docs: document fp16 single-file + precise two-path usage

README.md

@@ -72,9 +72,20 @@ CTC models are **non-autoregressive** — they decode a whole clip in one forwar
 
 FP16 and FP32 produce **identical transcripts**, so FP16 is the recommended default (half the VRAM, ~2× faster). Batched throughput on the 4090 reaches **~208× real time** (the 222-min FLEURS set transcribes in ~64 s). A **16 GB GPU** (e.g. Colab T4 / L4) is enough for fp16. CPU works but is slow.
 
+## Files in this repo
+
+| File | What it is | Size | When to use |
+|---|---|---|---|
+| **`model_fp16.pt`** | Consolidated **fp16** weights, single file | **~6.2 GB** | **Recommended.** Smaller, faster download; fp16 inference. |
+| `pp_00/tp_00/sdp_00.pt`, `sdp_01.pt` | Original **fp32** FSDP checkpoint shards | ~12 GB | If you want full fp32 precision weights. |
+| `config.json` | Model metadata (arch, tokenizer, vocab) | — | Read by tooling; you don't load it directly. |
+| `inference_colab.ipynb` | Ready-to-run Colab notebook | — | One-click demo. |
+
+> Both weight files produce **identical transcripts** at fp16. The single `model_fp16.pt` is just half the download — prefer it unless you specifically need the fp32 master weights.
+
 ## Usage
 
-Install the official Omnilingual ASR runtime (the model loads through fairseq2's asset system):
+**1) Install the Omnilingual ASR runtime** (the model loads through fairseq2's asset system):
 
 ```bash
 # system dependency for audio I/O
@@ -87,27 +98,34 @@ pip install torch==2.8.0 torchaudio==2.8.0 torchvision==0.23.0 --index-url https
 # On Colab/Jupyter, restart the runtime once after this install.
 ```
 
-Load **this** fine-tuned checkpoint by registering a small custom asset card that points fairseq2 at the weights in this repo, while reusing Omnilingual ASR's official Persian-capable tokenizer:
+**2) Download the weights + register a fairseq2 asset card** that points at them, reusing Omnilingual ASR's official Persian-capable tokenizer. The asset card's `checkpoint:` can point at **either** the single `model_fp16.pt` file **or** the shard directory — both work.
 
 ```python
 import pathlib, torch
-from huggingface_hub import snapshot_download
+from huggingface_hub import hf_hub_download
 
-# 1) Download this repo's weights (FSDP shards under pp_00/tp_00/)
-local_dir = snapshot_download("RadeAI/Rade-ASR-CTC-3B-fa")
+# --- Recommended: download ONLY the single fp16 file (~6.2 GB) ---
+ckpt = hf_hub_download("RadeAI/Rade-ASR-CTC-3B-fa", "model_fp16.pt")
 
-# 2) Register a fairseq2 asset card pointing at the local checkpoint dir
+# --- Alternative: full fp32 shards (~12 GB) ---
+# from huggingface_hub import snapshot_download
+# ckpt = snapshot_download("RadeAI/Rade-ASR-CTC-3B-fa", allow_patterns=["pp_00/tp_00/*"])
+
+# Register a fairseq2 asset card pointing at the downloaded checkpoint
 asset_dir = pathlib.Path.home() / ".config/fairseq2/assets"
 asset_dir.mkdir(parents=True, exist_ok=True)
 (asset_dir / "rade.yaml").write_text(f"""\
 name: rade_CTC_3B_fa
 model_family: wav2vec2_asr
 model_arch: 3b_v2
-checkpoint: "{local_dir}"
+checkpoint: "{ckpt}"
 tokenizer_ref: omniASR_tokenizer_written_v2
 """)
+```
 
-# 3) Transcribe (clips must be < 40 s)
+**3) Transcribe** (clips must be **< 40 s**):
+
+```python
 from omnilingual_asr.models.inference.pipeline import ASRInferencePipeline
 pipe = ASRInferencePipeline(
     model_card="rade_CTC_3B_fa",