Instructions to use IHaBiS/PiVoT-MoE-exl2 with libraries, inference providers, notebooks, and local apps. Follow these links to get started.

Libraries

How to use IHaBiS/PiVoT-MoE-exl2 with Transformers:

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

pipe = pipeline("text-generation", model="IHaBiS/PiVoT-MoE-exl2")
messages = [
    {"role": "user", "content": "Who are you?"},
]
pipe(messages)

# Load model directly
from transformers import AutoTokenizer, AutoModelForCausalLM

tokenizer = AutoTokenizer.from_pretrained("IHaBiS/PiVoT-MoE-exl2")
model = AutoModelForCausalLM.from_pretrained("IHaBiS/PiVoT-MoE-exl2")
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

vLLM

How to use IHaBiS/PiVoT-MoE-exl2 with vLLM:

Install from pip and serve model

# Install vLLM from pip:
pip install vllm
# Start the vLLM server:
vllm serve "IHaBiS/PiVoT-MoE-exl2"
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:8000/v1/chat/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "IHaBiS/PiVoT-MoE-exl2",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Use Docker

docker model run hf.co/IHaBiS/PiVoT-MoE-exl2

SGLang

How to use IHaBiS/PiVoT-MoE-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 "IHaBiS/PiVoT-MoE-exl2" \
    --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": "IHaBiS/PiVoT-MoE-exl2",
		"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 "IHaBiS/PiVoT-MoE-exl2" \
        --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": "IHaBiS/PiVoT-MoE-exl2",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Docker Model Runner
How to use IHaBiS/PiVoT-MoE-exl2 with Docker Model Runner:
```
docker model run hf.co/IHaBiS/PiVoT-MoE-exl2
```

Exl2 version of maywell/PiVoT-MoE

branch

main : 2.4bpw h8
3bh8 : 3bpw h8
4bh8 : 4bpw h8
6bh8 : 6bpw h8
8bh8 : 8bpw h8

Using ThePile 0007.parquet as dataset

Quantization settings : python convert.py -i models/maywell_PiVoT-MoE -o PiVoT-MoE-temp -cf PiVoT-MoE-8bpw-h8-exl2 -c 0007.parquet -l 8192 -b 8 -hb 8 -ml 8192
python convert.py -i models/maywell_PiVoT-MoE -o PiVoT-MoE-temp2 -cf PiVoT-MoE-6bpw-h8-exl2 -c 0007.parquet -l 8192 -b 6 -hb 8 -m PiVoT-MoE-temp/measurement.json -ml 8192
python convert.py -i models/maywell_PiVoT-MoE -o PiVoT-MoE-temp3 -cf PiVoT-MoE-4bpw-h8-exl2 -c 0007.parquet -l 8192 -b 4 -hb 8 -m PiVoT-MoE-temp/measurement.json -ml 8192
python convert.py -i models/maywell_PiVoT-MoE -o PiVoT-MoE-temp4 -cf PiVoT-MoE-3bpw-h8-exl2 -c 0007.parquet -l 8192 -b 3 -hb 8 -m PiVoT-MoE-temp/measurement.json -ml 8192
python convert.py -i models/maywell_PiVoT-MoE -o PiVoT-MoE-temp5 -cf PiVoT-MoE-2.4bpw-h8-exl2 -c 0007.parquet -l 8192 -b 2.4 -hb 8 -m PiVoT-MoE-temp/measurement.json -ml 8192

below this line is original readme

PiVot-MoE

Model Description

PiVoT-MoE, is an advanced AI model specifically designed for roleplaying purposes. It has been trained using a combination of four 10.7B sized experts, each with their own specialized characteristic, all fine-tuned to bring a unique and diverse roleplaying experience.

The Mixture of Experts (MoE) technique is utilized in this model, allowing the experts to work together synergistically, resulting in a more cohesive and natural conversation flow. The MoE architecture allows for a higher level of flexibility and adaptability, enabling PiVoT-MoE to handle a wide variety of roleplaying scenarios and characters.

Based on the PiVoT-10.7B-Mistral-v0.2-RP model, PiVoT-MoE takes it a step further with the incorporation of the MoE technique. This means that not only does the model have an expansive knowledge base, but it also has the ability to mix and match its expertise to better suit the specific roleplaying scenario.

Prompt Template - Alpaca (ChatML works)

{system}
### Instruction:
{instruction}
### Response:
{response}

Downloads last month: 3