Ctrl+K

1 contributor

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

9ec65a0 verified about 24 hours ago

.gitattributes

1.52 kB
initial commit 4 days ago
README.md

6.31 kB
card: unified LiteRT model card with soniqo.audio + ecosystem links 3 days ago
config.json

11.7 kB
INT8 prefill: 7.7 GB FP32 → 1.9 GB INT8 weight-only; audio_encoder/decoder stay FP32 (ai_edge_quantizer rejects Conv ops). Total bundle 10 GB → 4.3 GB, runtime working set 13 GiB → ~5-6 GiB. about 24 hours ago
generation_config.json

3 Bytes
Add VoxCPM2 LiteRT INT8 smoke-tested bundle 4 days ago
special_tokens_map.json

1.63 kB
Add VoxCPM2 LiteRT INT8 smoke-tested bundle 4 days ago
tokenization_voxcpm2.py

2.9 kB
Add VoxCPM2 LiteRT INT8 smoke-tested bundle 4 days ago
tokenizer.json

3.68 MB
Add VoxCPM2 LiteRT INT8 smoke-tested bundle 4 days ago
tokenizer_config.json

5.06 kB
Add VoxCPM2 LiteRT INT8 smoke-tested bundle 4 days ago
voxcpm2-audio-decoder.tflite

183 MB
xet

Add VoxCPM2 LiteRT INT8 smoke-tested bundle 4 days ago
voxcpm2-audio-encoder.tflite

192 MB
xet

Add VoxCPM2 LiteRT INT8 smoke-tested bundle 4 days ago
voxcpm2-text-prefill.tflite

2.08 GB
xet

INT8 prefill: 7.7 GB FP32 → 1.9 GB INT8 weight-only; audio_encoder/decoder stay FP32 (ai_edge_quantizer rejects Conv ops). Total bundle 10 GB → 4.3 GB, runtime working set 13 GiB → ~5-6 GiB. about 24 hours ago
voxcpm2-text-prefill_recipe.json

311 Bytes
INT8 prefill: 7.7 GB FP32 → 1.9 GB INT8 weight-only; audio_encoder/decoder stay FP32 (ai_edge_quantizer rejects Conv ops). Total bundle 10 GB → 4.3 GB, runtime working set 13 GiB → ~5-6 GiB. about 24 hours ago
voxcpm2-token-step.tflite

2.19 GB
xet

INT8 prefill: 7.7 GB FP32 → 1.9 GB INT8 weight-only; audio_encoder/decoder stay FP32 (ai_edge_quantizer rejects Conv ops). Total bundle 10 GB → 4.3 GB, runtime working set 13 GiB → ~5-6 GiB. about 24 hours ago
voxcpm2-token-step_recipe.json

311 Bytes
INT8 prefill: 7.7 GB FP32 → 1.9 GB INT8 weight-only; audio_encoder/decoder stay FP32 (ai_edge_quantizer rejects Conv ops). Total bundle 10 GB → 4.3 GB, runtime working set 13 GiB → ~5-6 GiB. about 24 hours ago