Instructions to use syvai/plapre-nano with libraries, inference providers, notebooks, and local apps. Follow these links to get started.

Libraries

How to use syvai/plapre-nano with Transformers:

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

pipe = pipeline("text-to-speech", model="syvai/plapre-nano")

# Load model directly
from transformers import AutoTokenizer, AutoModelForCausalLM

tokenizer = AutoTokenizer.from_pretrained("syvai/plapre-nano")
model = AutoModelForCausalLM.from_pretrained("syvai/plapre-nano")

llama-cpp-python

How to use syvai/plapre-nano with llama-cpp-python:

# !pip install llama-cpp-python

from llama_cpp import Llama

llm = Llama.from_pretrained(
	repo_id="syvai/plapre-nano",
	filename="gguf/plapre-nano.f16.gguf",
)

output = llm(
	"Once upon a time,",
	max_tokens=512,
	echo=True
)
print(output)

Notebooks
Google Colab
Kaggle
Local Apps

llama.cpp

How to use syvai/plapre-nano with llama.cpp:

Install from brew

brew install llama.cpp
# Start a local OpenAI-compatible server with a web UI:
llama-server -hf syvai/plapre-nano:Q4_K_M
# Run inference directly in the terminal:
llama-cli -hf syvai/plapre-nano:Q4_K_M

Install from WinGet (Windows)

winget install llama.cpp
# Start a local OpenAI-compatible server with a web UI:
llama-server -hf syvai/plapre-nano:Q4_K_M
# Run inference directly in the terminal:
llama-cli -hf syvai/plapre-nano:Q4_K_M

Use pre-built binary

# Download pre-built binary from:
# https://github.com/ggerganov/llama.cpp/releases
# Start a local OpenAI-compatible server with a web UI:
./llama-server -hf syvai/plapre-nano:Q4_K_M
# Run inference directly in the terminal:
./llama-cli -hf syvai/plapre-nano:Q4_K_M

Build from source code

git clone https://github.com/ggerganov/llama.cpp.git
cd llama.cpp
cmake -B build
cmake --build build -j --target llama-server llama-cli
# Start a local OpenAI-compatible server with a web UI:
./build/bin/llama-server -hf syvai/plapre-nano:Q4_K_M
# Run inference directly in the terminal:
./build/bin/llama-cli -hf syvai/plapre-nano:Q4_K_M

Use Docker

docker model run hf.co/syvai/plapre-nano:Q4_K_M

LM Studio
Jan
Ollama
How to use syvai/plapre-nano with Ollama:
```
ollama run hf.co/syvai/plapre-nano:Q4_K_M
```

Unsloth Studio new

How to use syvai/plapre-nano with Unsloth Studio:

Install Unsloth Studio (macOS, Linux, WSL)

curl -fsSL https://unsloth.ai/install.sh | sh
# Run unsloth studio
unsloth studio -H 0.0.0.0 -p 8888
# Then open http://localhost:8888 in your browser
# Search for syvai/plapre-nano to start chatting

Install Unsloth Studio (Windows)

irm https://unsloth.ai/install.ps1 | iex
# Run unsloth studio
unsloth studio -H 0.0.0.0 -p 8888
# Then open http://localhost:8888 in your browser
# Search for syvai/plapre-nano to start chatting

Using HuggingFace Spaces for Unsloth

# No setup required
# Open https://huggingface.co/spaces/unsloth/studio in your browser
# Search for syvai/plapre-nano to start chatting

Docker Model Runner
How to use syvai/plapre-nano with Docker Model Runner:
```
docker model run hf.co/syvai/plapre-nano:Q4_K_M
```

Lemonade

How to use syvai/plapre-nano with Lemonade:

Pull the model

# Download Lemonade from https://lemonade-server.ai/
lemonade pull syvai/plapre-nano:Q4_K_M

Run and chat with the model

lemonade run user.plapre-nano-Q4_K_M

List all available models

lemonade list

You need to agree to share your contact information to access this model

This repository is publicly accessible, but you have to accept the conditions to access its files and content.

Plapre Nano - Dansk Tekst-til-Tale

Dansk TTS-model med talerkonditionering og stemmekloningssupport. Genererer 24kHz lyd fra dansk tekst ved hjælp af autoregressiv lydtoken-prædiktion.

Modeldetaljer


Arkitektur	SmolLM2-360M (LLaMA-baseret, 32 lag, hidden_size=960)
Parametre	~334.6M (base) + 123K (talerprojektion)
Vocab-størrelse	20.802 (8000 BPE + 12.800 lyd + separatorer)
Lydtokenizer	Kanade (25 tokens/sek, 12.800 codebook)
Samplerate	24kHz
Præcision	bfloat16

Sådan virker det

Modellen tager dansk tekst, konverterer den til BPE-tokens, og genererer autoregressivt Kanade-lydtokens, som afkodes til en lydbølge.

Sekvensformat:

[speaker_embedding] <text> BPE tokens <audio> audio tokens <eos>

Talerkonditionering: En indlært lineær projektion (nn.Linear(128, 960)) mapper en 128-dimensionel Kanade-talerembedding til modellens skjulte dimension. Denne indsættes som det første token i sekvensen, så modellen kan konditionere på taleridentitet via attention. Til stemmekloning udtrækkes talerembeddingen fra et referenceaudioklip via Kanade-encoderen.

Træningsdata

NST-DA
FTSpeech

Installation

uv add git+https://github.com/syv-ai/plapre.git

Inferens

Grundlæggende brug

from plapre import Plapre

tts = Plapre("syvai/plapre-nano")
tts.speak("Hej, hvordan har du det?", output="output.wav")

Vis tilgængelige talere

print(tts.list_speakers())
# ['tor', 'ida', 'liv', 'ask', 'kaj']

Vælg en taler

tts.speak("Hej med dig.", output="output.wav", speaker="ida")

Stemmekloning

tts.speak("Hej med dig.", output="cloned.wav", speaker_wav="reference.wav")

Lange tekster med sætningsopdeling

tts.speak(
    "Første sætning. Anden sætning. Tredje sætning!",
    output="long.wav",
    split_sentences=True,
)

Genereringsparametre

tts.speak(
    "Hej verden.",
    output="output.wav",
    temperature=0.8,     # sampling-temperatur (standard: 0.8)
    top_p=0.95,          # nucleus sampling (standard: 0.95)
    top_k=50,            # top-k sampling (standard: 50)
    max_tokens=500,      # maks lydtokens at generere (standard: 500)
)

Udtræk en talerembedding

Udtræk en 128-dim talerembedding fra en wav-fil og genbrug den på tværs af flere genereringer:

speaker_emb = tts.extract_speaker("reference.wav")
tts.speak("Hej.", output="a.wav", speaker_emb=speaker_emb)
tts.speak("Farvel.", output="b.wav", speaker_emb=speaker_emb)

Returværdi

speak() returnerer lyden som et numpy-array (24 kHz, float32) ud over at gemme filen:

audio = tts.speak("Hej.", output="output.wav")
print(f"Varighed: {len(audio) / 24000:.2f}s")

Downloads last month: 59,069

Safetensors

Model size

0.3B params

Tensor type

BF16

syvai
/

plapre-nano