aufklarer/PersonaPlex-7B-MLX-8bit

PersonaPlex-7B MLX 8-bit

PersonaPlex 7B full-duplex speech-to-speech model converted to MLX safetensors with 8-bit quantization for Apple Silicon.

Converted from nvidia/personaplex-7b-v1 (based on Kyutai Moshi architecture).

Swift inference: soniqo/speech-swift

Model Details

Component	Architecture	Size
Temporal Transformer	32-layer, 4096d, 32 heads (7B params)	~6.5 GB (8-bit)
Depformer	6-layer, 1024d, 16 heads, per-codebook weights	~1.3 GB (8-bit)
Mimi Codec	SEANet encoder/decoder + 8L transformer + 16 RVQ codebooks	~370 MB (fp16)
Embeddings	Text + 16 audio embeddings + output heads	~940 MB (fp16)
Total		~9.1 GB

Usage

let model = try await PersonaPlexModel.fromPretrained(
    modelId: "aufklarer/PersonaPlex-7B-MLX-8bit"
)
let response = model.respond(audio: samples, voice: .NATF0, steps: 100)

audio personaplex input.wav --model aufklarer/PersonaPlex-7B-MLX-8bit -o output.wav

Variants

Variant	Quantization	Size	Model ID
4-bit	4-bit	~4.9 GB	aufklarer/PersonaPlex-7B-MLX-4bit
8-bit	8-bit	~9.1 GB	aufklarer/PersonaPlex-7B-MLX-8bit

Voices

18 voice presets available: NATF0-3, NATM0-3, VARF0-4, VARM0-4

---

Links

• Blog: blog.ivan.digital
• Library Docs: soniqo.audio