Instructions to use royallab/Pygmalion-2-13b-SuperCOT-weighed with libraries, inference providers, notebooks, and local apps. Follow these links to get started.

Libraries

How to use royallab/Pygmalion-2-13b-SuperCOT-weighed with Transformers:

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

pipe = pipeline("text-generation", model="royallab/Pygmalion-2-13b-SuperCOT-weighed")

# Load model directly
from transformers import AutoTokenizer, AutoModelForCausalLM

tokenizer = AutoTokenizer.from_pretrained("royallab/Pygmalion-2-13b-SuperCOT-weighed")
model = AutoModelForCausalLM.from_pretrained("royallab/Pygmalion-2-13b-SuperCOT-weighed")

Inference
Notebooks
Google Colab
Kaggle
Local Apps Settings

vLLM

How to use royallab/Pygmalion-2-13b-SuperCOT-weighed with vLLM:

Install from pip and serve model

# Install vLLM from pip:
pip install vllm
# Start the vLLM server:
vllm serve "royallab/Pygmalion-2-13b-SuperCOT-weighed"
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:8000/v1/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "royallab/Pygmalion-2-13b-SuperCOT-weighed",
		"prompt": "Once upon a time,",
		"max_tokens": 512,
		"temperature": 0.5
	}'

Use Docker

docker model run hf.co/royallab/Pygmalion-2-13b-SuperCOT-weighed

SGLang

How to use royallab/Pygmalion-2-13b-SuperCOT-weighed 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 "royallab/Pygmalion-2-13b-SuperCOT-weighed" \
    --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": "royallab/Pygmalion-2-13b-SuperCOT-weighed",
		"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 "royallab/Pygmalion-2-13b-SuperCOT-weighed" \
        --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": "royallab/Pygmalion-2-13b-SuperCOT-weighed",
		"prompt": "Once upon a time,",
		"max_tokens": 512,
		"temperature": 0.5
	}'

Docker Model Runner
How to use royallab/Pygmalion-2-13b-SuperCOT-weighed with Docker Model Runner:
```
docker model run hf.co/royallab/Pygmalion-2-13b-SuperCOT-weighed
```

Model Card: Pygmalion-2-13b-SuperCOT-weighted

This is an experimental weighted merge between:

Quantizations provided by us and TheBloke:

The merge was performed by a gradient merge script (apply-lora-weight-ltl.py) from zaraki-tools by Zaraki.

Thanks to Zaraki for the inspiration and help.

This merge differs from the previous Pyg-2-SuperCOT merges. The first iteration of the SuperCOT loras were used here since it performed better than SuperCOT2.

The SuperCOT lora was merged with the following layer weights (basically 50/50. The exact ratio is 0.51)

0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0.5,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1

Here is an image to help visualize this merge. The light blue is Pygmalion-2-13b and the light green is the SuperCOT lora:

Usage:

Since this is an experimental weight merge between Pygmalion-2 and SuperCOT, the following instruction formats should work:

Metharme:

<|system|>This is a text adventure game. Describe the scenario to the user and give him three options to pick from on each turn.<|user|>Start!<|model|>

Alpaca:

### Instruction:
Your instruction or question here.
### Response:

Bias, Risks, and Limitations

The model will show biases similar to those observed in niche roleplaying forums on the Internet, besides those exhibited by the base model. It is not intended for supplying factual information or advice in any form.

In addition, this merge is experimental from our own testing. Your results may vary.

Training Details

This model is merged and can be reproduced using the tools mentioned above. Please refer to all provided links for extra model-specific details.

Downloads last month: 5

Safetensors

Model size

13B params

Tensor type

F16

Collection including royallab/Pygmalion-2-13b-SuperCOT-weighed

Models

Collection

FP16 models and Loras that we've created • 7 items • Updated Feb 15, 2024