		"temperature": 0.5
	}'

Use Docker

docker model run hf.co/AzureBlack/opus-v0-70b-exl2

SGLang

How to use AzureBlack/opus-v0-70b-exl2 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 "AzureBlack/opus-v0-70b-exl2" \
    --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": "AzureBlack/opus-v0-70b-exl2",
		"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 "AzureBlack/opus-v0-70b-exl2" \
        --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": "AzureBlack/opus-v0-70b-exl2",
		"prompt": "Once upon a time,",
		"max_tokens": 512,
		"temperature": 0.5
	}'

Docker Model Runner
How to use AzureBlack/opus-v0-70b-exl2 with Docker Model Runner:
```
docker model run hf.co/AzureBlack/opus-v0-70b-exl2
```

ExllamaV2 version of the model created by dreamgen!

Original Model https://huggingface.co/dreamgen/opus-v0-70b

Requires ExllamaV2, which is being developed by turboderp https://github.com/turboderp/exllamav2 under an MIT license.

Main branch is 5bpw 8h

6b8h is 6bpw and 8h

4.6b8h is 4.6bpw and 8h

2.5b8h is 2.5bpw and 8h

DreamGen Opus V0 70B

DreamGen Opus is a family of uncensored models fine-tuned for (steerable) story writing and the model also works great for chat / RP. The DreamGen Opus V0 70B model is derived from meta-llama/Llama-2-70b-hf.

You can try the Opus V0 70B (AWQ) model for free on dreamgen.com.

Quantized versions:

AWQ: dreamgen/opus-v0-70b-awq
GGUF: dreamgen/opus-v0-70b-gguf

Other sizes:

7B: dreamgen/opus-v0-7b

Prompting

Please see the official documentation for more detailed guide, including how to prompt the model for chat / RP.

The (collaborative / steerable) story writing task teaches the model to respect <setting> and <instruction> inserted into the prompt.

Example prompt:

<setting>
(Setting provides general overview of the story and characters)
This story is a twist on the traditional Little Red Riding Hood story.
In this variation, the Little Red Riding Hood and her grandma are secretely werevoles.
</setting>

(Previous part of the story, potentially empty)

<instruction>
(Setting tells the model what should happen in the next few sentences / paragraphs)
The Little Red Riding hood confronts The Big Bad Wolf, transforming into her wolf form.
</instruction>

Dataset

The fine-tuning dataset consisted of >1M tokens of collaborative writing task examples, each example being up to 4096 tokens. On top of that, >20M tokens of more general, but less instructed examples were included to help preserve generalization.

Community

Join the DreamGen community on Discord, or follow our X/Twitter account for new model releases and other news. We will soon be releasing models with longer context window, as well as models specifically fine-tuned for character chat & roleplay.

Help us shape the future of DreamGen.

Running the model

The model is should be compatible with any software that supports meta-llama/Llama-2-70b-hf. Note that because this is a 70B model, the resource requirements are large. You can try the quantized versions linked at the top, but expect a quality drop.

Running on DreamGen.com (free)

You can try the 70B (AWQ) model for free at dreamgen.com — note that an account is required. The version used for the website is the official AWQ 4bit quant dreamgen/opus-v0-70b-awq.

License

For personal and academic use: Same license as the base model, in this case https://ai.meta.com/resources/models-and-libraries/llama-downloads/.
For commercial use: Please reach out to hello@dreamgen.com.

Downloads last month: 5