chore: readme

README.md

@@ -1,3 +1,34 @@
 ---
 license: mit
 ---
+
+# AniTag2Vec
+
+Training and inference examples are all available on [my github](https://github.com/michael-0acf4/anitag2vec).
+
+Implementation is detailed in [this blog post](https://blog.afmichael.dev/posts/2026/set-embeddings-and-anitag2vec/).
+
+```python
+device = torch.device("cuda" if torch.cuda.is_available() else "cpu")
+cfg = SetupConfig.load_from_file("./pytorch/setup_params.json")
+data, tagtok, anitag2vec = get_setup(
+    cfg,
+    device=device,
+    prefix_path= "."
+)
+# Load model
+anitag2vec.load_state_dict(torch.load("./pytorch/anitag2vec_e15_s50000_p1871744.pth"))
+anitag2vec.to(device)
+anitag2vec.eval()
+runner = AniTag2VecRunner(tagtok, anitag2vec)
+
+# Inference
+def compare(a: str, b: str):
+    ax = runner.run_inference_human([a])
+    bx = runner.run_inference_human([b])
+    howmuch = ((F.normalize(ax) @ F.normalize(bx).T).item())
+    print(f"{howmuch:.2f}: '{a}' vs '{b}'")
+
+compare("#1girl #1boy", "#1boy #1girl")
+# 1.00: '#1girl #1boy' vs '#1boy #1girl'
+```

pytorch/setup_params.json

@@ -0,0 +1,12 @@
+{
+  "TRAINING_TAKE_EXAMPLES": 50000,
+  "TRAINING_BATCH_SIZE": 256,
+  "HYPERP_TAGTOK_MAX_TOKEN_CLAMP": 128,
+  "HYPERP_TAGTOK_VOCAB_SIZE": 5000,
+  "HYPERP_TAGTOK_MIN_FREQ": 3,
+  "HYPERP_TRANSFORMER_D_MODEL": 128,
+  "HYPERP_TRANSFORMER_N_HEADS": 8,
+  "HYPERP_TRANSFORMER_N_LAYERS": 2,
+  "HYPERP_OUTPUT_EMB": 128,
+  "HYPERP_EPOCHS": 15
+}