← Catalog

Fish Audio S2 Pro

safetensors · fishaudio/s2-pro

queued non-commercial

Publisherfishaudio
Modalitytext-to-speech
Parameters4.56B
Card: 4B slow AR + 400M fast AR. HF safetensors count 4.56B. Source: model card.
Formatsafetensors
Size on disk10.26 GiB (11,011,629,649 bytes)
Estimated memory to run~15.0 GB (estimate; see method)
Native contextn/a
Enginefish-speech · recipe
Licensefish-audio-research-license
Research/non-commercial free; commercial use needs a separate Fish Audio license (card).
Archive foldermedia
NotesCard: 4B slow AR + 400M fast AR. HF safetensors count 4.56B.

Where it runs

MachineVerdictBudget used
Bosgame M5 (x1)Fits~108-110 GB GPU-usable on Linux with GTT raised (96 GB cap on Windows). Fit budget used here: 108 GB.
Bosgame M5 (x2, RPC)Fits~200-225 GB usable across two nodes. Fit budget used here: 210 GB; 210-225 GB = tight.
MacBook Pro (M4 Max)FitsmacOS caps GPU wired memory by default (~75% = ~27 GB). Raising iogpu.wired_limit_mb to ~32 GB is possible but leaves little for the OS.
Mac mini (M4)No~2/3 of 16 GB usable by the GPU by default. Already busy as build box + Hermes/Ollama.
omarchy (AMD desktop)No7 GB of the 8 GB VRAM for weights. Offload capacity depends on system RAM, which is unknown today; after the 128 GB upgrade ~110 GB of MoE experts could sit in RAM.
Ugreen DXP2800 NASNoStorage host. Already runs ollama/open-webui containers; only tiny models (e.g. Whisper on CPU) are realistic.
MacBook NeoUnknownSpecs unknown: machine was offline/unreachable on Nov 14 2026.

Token/s = theoretical ceiling from memory bandwidth and active weight bytes; not a benchmark. Nothing on this page is a benchmark score.