Publish accepted A75 checkpoint with bound evaluation evidence

README.md

@@ -95,6 +95,7 @@ model = AutoModelForCausalLM.from_pretrained("limoXD/romaji2ja")
 bos_in, bos_out = chr(0xEE00), chr(0xEE01)   # romaji-start, japanese-start
 prompt = bos_in + "kyouhaiitenkidesu" + bos_out
 ids = tok(prompt, return_tensors="pt", add_special_tokens=False)
+ids.pop("token_type_ids", None)  # Qwen-style model does not consume this field
 out = model.generate(**ids, max_new_tokens=64, do_sample=False)
 print(tok.decode(out[0][ids.input_ids.shape[1]:], skip_special_tokens=True))
 # 今日はいい天気です

release_manifest.json

@@ -1,6 +1,6 @@
 {
   "schema_version": 1,
-  "generated_at_utc": "2026-07-12T14:35:01.736646+00:00",
+  "generated_at_utc": "2026-07-12T15:25:24.814747+00:00",
   "model_repo_id": "limoXD/romaji2ja",
   "space_repo_id": "limoXD/romaji2ja",
   "source_repository": "https://github.com/dj-thank/romaji2ja",
@@ -93,8 +93,8 @@
       "sha256": "3f749ffa67a33b6e56825767490e1d4e361c1f29d9e56a9be94b1ae044e68ea6"
     },
     "README.md": {
-      "bytes": 5466,
-      "sha256": "23ab66ec1ec289b9d1b24c36f4f473e77a8f124416c2fe6ed94d834515e58022"
+      "bytes": 5546,
+      "sha256": "f8d903104fa7ace9ad8c09800724d8f5678557e9d34fd76528114785beba1ce0"
     },
     "NOTICE.md": {
       "bytes": 691,