Add config.json so Hugging Face can count downloads

Metadata only; weights unchanged. Quick start now downloads through huggingface_hub.

README.md

@@ -128,13 +128,15 @@ FLOPs. Raspberry Pi 5 measurements are now recorded above; Jetson Orin Nano rema
 ## Quick start
 
 ```bash
-pip install numpy
-git clone https://huggingface.co/sraivante/superfast-tiny-home-robotics-json-1m-v3
+pip install numpy huggingface_hub
+python -c "from huggingface_hub import snapshot_download; snapshot_download('sraivante/superfast-tiny-home-robotics-json-1m-v3', local_dir='superfast-tiny-home-robotics-json-1m-v3')"
 cd superfast-tiny-home-robotics-json-1m-v3
 python engine.py "kitchen ki light band kar do"
 # {"activity":"light","subject":"kitchen_light","action":"OFF"}
 ```
 
+`git clone https://huggingface.co/sraivante/superfast-tiny-home-robotics-json-1m-v3` also works.
+
 In Python:
 
 ```python
@@ -286,6 +288,7 @@ describes the train/eval construction. The preserved 340,000-row training mix di
 
 | Path | What |
 | --- | --- |
+| `config.json` | release metadata; also the file Hugging Face uses to count downloads |
 | `model/model_v3_fp16.npz` | weights, fp16, 2.1 MB |
 | `model/model_v3_config.json` | vocab (`itos`) + architecture config |
 | `model/devices-catalog.json` | the 1,004 subjects with allowed actions |

config.json

@@ -0,0 +1,20 @@
+{
+  "name": "superfast-tiny-home-robotics-json-1m-v3",
+  "version": "v3.0.0",
+  "task": "text-generation",
+  "architecture": "gpt-style-causal-decoder-char-level",
+  "languages": ["en", "hi"],
+  "runtime": "numpy",
+  "entry_point": "engine.py",
+  "weights": "model/model_v3_fp16.npz",
+  "model_config": "model/model_v3_config.json",
+  "catalog": "model/devices-catalog.json",
+  "d_model": 160,
+  "layers": 4,
+  "heads": 4,
+  "ffn": 512,
+  "max_len": 360,
+  "vocab_size": 78,
+  "parameters": 1143328,
+  "dataset": "sraivante/home-commands-json-v3"
+}