Fix README.md benchmark methodology

README.md

@@ -50,18 +50,28 @@ reconstructions, RGB error deltas, and latent PCA side by side, with FLUX.2 VAE
 included for comparison:
 [semdisdiffae_p32_v2 reconstruction viewer](https://huggingface.co/spaces/data-archetype/semdisdiffae_p32_v2-results).
 
-## Throughput
+## Encode Throughput
 
-Measured on an `NVIDIA GeForce RTX 5090` in `bfloat16`, with `5` warmup batches
-and `20` timed batches. Decode uses the default 1-step sampler with PDG
-disabled.
+Measured on an `NVIDIA GeForce RTX 5090` in `bfloat16`, averaging `20`
+repeated batched `encode()` calls after `5` warmup batches.
 
-| Operation | Resolution | Batch Size | Mean (ms/batch) | Images/s | Peak Allocated VRAM |
-|---|---:|---:|---:|---:|---:|
-| Encode | `256x256` | `128` | `12.57` | `10186.8` | `574 MiB` |
-| Decode | `256x256` | `128` | `98.93` | `1293.9` | `1042 MiB` |
-| Encode | `512x512` | `32` | `12.08` | `2649.9` | `579 MiB` |
-| Decode | `512x512` | `32` | `100.36` | `318.8` | `1042 MiB` |
+| Resolution | Batch Size | Mean (ms/batch) | ms/image | Images/s | Peak Allocated VRAM |
+|---:|---:|---:|---:|---:|---:|
+| `256x256` | `128` | `12.54` | `0.098` | `10206.3` | `567.8 MiB` |
+| `512x512` | `32` | `12.09` | `0.378` | `2647.2` | `563.8 MiB` |
+
+## Decode Latency
+
+Measured on the same `NVIDIA GeForce RTX 5090` in `bfloat16`. This is
+decode-only latency: images are encoded once, latents are cached, and timing is
+sequential batch-1 `decode()` over the cached latent set with the default 1-step
+sampler and PDG disabled.
+
+| Resolution | Batch Size | Images | Mean (ms/image) | Images/s | Peak Allocated VRAM |
+|---:|---:|---:|---:|---:|---:|
+| `512x512` | `1` | `20` | `5.11` | `195.6` | `340.8 MiB` |
+| `1024x1024` | `1` | `20` | `10.14` | `98.6` | `409.6 MiB` |
+| `2048x2048` | `1` | `20` | `53.86` | `18.6` | `720.9 MiB` |
 
 ## Latent Interface