README: use Scicom-intl/neucodec-44k-d20 + correct _from_pretrained(decoder_depth=20) + link GitHub #load-from-hugging-face

README.md

@@ -27,11 +27,14 @@ NeuCodec, so this model is a drop-in higher-fidelity decoder.
 Training pipeline, configs and RunPod automation:
 **https://github.com/Scicom-AI-Enterprise-Organization/neucodec-44k**
 
+Load instructions (canonical):
+**https://github.com/Scicom-AI-Enterprise-Organization/neucodec-44k#load-from-hugging-face**
+
 ## Files
 
 | file | what |
 |---|---|
-| `pytorch_model.bin` | decoder weights for **inference** (load with `NeuCodec.from_pretrained`) |
+| `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 data
@@ -404,17 +407,23 @@ and SR-verified ≥ 44 kHz. Original sources:
 
 ## Usage
 
+See the canonical
+[**Load from Hugging Face**](https://github.com/Scicom-AI-Enterprise-Organization/neucodec-44k#load-from-hugging-face)
+section in the source repo. **`decoder_depth=20` is required** — the weights are a
+depth-20 decoder, so loading with any other depth mismatches the architecture.
+
 ```python
-import torch, soundfile as sf
+import soundfile as sf
 from neucodec import NeuCodec
 
-model = NeuCodec.from_pretrained("Scicom-intl/neucodec-44k-d20")
-model.eval().cuda()
+# decoder_depth=20 MUST match this repo; pass token=... (or hf login) for access
+model = NeuCodec._from_pretrained(model_id="Scicom-intl/neucodec-44k-d20", decoder_depth=20)
+model = model.eval().cuda()
 
 # encode (16 kHz path, frozen) -> codes -> decode (44.1 kHz, this finetune)
-codes = model.encode_code(input_waveform=wav_16k)        # [B, 1, T], identical to base NeuCodec
-recon_44k = model.decode_code(codes).cpu().numpy()[0, 0] # 44 100 Hz
-sf.write("recon.wav", recon_44k, 44_100)
+codes = model.encode_code("input.wav")                   # [1, 1, T], identical to base NeuCodec
+recon_44k = model.decode_code(codes).squeeze().cpu().numpy()
+sf.write("recon.wav", recon_44k, model.sample_rate)      # 44100
 ```
 
 The discrete codes are **identical to base `neuphonic/neucodec`** — only the reconstruction