Commit History

INT8 prefill: 7.7 GB FP32 β†’ 1.9 GB INT8 weight-only; audio_encoder/decoder stay FP32 (ai_edge_quantizer rejects Conv ops). Total bundle 10 GB β†’ 4.3 GB, runtime working set 13 GiB β†’ ~5-6 GiB.
9ec65a0
verified

aufklarer commited on

card: unified LiteRT model card with soniqo.audio + ecosystem links
2536d07
verified

aufklarer commited on

card: unified LiteRT model card with soniqo.audio + ecosystem links
97d10e4
verified

aufklarer commited on

card: unified LiteRT model card with soniqo.audio + ecosystem links
ecc3f03
verified

aufklarer commited on

Fix VoxCPM2 LiteRT semantic roundtrip
3c9bcbf
verified

aufklarer commited on

Update VoxCPM2 LiteRT bundle to 512 text tokens
96978c7
verified

aufklarer commited on

Add VoxCPM2 LiteRT INT8 smoke-tested bundle
dd17e92
verified

aufklarer commited on

initial commit
7982a1e
verified

aufklarer commited on