Instructions to use idle-intelligence/personaplex-24L-q4_k-webgpu with libraries, inference providers, notebooks, and local apps. Follow these links to get started.

Libraries

How to use idle-intelligence/personaplex-24L-q4_k-webgpu with Moshi:

# pip install moshi
# Run the interactive web server
python -m moshi.server --hf-repo "idle-intelligence/personaplex-24L-q4_k-webgpu"
# Then open https://localhost:8998 in your browser

# pip install moshi
import torch
from moshi.models import loaders

# Load checkpoint info from HuggingFace
checkpoint = loaders.CheckpointInfo.from_hf_repo("idle-intelligence/personaplex-24L-q4_k-webgpu")

# Load the Mimi audio codec
mimi = checkpoint.get_mimi(device="cuda")
mimi.set_num_codebooks(8)

# Encode audio (24kHz, mono)
wav = torch.randn(1, 1, 24000 * 10)  # [batch, channels, samples]
with torch.no_grad():
    codes = mimi.encode(wav.cuda())
    decoded = mimi.decode(codes)

llama-cpp-python

How to use idle-intelligence/personaplex-24L-q4_k-webgpu with llama-cpp-python:

# !pip install llama-cpp-python

from llama_cpp import Llama

llm = Llama.from_pretrained(
	repo_id="idle-intelligence/personaplex-24L-q4_k-webgpu",
	filename="personaplex-24L-q4_k.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 idle-intelligence/personaplex-24L-q4_k-webgpu with llama.cpp:

Install from brew

brew install llama.cpp
# Start a local OpenAI-compatible server with a web UI:
llama-server -hf idle-intelligence/personaplex-24L-q4_k-webgpu
# Run inference directly in the terminal:
llama-cli -hf idle-intelligence/personaplex-24L-q4_k-webgpu

Install from WinGet (Windows)

winget install llama.cpp
# Start a local OpenAI-compatible server with a web UI:
llama-server -hf idle-intelligence/personaplex-24L-q4_k-webgpu
# Run inference directly in the terminal:
llama-cli -hf idle-intelligence/personaplex-24L-q4_k-webgpu

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 idle-intelligence/personaplex-24L-q4_k-webgpu
# Run inference directly in the terminal:
./llama-cli -hf idle-intelligence/personaplex-24L-q4_k-webgpu

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 idle-intelligence/personaplex-24L-q4_k-webgpu
# Run inference directly in the terminal:
./build/bin/llama-cli -hf idle-intelligence/personaplex-24L-q4_k-webgpu

Use Docker

docker model run hf.co/idle-intelligence/personaplex-24L-q4_k-webgpu

LM Studio
Jan
Ollama
How to use idle-intelligence/personaplex-24L-q4_k-webgpu with Ollama:
```
ollama run hf.co/idle-intelligence/personaplex-24L-q4_k-webgpu
```

Unsloth Studio new

How to use idle-intelligence/personaplex-24L-q4_k-webgpu 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 idle-intelligence/personaplex-24L-q4_k-webgpu 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 idle-intelligence/personaplex-24L-q4_k-webgpu to start chatting

Using HuggingFace Spaces for Unsloth

# No setup required
# Open https://huggingface.co/spaces/unsloth/studio in your browser
# Search for idle-intelligence/personaplex-24L-q4_k-webgpu to start chatting

Docker Model Runner
How to use idle-intelligence/personaplex-24L-q4_k-webgpu with Docker Model Runner:
```
docker model run hf.co/idle-intelligence/personaplex-24L-q4_k-webgpu
```

Lemonade

How to use idle-intelligence/personaplex-24L-q4_k-webgpu with Lemonade:

Pull the model

# Download Lemonade from https://lemonade-server.ai/
lemonade pull idle-intelligence/personaplex-24L-q4_k-webgpu

Run and chat with the model

lemonade run user.personaplex-24L-q4_k-webgpu-{{QUANT_TAG}}

List all available models

lemonade list

code example

by Shadow0482 - opened 30 days ago

Discussion

Shadow0482

30 days ago

Guys any interface codes to run the model? Like Python codes or something like that?

ilnmtlbnm

Idle Intelligence org 29 days ago

Hey Shadow,

I'm not sure you've seen the disclaimer on the Model car :
"Work in progress. This model exists primarily to test layer pruning + QLoRA recovery for browser deployment. Model behavior may differ from the original 32L model. No guarantees — use at your own discretion."

That being said, I built this to play with Speech to Speech in the browser, you can find some inference code (in rust, sorry) over there: https://github.com/idle-intelligence/sts-web and there's a demo link in there too.

I guess if you're interested I do have, somewhere, a version of nvidia's python inference that will run this model, let me know!

Shadow0482

27 days ago

Hey ilnmtlbnm , thanks for sharing this looks really interesting.

I don’t think I’m seeing the model card on my end, not sure if I missed it
Also, I’d definitely be interested in the Python inference version (the NVIDIA one you mentioned), if you don’t mind sharing it. That would be super helpful for me to experiment with.

Appreciate it

ilnmtlbnm

Idle Intelligence org 25 days ago

Hey Shadow, I appreciate your interest. I meant the model card here: https://huggingface.co/idle-intelligence/personaplex-24L-q4_k-webgpu

For the python inference... I didn't think this through when answering initially (I did this thing a few weeks back already 😅 .
This GGUF is custom (layer-pruned + LoRA-recovered + Q4_K with Moshi-style tensor names), so it can't be loaded with NVIDIA's PersonaPlex Python inference as-is.

I added a native Rust CLI that runs the model end-to-end (Vulkan on Linux, Metal on macOS, no Python required):
https://github.com/idle-intelligence/sts-web#native-cli-sts

huggingface-cli download idle-intelligence/personaplex-24L-q4_k-webgpu
--local-dir personaplex-24L-q4_k-webgpu

cargo run --release --features "wgpu,cli" --bin sts -- \
--model-dir ./personaplex-24L-q4_k-webgpu \
--input question.wav \
--output response.wav
--voice NATF2

Takes ~5 min for the first build (Burn + cubecl), then ~64 ms/frame on a 3080. Voices: NATF0..3, NATM0..3, VARF0..4, VARM0..4.

Upload images, audio, and videos by dragging in the text input, pasting, or clicking here.

Tap or paste here to upload images

· Sign up or log in to comment