-
Mirror OpenMOSS-Team/MOSS-Audio-4B-Instruct @ 6907a499dc0e for MAESTRO
- 3 bf16 safetensors shards VERBATIM (LLM-text output keeps
bf16; sha256-checked against upstream LFS metadata)
- tokenizer + processor configs bundled for fully-offline runtime
- HF remote-code .py files dropped: MAESTRO vendors the GitHub
src/ inference package (ai/moss_audio_vendor), never
trust_remote_code
- structural verify (meta-skeleton key equivalence + processor
load) through the vendored graph before upload
Loaded by MAESTRO via backend/ai/models/moss_audio.py.
