README: document stage-1 (noisy) revision + stage-2 latest

README.md

@@ -37,6 +37,24 @@ Load instructions (canonical):
 | `pytorch_model.bin` | decoder weights for **inference** (load with `NeuCodec._from_pretrained(..., decoder_depth=20)`) |
 | `last.ckpt` | full PyTorch-Lightning checkpoint **with optimizer states + LR schedulers** for **resuming training** |
 
+## Training stages / which revision to pull
+
+This model is trained in two stages. `main` always points at the **latest** stage.
+
+| stage | data | how to load |
+|---|---|---|
+| **Stage 1 — noisy mix** (steps 0→1.58M, `epoch=9-step=1580000`) | 8 base corpora + `scale44k` (~6,500 h, mostly noisy crowdsourced/podcast) | pin revision **`d2c612f4319c1c1e0c65a9431a7d6db1b7955ce0`** |
+| **Stage 2 — clean finetune** (from 1.58M, clean ≥44.1 kHz only; in progress) | TTS-Clean44k + Clean-Podcast + clean-teacher pool + EARS/Expresso | `main` (default) |
+
+```python
+# Stage 1 (first-stage noisy training) — pin the exact revision:
+model = NeuCodec._from_pretrained(
+    model_id="Scicom-intl/neucodec-44k-d20",
+    decoder_depth=20,
+    revision="d2c612f4319c1c1e0c65a9431a7d6db1b7955ce0",
+)
+```
+
 ## Training data
 
 Diverse, predominantly high-sample-rate human speech (no synthetic TTS). The **scale-up group is