Model card: the starting point (PyTorch on the Mac GPU) and a 3090 for scale

README.md

@@ -31,9 +31,13 @@ The backbone, the text encoder and every embedding stay **bf16, exactly as in [m
 
 | Build (Apple M4 Pro, 48 GB) | First audio, plain / directed | RTF, directed | Memory |
 |---|---|---|---|
+| BreezeBlue's PyTorch, bf16, on the Mac GPU (`model.generate`; its streaming path is CUDA-only) | no streaming: a one-sentence line takes ~6 s | 2.26 (plain lines) | 11 GB |
 | [mlx-community 8-bit](https://huggingface.co/mlx-community/Breeze-TTS-2-mlx-8bit) (mxfp8) | 212 / 297 ms | 0.58 | 13 GB |
 | [mlx-community bf16](https://huggingface.co/mlx-community/Breeze-TTS-2-mlx) | 380 / 456 ms | 0.86 | 15 GB |
 | **this checkpoint**, with the streaming server's speed settings | **139 / 207 ms** | **0.62** | **8.9 GB** |
+| *For reference: RTX 3090 (power-capped at 250 W of 350 W), BreezeBlue's CUDA fast path via a small server wrapper, re-processing the reference on every line* | *87 / 160 ms* | *0.50* | *19.7 GB VRAM + 4 GB RAM* |
+
+On the same Mac, BreezeBlue's own PyTorch code runs at RTF 2.26 (over twice slower than real time) and can't stream; this build is ~3.6x faster and streams its first audio in 139 / 207 ms. For scale, a power-capped RTX 3090 running BreezeBlue's CUDA fast path measured 87 / 160 ms in our setup. The MLX builds' memory includes MLX's buffer cache (uncapped in the mlx-community builds, capped at 1 GB here).
 
 Quality against bf16 on the same tokens (KL of the samplers' logits, lower is closer; lines with a voice direction at CFG 2):