Instructions to use cubbk/orpheus-swedish with libraries, inference providers, notebooks, and local apps. Follow these links to get started.

Libraries

How to use cubbk/orpheus-swedish with Transformers:

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

pipe = pipeline("text-generation", model="cubbk/orpheus-swedish")
messages = [
    {"role": "user", "content": "Who are you?"},
]
pipe(messages)

# Load model directly
from transformers import AutoTokenizer, AutoModelForCausalLM

tokenizer = AutoTokenizer.from_pretrained("cubbk/orpheus-swedish")
model = AutoModelForCausalLM.from_pretrained("cubbk/orpheus-swedish", device_map="auto")
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 cubbk/orpheus-swedish with vLLM:

Install from pip and serve model

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

Use Docker

docker model run hf.co/cubbk/orpheus-swedish

SGLang

How to use cubbk/orpheus-swedish 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 "cubbk/orpheus-swedish" \
    --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": "cubbk/orpheus-swedish",
		"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 "cubbk/orpheus-swedish" \
        --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": "cubbk/orpheus-swedish",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Docker Model Runner
How to use cubbk/orpheus-swedish with Docker Model Runner:
```
docker model run hf.co/cubbk/orpheus-swedish
```

Orpheus-swedish

A text-to-speech model based on Orpheus-TTS finetuned with just ~50h of swedish parliament speeches to be able to generate swedish audio.

While the dataset contains 5000h I used only 50h as a test. https://huggingface.co/datasets/KBLab/rixvox

Here are some results:

Billströms kritik är extra känslig för statsministern i och med att Billström tillhört den innersta kretsen i regeringen och partiet.`

Möjligen har också Billström känt sig trampad på tårna och kritiken mot NSR kan ha varit en bidragande orsak till att han förra året oväntat avgick som utrikesminister.Tack

Hej, hur mår du?

Vad är klockan?

Vilken färg har himlen?

Vad är meningen med livet?

How to use

%pip install transformers snac soundfile torch torchaudio

from transformers import pipeline

# Use your custom task name and model repo id
pipe = pipeline(task="orpheus-swedish", model="cubbk/orpheus-swedish", trust_remote_code=True)

from IPython.display import Audio, display

prompt = [
        "Enligt brittiska medier kommer ledarna att presentera en techpakt som ska stärka ländernas samarbete kring AI, kvantfysik och kärnkraft."
    ]
outputs = pipe(prompt)

for i in range(len(outputs)):
    print(prompt[i])
    samples = outputs[i][0]
    display(Audio(samples.detach().squeeze().to("cpu").numpy(), rate=24000))

Good to know: performs poorly on short text Cuts out at 14 sec.(didn't figure out why)

Downloads last month: 162

Safetensors

Model size

3B params

Tensor type

BF16