Kokoro-82M-GGUF

Kokoro-82M is a lightweight, open-weight text-to-speech (TTS) model with 82 million parameters released under an Apache license, making it free for both personal and commercial deployment. Built on a decoder-only architecture based on StyleTTS 2 and ISTFTNet (without diffusion or an encoder release) and utilizing the Misaki grapheme-to-phoneme library, its v1.0 release supports 8 languages and 54 voices while delivering audio quality comparable to much larger models at a fraction of the compute costβ€”averaging under $1 per million characters of input or $0.06 per hour of generated audio via API. Remarkably efficient to develop, the model was trained for roughly $1,000 across 1,000 hours on A100 80GB GPUs using a few hundred hours of exclusively permissive, public domain, and synthetic audio paired with IPA phoneme labels.

Use F32 for the best quality. Supports llama-tts via llama.cpp.

Model Files

File Name Quant Type File Size File Link Description
Kokoro-82M.F32.gguf F32 23.3 MB Link Full 32-bit float weights. Maximum precision, largest file size.
Kokoro-82M.BF16.gguf BF16 11.8 MB Link Full BF16 weights. High quality, excellent balance of precision and size.
Kokoro-82M.Q3_K_L.gguf Q3_K_L 3.55 MB Link Lower quality but usable, good for very low RAM availability.
Kokoro-82M.Q3_K_M.gguf Q3_K_M 3.28 MB Link Low quality, smallest file size.
Kokoro-82M.Q4_K_M.gguf Q4_K_M 4.14 MB Link Good quality, default size for most use cases, recommended.
Kokoro-82M.Q4_K_S.gguf Q4_K_S 3.86 MB Link Slightly lower quality with more space savings, recommended.
Kokoro-82M.Q5_K_M.gguf Q5_K_M 4.57 MB Link High quality, recommended.
Kokoro-82M.Q5_K_S.gguf Q5_K_S 4.28 MB Link High quality with slight space savings, recommended.
Kokoro-82M.Q6_K.gguf Q6_K 5.04 MB Link Very high quality, near perfect, recommended.
Kokoro-82M.Q8_0.gguf Q8_0 6.41 MB Link Extremely high quality, generally unneeded but max available standard quant.
Kokoro-82M.mmproj-bf16.gguf mmproj-bf16 208 MB Link Multimodal projection file in BF16 format. Used for audio/voice capabilities.

Run with llama.cpp

Install dependencies and clone llama.cpp

!pip install transformers huggingface-hub

!git clone https://github.com/ggml-org/llama.cpp.git
%cd /data/llama.cpp

Download GGUF model files

Download the BF16 model and BF16 multimodal projector:

from huggingface_hub import snapshot_download

# Target local directory
local_dir = "./model/Kokoro-82M-GGUF"

# Download only the specified files
snapshot_download(
    repo_id="prithivMLmods/Kokoro-82M-GGUF",
    local_dir=local_dir,
    allow_patterns=[
        "Kokoro-82M.BF16.gguf",
        "Kokoro-82M.mmproj-bf16.gguf",
    ],
    resume_download=True,
)

Build llama-tts

Build the TTS CLI with the CUDA backend enabled:

!cmake --build . --config Release -j$(nproc) --target llama-tts

Text-to-speech generation (GPU offloading)

Run inference using llama-tts with GPU offloading enabled:

!/data/llama.cpp/build/bin/llama-tts \
    -m /data/llama.cpp/model/Kokoro-82M-GGUF/Kokoro-82M.BF16.gguf \
    -mm /data/llama.cpp/model/Kokoro-82M-GGUF/Kokoro-82M.mmproj-bf16.gguf \
    --tts-voice am_michael \
    -p "Kokoro is an open-weight TTS model with 82 million parameters. Despite its lightweight architecture, it delivers comparable quality to larger models while being significantly faster and more cost-efficient. With Apache-licensed weights, Kokoro can be deployed anywhere from production environments to personal projects." \
    -ngl 99 \
    -n 1000 \
    --seed 42 \
    --output /data/llama.cpp/model/Kokoro-82M-GGUF/out.wav

Output

The generated audio file will be saved at:

/data/llama.cpp/model/Kokoro-82M-GGUF/out.wav

You can play the generated WAV file using an audio player or a notebook audio widget.

llama.cpp

LLM inference in C/C++ β€” https://github.com/ggml-org/llama.cpp

Quantized using llama.cpp by prithivMLmods.

Downloads last month
549
GGUF
Model size
5.82M params
Architecture
kokoro
Hardware compatibility
Log In to add your hardware

3-bit

4-bit

5-bit

6-bit

8-bit

16-bit

32-bit

Inference Providers NEW
This model isn't deployed by any Inference Provider. πŸ™‹ Ask for provider support

Model tree for prithivMLmods/Kokoro-82M-GGUF

Quantized
(130)
this model

Collections including prithivMLmods/Kokoro-82M-GGUF